Read my full disclaimer here.

My previous post on ride-hailing examined Waymo’s positioning for the EV era and whether Uber would be a winner. My opinion is that Uber is positioned to succeed based on the ability to quickly deploy AVs to its existing user base, the poor economics of supplying all ride-hailing demand with AVs, and the abundance of Waymo alternatives for Uber to utilize. I have since continued to follow Uber and its share price, inspiring me to write a second piece focusing on Uber’s two business lines and whether they can continue to grow, ignoring the AV factor.

Where is the growth coming from?

Uber’s growth, given its size, has been nothing short of outstanding. 1Q26 earnings were stronger than expected at 25% bookings growth (the total value of all trips and orders) and 42% adjusted operating income growth.

Gross bookings and margins have continued to increase despite heightened competition and relatively high market penetration. For mobility, while the core markets continue to grow, much of the growth has been driven by a focus on underserved communities (i.e., suburbs) and wealth classes. In addition to being underpenetrated, the “sparse markets” and premium products garner higher margins, making them an incredible source of future growth.

The high and low-end products, such as Black and Moto, have seen 40%+ growth in recent years, aided by investments in driver acquisition and geographic availability.

Another valuable aspect of Uber’s growth is its consistency. Competitors like Lyft and DoorDash have single-segment businesses that offer no insulation if the underlying market slows. In comparison, Uber’s combination of Delivery + Mobility has led to bookings growth hovering around the low twenties for years, despite much greater fluctuations across its business lines.

Previously, Uber operated these as separate segments, but has moved to unify the tech stack, customer experience, and decision making. In accordance with their one-app strategy, it’s believed that a centralized focus will help leverage cross-product benefits that many competitors can’t. As we’ll discuss later, this further supports the argument that Uber still has room for growth.

But when considering growth rates, it’s important to recognize that high historical growth rates, particularly in consumer tech, don't strongly correlate with sustainable future growth. Many consumer companies grow rapidly before suddenly hitting market saturation or rampant competition. Moto and Black's high growth imply those customer segments were long underserved, but how much runway remains is an open question.

What is clear is that Uber is still investing aggressively to grow users and frequency. The March 2026 acquisition of Blacklane, a leading chauffeur service, signals continued conviction in the high-end segment, and management sees significant growth opportunities ahead.

Despite a supposed 65% of consumers having used a ride-hailing app in the last year, their usage varies widely by region, suggesting room for higher trip frequency through expanded use cases. Ride-hailing has become cheap and safe enough that 42% of Latin Americans view it as a daily commute option, but in Europe and North America, it is mainly used for social and leisure purposes.

The delivery business presents another situation. Uber is falling behind DoorDash in bookings, growth, and US leadership. In 2022, Uber Delivery had $56B in bookings while DoorDash had $53B. Over the following three years, DoorDash compounded at 24% to $110B for TTM 1Q26, while Uber Eats compounded at only 18.4% to $96B. While the exact numbers are unclear, third-party US market share statistics invariably show DoorDash with ~2x Uber Eats’ market share, 56-67% versus 23-26%.

The split isn’t as drastic in all US cities, however. For instance, Uber Eats is estimated to hold a leading share in cities like Los Angeles. But the general trend remains that Uber and DoorDash largely dominate US delivery. Both are still growing their share in practically all cities, overtaking the smaller competitors.

Primarily, two trends have led to this varied market share development:

• Uber Eats and DoorDash established early market positions in different cities or acquired companies to expand their market share. This led to continued growth and market-share leadership in those cities. Since their network effects are primarily local, it’s hard to take market share from the other in a city where the competitor dominates.

• Uber Eats historically neglected suburbs, just as Uber Mobility did, leaving Uber Eats an urban winner and DoorDash dominant in the suburbs. Presumably, Uber Mobility’s push into the suburbs and the expansion of UberOne (+55% YoY) should help change this.

  • Management indicated on the first-quarter earnings call that it’s “very, very early innings” for the US and global suburban Delivery efforts.

Although DoorDash’s Western dominance continues, Uber Eats' growth opportunity remains clear. The food delivery market has much lower penetration/adoption than rideshare, greatly benefits from any change in prices and restaurant selection, and marginal improvements in delivery time and driver quality are easily noticed by consumers.

There’s also geographic expansion available. While Uber Mobility is in 70+ countries, Delivery is in 45 countries. In 2026, Uber plans to expand Eats into Eastern Europe, starting operations in the Czech Republic, Greece, Romania, Austria, Denmark, Finland, and Norway. Delivery competitors already exist, but Uber’s scale, expertise, and enhanced focus on cross-product benefits for Uber One should help it gain share.

Once you fold in grocery delivery, non-food store delivery, and UberOne, the delivery side of the business has several areas for continued expansion. CEO Dara Khosrowshahi highlighted the magnitude in his 3Q25 remarks, stating that delivery is at an $80B annual run rate (now >$100B) while the global food delivery opportunity is $2 trillion. While Uber’s reasonably attainable slice of the $2 trillion is likely small, it still conveys the extent of Uber’s opportunity ahead.

Addressable Market Math

But for this to be a worthwhile investment, double-digit free cash flow growth needs to be attainable over the next decade. So what would Uber have to do to make that happen?

Start with Dara’s claim that Uber is “still in the early stages of customer acquisition,” and more importantly, customer use expansion. 40% of customers regularly use more than one Uber product, yet only 20% of eligible users use both Delivery and Mobility (as of 3Q25), and 30% have never even tried Delivery (as of 1Q26).

Zooming out further, Uber operates in ~73 countries, which contain an urban/suburban population of ~2.4B. Uber’s 199M MAPCs are just 8% of that population. Not all of that population is reasonable potential Uber users, given income levels or lifestyle choices, but 8% still leaves significant room for growth.

It’s also important to consider previous users who churned. Assume 2.5x as many people have used Uber at some point as are currently active MAPCs, and the penetration rate jumps to 21%. The question is then why those customers churned, and whether Uber’s product line expansion, price improvements, and quality improvements can win them back.

Assuming 35% of those previous users are recoverable over the next ten years, 101M MAPCs can be gained from them. Next, Uber could reasonably attain 5% of the remaining urban population, adding another 116M MAPCs. Taken together, this would lead to year 10 MAPCs of 409M, with a 7.9% CAGR. And arguably even higher MAPC assumptions are reasonable.

Once you factor in the historical average 2% Trips/MAPC growth and even a mere 1% $ Bookings/Trip annual growth, then you get to a ten-year 12% bookings CAGR.

FCF margin will also likely continue to increase as higher-margin business lines (Reserve, Black, and Uber for Business) grow faster than the overall business. Fixed cost leverage and Delivery margin maturation add to this. Modeling FCF margin expanding from 10% to 15% over 10 years (or bookings margin from 2.7% to 4%) is a reasonable outcome, resulting in a 16% FCF CAGR.

These projections are by no means concrete, but they illustrate how Uber has the potential to grow cash flow in the double digits over the next decade. To make this a reality, Uber must maintain dominant market share in markets where competition is intensifying (India, Europe, Africa, and LATAM) and where delivery fragmentation is greater than mobility.

Estimating Uber having a 60% global market share by year 10, the implied ride-hailing/delivery penetration rate is 24%, which is reasonable. But the projections become more demanding the lower the share figure goes, which is why the international competition picture matters so much.

Case Study of India

India has become a glaring example of the international competition Uber is facing. Ola was the main domestic competitor, launching in 2010 and reaching an estimated >45% share by 2018. Uber entered India in 2013 and attained a 35% share of the taxi market by 2018. Over the next five years (2018-2023), Uber continued to gain market share in the cab market, reaching an estimated 45-50%, while Ola fell behind to 25-30%.

During this time, a third competitor, Rapido, emerged under their noses. While Uber and Ola focused on pricier taxis in Tier-1 cities, the Rapido founders (based in India) recognized a clear service gap. Rapido began by offering solely bike taxis in 2016, which were 50% cheaper than car-hailing services. They focused on Tier-2 cities where car-hailing would be too expensive, leaving Uber and Ola to fight for high-income riders in Mumbai and Delhi.

Bikes were the perfect product for quick, cheap rides that the average Indian needed. Rapido also navigated regulatory hurdles that Uber and Ola failed to, allowing them to quietly build a massive user base they could later upsell into.

Then Rapido expanded into auto-rickshaws in mid-2022 and cabs in late 2023 with a differentiated business model. Rapido charged cab drivers a flat access fee as low as 9 rupees per day, the equivalent of just $2.85 per month. This made cabs 10-15% cheaper and drew drivers to the platform. In a few years, they have already reached 20% share of the four-wheel ride-hailing market, while Uber still leads this segment at 45%.

In overall trip volume, Rapido has overtaken Uber, capturing an estimated 50% market-share versus 40% for Uber. Revenue for Rapido has grown quickly, rising from ₹145 crore in FY22 to ₹934 crore ($97M) in FY25. At the same time, operating losses decreased 45% in FY24 to ₹371 crore and another 30% in FY25 to ₹258 crore ($29M) in losses.

In contrast, Uber’s smaller trip share resulted in much higher revenue of ₹3,849 crore ($400M) in FY25, but overall growth was just 2.3% YoY, according to government filings. At the same time, Uber India’s losses have skyrocketed from ₹89 crore in FY24 to ₹1,512 crore ($157M) in FY25, driven by increased promotions and a shift to a driver subscription model. Shifting to a subscription model means Uber’s revenue growth in India depends solely on driver growth, not on the number of rides, unlike in most markets.

It’s quite telling that an overlooked startup has pushed the largest ride-hailing company to change its business model. This tale raises two questions. Can Uber be profitable in a market where domestic competitors have much lower cost bases? For a domestic company, Rapido clearly needs much lower fees to be profitable than Uber does. Second, is Uber too out of touch with some of its foreign markets to avoid being disrupted? If this could happen in India, why can’t it happen in more of Uber’s low-income markets?

On the other hand, Uber is also now in a better position with clearer regulations and the ability to undercut the competition and/or provide better service. India is already Uber’s third-largest market in trip volume, with 1.5M monthly drivers and 1.2B annual trips, and there’s seemingly room to grow. Even without Moto, the Uber India President said Uber Black and Reserve are growing so fast that there aren’t enough vehicles to meet demand.

Uber started in the US with premium products before it moved to the mass-market Uber product. With the even greater wealth disparity in India, perhaps only the high-end of the market is needed to succeed. Uber's customer experience advantage is still a clear differentiator in India. From the customer testimonials I can find, Rapido may be a few rupees cheaper, but Uber’s reliability is still unmatched, and the Uber One membership makes it a no-brainer.

The India story illustrates both sides of the moat question, where Uber was vulnerable and where it held. So how durable is Uber’s moat?

How strong is Uber’s moat and business model?

Uber’s main moat—what protects it from competitors stealing customers and eroding margins—is its network effect. Uber needs quality users on both ends (customers and drivers) to make the platform attractive. The more scale they achieve, the better the product: more supply, faster matching, better pricing, and more data to improve the experience. But network effects in Uber’s case are also overrated.

As seen in the India case study, Uber's moat struggles because the driver and rider networks it builds are local—the network must be established city by city, country by country. Dominance in Mumbai offers little protection when a competitor enters Delhi. This makes Uber's network effect fundamentally different from that of platforms like Facebook, Airbnb, or eBay, where every new user strengthens the network for existing users, nationally or globally.

Rapido’s market disruption also highlighted that Uber’s moat holds only if neither side of the platform has a better alternative, which arguably isn’t much of a moat at all. Given that many drivers already use multiple platforms (“multihoming”), capturing drivers isn’t as hard as it’s made out to be. If the supply side is easy to solve and the value proposition to customers is attractive (i.e., lower prices), then Uber’s business is susceptible to competition.

On the food delivery side, the network moat is arguably stronger, though still geographically constrained. The delivery business adds a third layer of complexity, necessitating an ample supply of drivers to pick up food quickly, a sufficient quality and variety of restaurants, and customers who trust the platform. Multihoming is also less prevalent among restaurants; many lack the necessary technology to integrate easily with delivery apps, and leading POS systems already have deep integrations with delivery platforms like Uber Eats, DoorDash, and Grubhub.

So why does Uber stay so dominant?

Uber benefits greatly from scale. Despite being a local-network business, travelers and businesspeople still derive value from Uber’s near-global availability (~10% of trips in 2025 were taken outside their home city). The sheer level of data and profits Uber generates also provides greater leeway to test out and deploy new products and promotions, compete on price, and continuously improve routing and matching algorithms.

But one of the greatest strategic decisions Uber management has made is investing heavily in Uber One. The subscription product reduces multihoming, increases stickiness, and leverages what no pure-play competitor can match—a bundled delivery and mobility offering. As of 1Q26, Uber One has reached an astonishing 50M members, accounting for over 50% of booking dollars, and continues to grow rapidly.

For the 20%+ of users who already use both products, one app and one subscription is a no-brainer. But Uber One clearly provides value beyond the cross-product users, too, with subscribers now representing 25% of all MAPCs. With no delivery fees and ride discounts, the chance a user finds a cheaper offering elsewhere is quite low.

The economics work in Uber’s favor in a way no single-product competitor can replicate. Because subscribers don’t use both products equally (one may only order delivery while the other uses only mobility), Uber can offer industry-leading discounts across both without the margin pressure a pure-play food delivery or ride-hailing app would face when offering the same discount.

The result is a flywheel. More subscribers drive more volume, improving the platform for drivers and restaurants and making the subscription more valuable.

International Risk

What happened in India is happening elsewhere. Bolt and DoorDash pose significant competition across Europe, and inDrive and DiDi threaten Uber’s position in Latin America. Differentiated business models and lower commission structures are the common thread. And since Uber is a publicly traded company with established take rates, pivoting to a new revenue model is structurally harder for it than for a private startup with greater flexibility.

That said, Uber has shown it can fight back when it invests. Eastern Europe grew 84% YoY in 2025, Uber’s UK share jumped from 28% to 38% between 2022 and 2025, and Germany’s share rose from 10% to 26%, largely on the back of Uber One. Market position is also improving in Spain, Australia, Japan, and Canada.

International competition is the biggest risk to Uber’s long-term thesis, but it isn’t a one-way story. Uber’s core businesses are healthier than the market seems to appreciate. Growth is broad-based, the delivery opportunity is underpenetrated, and Uber One is quietly becoming the moat that the network effect alone never was. But the international picture is genuinely uncertain, and the India case study is a reminder that Uber’s dominance is never guaranteed.

What makes Uber a fascinating investment isn’t the next two years; it’s the terminal value question. In the next post, I’ll examine how autonomous vehicles reshape the bull and bear cases, and what the stock is actually pricing in today.

Subscribe now

Share

Interested in more?

Read my most popular posts:

• Topicus: Falling Knife or Undervalued?

• Dino Polska Q3 Update and Q4 Update

My thoughts on Uber and Waymo

Read my full disclaimer here.

I have been a long-term believer in Dino Polska’s potential, especially at current prices. But Q4 disappointed markets as expenses grew much faster than revenue. While I still have confidence that Dino Polska is undervalued, it is definitely testing investors’ patience with continued margin contraction and lack of in-depth discussion on the topic.

Besides a short letter from the CFO, which didn’t fully address many of the issues Dino is facing, management’s communication is overall disappointing. Lack of communication is easier to be passive about when times are good, but when several years have seen stunning margin contraction, there should be more disclosure.

The lack of disclosure and continued earnings misses have destroyed Dino’s stock returns over the past five years. The reality with markets is that even if a company is undervalued, a disappointing quarter can keep the stock under pressure.

With continued stock price declines, it’s important to consider how much upside the investment truly has and how long that upside will take to materialize. That’s what I aim to do with this update.

The Good

Store rollout continued with an acceleration in new store additions from 283 added in 2024 to 345 added in 2025. As a result, net sales area grew 13.1% YoY, and sales grew 14.9% YoY. LFL growth was strong at 4.4%, but weaker once you consider 4.7% food inflation for Poland.

Two additional distribution centers were also added, now totaling 12 throughout Poland. These can support 4,200-4,800 stores versus the 3,033 open at year-end, meaning the existing DC network is up to 37% underutilized. Although underutilization is generally a bad thing, for growing companies it has two positive indicators. Growth is expected to continue, and mature margins are significantly higher than current levels.

With 345 new stores that aren’t at mature earnings capacity and underutilized DCs, the current earnings power of Dino’s network is underestimated two-fold. They are paying for labor that may not be working at full capacity and have stores that aren’t earning up to their full potential yet.

The Bad

While government-mandated minimum wage hikes have slowed, as discussed in the last Dino Polska update, labor unions have taken their place. In December 2025, OPZZ called for a monthly salary increase of PLN 900 (18% increase), introduction of the Company Social Benefits Fund (which is legally required for a company of Dino’s size), and increased employment in non-managerial positions by 30%.

OPZZ, other unions, and left-wing Polish politicians have accused Dino of overworking employees, providing inhumane working conditions, and having the lowest wages for the sector. It’s believed that since Dino primarily exists in rural areas, they have been able to attract workers despite low wages and poor work environments.

Although the salary increase was initially demanded to be implemented in February 2026, it seems no action has been taken so far. With 55,856 employees at year-end (and more now), a PLN 900 increase would substantially impact margins. If that full increase were to take effect (unlikely), the EBITDA margin could contract from 7.6% for FY25 to 5.8%, before even considering the salary increase that would come from increasing non-managerial employees by 30%.

With that being said, union demands generally start much higher than the expected results. If only 25% of the demanded wage increase were to take place, it would instead lower margins from 7.6% to 7.1%. If that were to come true, then EBITDA would only grow ~8% in FY26, assuming revenues grow 15%.

Dino management is in a predicament. Many stores staff as few as 5–10 employees; if even a few decide to go on strike, it could substantially disrupt operations. However, the risk is somewhat mitigated by the number of stores Dino has (as the majority going on strike is unlikely) and the fact many employees don’t have better alternatives. Just because a union is pushing for higher wages doesn’t mean it will happen.

Profitability

The employee wages issue has lasted longer than many expected. Again, employee expenses far outpaced revenues. In FY25, employee count grew 12%, while employment expenses grew 21%—slower than the 28% of FY24 but still outpacing revenue growth of 15%.

For a company of Dino’s scale, you would generally expect operating leverage, but practically every expense line has increased as a percentage of revenue, mainly from wage growth.

The other surprising result is that Biedronka gained margin in FY25 and has surpassed Dino in terms of EBITDA margin, increasing from 7.7% in 2024 to 7.9% in 2025. Conversely, Dino’s contracted from 7.9% to 7.6%. This comparison doesn’t capture Dino’s store ownership advantage over Biedronka’s leasing model, but it still shows how Dino’s margin deterioration is out of step with other leading Polish grocers.

This is a period where Biedronka’s larger store format and higher sales provide an advantage. Although Dino employs fewer people per store, Biedronka employs fewer per square meter and sales dollar generated. Dino’s total employee count averages 18 employees per store, while Biedronka’s is ~22, but Biedronka’s stores are 80% bigger and generate 3.6x more revenue per store. This means wages are a smaller portion of Biedronka’s operating expenses, and thus wage increases have less of an effect on Biedronka’s results.

Under a regular economic environment, Dino’s distribution and store model have several cost advantages, but with rapidly rising wages, Dino is being hurt more.

Beyond store format dynamics, Biedronka and their parent company’s bargaining power is higher. Jeronimo Martins generates 4.6x as much revenue as Dino, and Biedronka alone generates 3x as much. With that much greater scale, they undeniably have more negotiating power with suppliers and more geographic variability and size to shield cost increases.

This could mean Dino will struggle more to match prices close to Biedronka. But the concern is lessened by the fact they appeal to somewhat different markets, sell different brands, and Biedronka has historically discounted more than Dino. Additionally, if Biedronka has such lower employment per square meter, then the union argument that Dino understaffs stores is weak. Lastly, Dino’s offering to customers is clearly still attractive, signaled by Dino growing LFL by 4.4% versus Biedronka’s mere 1.9%.

On the positive side, the main concern of Dino’s in terms of margins is regulatory and union-based rather than competitive. In years prior, inflation and deflation impacted gross margins, and now it’s affecting wages, but it seems Dino’s competitive positioning is still strong and not a core area of concern.

FY26 Outlook

First, let’s consider management’s projections from the FY24 annual report and how that measured up to reality.

• ✅Acceleration of store count openings in the high teens—This projection was exceeded. Store openings grew 22% for FY25, with Dino opening 345 stores versus 283 in the year prior.

• ❌High single-digit LFL sales growth—LFL sales only grew 4.4%, and food inflation was 4.7%.

• ❌Bring EBITDA margin back onto the path of gradual growth—Although margin deterioration was less than the year prior, EBITDA margin still declined from 7.9% to 7.6%. 4Q25 was particularly horrible at 6.9% versus 8.1% in 4Q24.

  

• ❌Slower OPEX growth than revenue growth—Management could not have been more wrong. Besides materials and energy, every expense area grew faster than revenue.

Based on FY25 performance, even management’s vague guidance turned out to be incorrect. The only guide they achieved was the one in Dino’s direct control, how many stores they opened. With how unexpected and variable Poland’s macroeconomic environment has been, it’s unsurprising the rest would be challenging to predict.

Unsurprisingly, this year’s CFO letter had even less guidance. Although they believe the 4.4% LFL growth was strong given the economic environment, they’re aiming for higher this year. Overall for FY26, they expect “a moderately positive scenario.”

Positively, store openings will increase by “a percentage in the teens.” If the LFL growth and higher store openings come true, then Dino’s sales growth should be in the strong mid-to-high teens. But margins may still suffer. Given the still cautious consumer and deflationary environment, management plans to prioritize sales volume and customer attention, which will influence their pricing policy and impact EBITDA margins.

Deflation

In an earnings call with institutional investors, management noted the single-digit deflation in late 2025 has deepened in early 2026, especially in the meat and produce categories, which are large drivers of sales for Dino. Furthermore, raw material costs are rising, and the Iran war could worsen this by pressuring fuel costs for receiving and delivering food to stores. It’s still early in the year, but these aren’t positive developments and may lead to a weak Q1.

Once again, Dino seems to be a mixed bag. Although their growth runway is still intact, the macroeconomic environment is dampening earnings. A deflationary environment will likely mean sales growth is depressed by lower prices for the goods, needing volume to grow quicker to offset it. So, management’s priority for FY26 of increasing volumes and attention is reasonable. While this will likely hurt margins (again) in the short term, it is arguably better for the lifetime value of the business.

Valuation

I wasn’t originally planning on including a valuation section in this update, but the valuation was so astonishingly disjointed that I felt I had to include it. Dino trades at 21x reported net income, which is depressed by margin contraction and growth investments. Based on my calculation of true maintenance free cash flow (OCF minus maintenance capex), Dino is priced at a mere 13x FCF.

To further illustrate how absurd the current valuation is, I presented models based on three different measurements of FCF. These models only project out ten years of growth, then a terminal multiple of just 12x FCF. I projected Dino to continue opening nearly 400 stores a year, like they expect to do in 2026. This would roughly double Dino’s footprint in that timeframe.

The first sensitivity table is my calculation of FCF based on maintenance capex, second is using depreciation as a proxy for capex, and third is using total capex (which I don’t recommend using but present here to show how low current prices are).

The conclusion to gather from the models is two-fold:

• If margins continued to contract through year 11, Dino would still likely yield double-digit returns with even modest same-store sales growth and continued store rollout. Under the maintenance capex model, 2% revenue per store growth and margin contraction from 7.7% to 4.5% could still lead to >12% annual returns.

• Regardless of what calculation of profits you choose, Dino’s valuation versus potential outcomes is clearly dislocated. The stock is priced as though margin deterioration will continue indefinitely and for LFL growth to stall. While this is a possibility, Dino’s margin contracting by 50% is quite a bearish scenario, but is being priced as though it’s a likely reality.

Ultimately, the question is whether you believe that Dino can get OPEX under control or you think the government and unions will squeeze Dino so much that over half their margin gets eroded.

Conclusion

Once again, Dino presents a strong underlying business model impacted by macroeconomic fluctuations. From the Ukraine war to inflation, deflation, and massive wage hikes, Dino has still held relatively strong margins. But the struggle doesn’t seem to be over. 2026 will continue to test margins with deflation while wages and fuel costs might increase.

Nonetheless, at current prices the downside appears limited and the upside substantial. But the timeline is uncertain. I'm continuing to hold, though I believe it’s worth considering other investments that may materialize quicker. If you have the desire to hold Dino for at least five to seven years, I believe the chances are quite strong that your shares will be worth much more than they are today.

Subscribe now

Share

Interested in more?

Read my most popular posts:

• Topicus: Falling Knife or Undervalued?

• Dino Polska Update

My thoughts on Uber and Waymo

Read my full disclaimer here.

Hi all. It’s been some time since my last post, but I’ve been working on the fintech space recently and have some posts in the pipeline. I also plan to release a follow-up on Uber soon.

This post examines BNPL through the lens of Klarna. Posts on other BNPLs and neobanks will follow.

Fintech’s rocky past

Despite being viewed as a high-growth industry with secular tailwinds, fintech performance in the public markets has been weak. With sky-high valuations, including many COVID-era IPOs, nearly all have lagged the market. Even investments in fast-growing fintechs like Adyen or SoFi would have lost money over the past 4-5 years.

Interestingly, the lackluster performance spans fintech categories, including BNPL players Klarna and Affirm; payment tech like PayPal, Adyen, and Wise; and neobanks Nubank, SoFi, Chime, Ally, and Dave. As of January 2026, a couple had annual returns over 10%, but sentiment has pulled back since then.

The only time period in which the majority of this basket outperformed was if you invested in late 2022 to early 2023. Many crashed after COVID and have reached profitability in the last year or two.

I believe the outlook for fintech continues to be quite interesting. With expectations somewhat tempered, there seems to be opportunity for reasonable entry prices.

But there is still plenty of risk and uncertainty.

Neobanks Revolut and Nubank are eyeing the US market, regulation concerns are heating up, MELI is pushing harder into LATAM banking, and Klarna seeks to transition from the leading BNPL to a full-fledged neobank. The main question I’m aiming to answer in this series is why one fintech should succeed over another and if any have a sustainable competitive advantage.

BNPL offers a useful entry point into this question. It’s the fintech category most directly challenging the credit card incumbents, with stark claims about consumer benefit and hotly contested economics. Below is a brief history of the category and a deep dive into its largest player, Klarna.

Buy Now Pay Later

Beginnings with Klarna

BNPL is considered to have started with Klarna, founded in Sweden in 2005.

In the early days of e-commerce, handing over card information to an obscure online merchant felt risky. Consumers had little control if the merchant didn't ship the product or stole their payment information. The issue was even greater in Sweden, where credit card use was low and buyer protections were limited.

While working at a factoring company, Klarna co-founder Sebastian Siemiatkowski observed how e-commerce businesses were handling this. Retailers were extending credit to customers and deferring payment until after delivery, then selling those receivables at a discount just to stay liquid. The opportunity became clear. A dedicated third party could step in to own the payment processing and financing, bringing both consistency and efficiency to a process that was largely improvised.

At the time, Siemiatkowski was attending the Stockholm School of Economics. Along with cofounders Niklas Adalberth and Victor Jacobsson, they pitched the idea at the school’s annual entrepreneurship competition. Despite losing the competition, an angel investor provided €60,000 in seed money and helped them hire their first five programmers.

The business model was simple: Klarna would pay the merchant for the order (with a service fee deducted), take on the accounts receivable risk, and invoice customers upon delivery. Even better, since most merchants cared about sales more than cash flow, Klarna could defer paying merchants for weeks. Customers often paid their balances early as well, making Klarna cash flow positive from quite early on.

After scaling in Sweden, Klarna expanded across the Nordics and into Germany by 2010, adapting to each country’s regulatory requirements. After seven years, similar credit offerings began to emerge, including Affirm in the US in 2012 and Afterpay in Australia in 2014.

What started as a way to protect payments in e-commerce morphed into a different form of credit with a unique value proposition. Klarna now positions itself as a better credit option. It helps consumers avoid snowballing debt, underwrites each transaction, and makes e-commerce payments seamless.

Strategic Pivot

Initially, part of Klarna’s strategy was owning the e-commerce checkout technology. But international expansion was challenging and their strategy became fragmented. Despite success in Sweden, Stripe and Adyen were far outpacing them globally. Between Stripe partnering with Shopify in 2013 and Klarna’s weak second attempt at US checkout in 2015, management decided it was time to pivot.

After the checkout failure, management’s vision for the company shifted to consumers. Powered by deeper customer data, Klarna wants to become your personal banker and shopping assistant—saving you money and recommending better products. With recent advancements in AI, the vision is close to becoming a reality.

To extend this strategy, Klarna became a licensed bank in Europe in 2017, allowing it to accept consumer deposits to help fund lending globally and expand beyond credit.

By 2019, Affirm had gained traction through big-ticket financing. Afterpay also entered the US, partnering with retailers Urban Outfitters and Anthropologie to finance low-value items. While Klarna had entered the UK in 2014, management eschewed the US, viewing it as unattractive due to heavy credit card use. But after Affirm and Afterpay’s early success, they realized the market was ready for BNPL, and the race for the US was underway.

Despite being far behind Afterpay in the US, they fought hard and began cementing large partnerships including Macy’s and Sephora in 2020. Their aggressiveness appears to have worked, reaching the highest US gross merchandise volume (GMV) for a BNPL. Klarna now has 118M customers and 966,000 merchants globally, while Affirm has 478,000 merchants and 26M customers. Other payment providers have also thrown their hat in the ring. PayPal launched Pay-in-4 in 2020 and has quickly caught up with $33B in GMV in 2024.

Significantly, Klarna won over Walmart from Affirm in early 2025, which represented 5% of Affirm’s GMV. Previously, Walmart offered Affirm-branded financing; they now exclusively offer Walmart OnePay powered on the back-end by Klarna.

Growth Over Profitability

For Klarna, though, pursuing US growth had a cost. Transaction fees, funding costs, and loss provisions are all higher in the US. But they got access to the largest payments market in the world—a non-negotiable if they want to be a market leader.

Klarna's business had been profitable from 2005 to 2018, albeit at single-digit margins and low ROE, but began recognizing operating losses after entering the US. These losses have begun to moderate, but Klarna is still not profitable as of 4Q25, nearly seven years after entering the US.

At first glance, investors could view Klarna as a behemoth versus other BNPLs, with 3 times the global payment volume of Affirm ($128B vs $43B). But in CY25, Klarna’s total revenues were 5% lower than Affirm’s at $3.5B vs $3.7B, meaning Klarna earns 2.7% on each transaction while Affirm earns 8.5%.

Management argues their lower GMV take-rate is a virtue of their more consumer-friendly business model, while Affirm claims their business is better with no ‘junk fees’ and higher-income customers.

To understand why Klarna's take-rate sits so far below Affirm's, you have to look at Klarna’s product mix and who they make money off of.

Business Model

Starting with Pay Later in 2005, Klarna has expanded to several debit and credit offerings. The core products are integrated directly into merchant checkouts as a branded payment option. Unlike Visa cards, which work at all merchants that accept Visa, each merchant must opt in to offer Klarna. Merchants offer Klarna alongside cards and digital wallets because it demonstrably increases conversion, order values, and purchase frequency.

• Pay Later defers payment to a later date (Pay in 30) or breaks it into installments (Pay-in-3 or Pay-in-4). It’s interest-free financing, meaning most of the revenue comes from merchant fees.

  • Merchants are charged typically 3.29%-5.99% of the order value + a 30-cent fixed fee.

  • Consumer fees are charged only if a customer misses a payment (Klarna calls them “reminder fees”) or wants to defer payment by 10 days (“snooze fee”). Late fees are charged several days after a payment is missed, and after multiple notifications to pay have been sent. The late fee is flat at $7 and maxes out at 25% of the order value.

  • Management said they tested no late fees but realized small late fees incentivize on-time payment. In contrast, credit card late fees average over 4x Klarna’s at $32. The CFPB has tried to limit credit card fees to $8, saying that’s enough to cover collection costs, but ongoing lawsuits have allowed credit card companies to continue charging the high fees.

• Pay-in-Full is basically a regular debit transaction facilitated by Klarna. Customers use this for similar reasons to using PayPal—they don’t have to enter card information and can use their Klarna balance. But with Klarna, users also benefit from more detailed “SKU-level” information on the products they bought.

  • As long-term financing has grown rapidly recently, Pay-in-Full has shrunk to 10% of Klarna GMV and 18% of transactions. Since it’s just debit, this category has a lower take rate but better cash flow and minimal funding costs.

• Fair Financing is a longer-term interest-bearing loan with scheduled payments. This is a newer product to grow high-ticket GMV and compete with Affirm, offering 6-24 month financing with monthly interest payments at a 0%-35.99% APR.

  • Introduction into the US has supercharged growth, with Fair Financing GMV growing 165% in 4Q25 to $4.5B and 193% in December 2025.

  • Fair financing is now 11.5% of GMV as of 4Q25, with room to grow, as only 20% of Klarna merchants offer Fair Financing.

• One-Time Virtual Card is Klarna’s solution to facilitate BNPL transactions at merchants that don’t accept Klarna, using a temporary Visa card.

  • Since Klarna only earns the Visa take-rate with the card, a $1.29-$5.99 issuance fee is paid by the user each time they request a one-time card.

  • The card can be activated in the Klarna app for a specific credit amount or directly through their Apple or Google wallet in-store or online.

  

• Klarna Card is a permanent Visa card enabling debit and credit use. It launched in the US less than a year ago (July 2025) to expand Klarna’s positioning as a bank, now expanding across the EU. It defaults as a debit card, but users can opt for a credit-installment plan before or after making a purchase (for a service fee). The downside is there’s no SKU-level data, since it runs on Visa rails, similar to traditional debit and credit cards.

  • It’s offered in five tiers, ranging from free to $44.99/mo. The free tier is similar to the one-time card while paid tiers offer perks like waived service fees and free subscriptions. The take-rate from merchants is lower, so consumers have to make up some of the difference either through fees for each purchase or a higher membership tier.

  • Unlike traditional debit cards, but similar to Neobanks, Klarna pays a 2.3%-3.28% APY on the deposit balance (for paid tiers) and provides cashback on debit purchases.

  • The Klarna card is already seeing great success in the US, growing Card GMV 209% in 4Q25 and making up 15% of all transactions. It’s also helping Klarna break into in-store spending, with 25% of US Klarna card spending being in-store.

  • This channel + the one-time card made up 12% of GMV for 2025.

• Klarna Balance functions as a stored-value account. Users can hold funds, receive refunds, and pay for transactions directly. In Europe, Klarna goes further with insured savings accounts (up to ~€105k), offering both flexible and fixed-term options at varying interest rates. Fixed-term deposits now make up 58% of total deposits, providing a funding base longer-dated than the loans they're issuing.

• Klarna Mobile is a new $40/mo mobile plan powered by AT&T with priority 5G data, available for their US consumer base.

Differentiated Data

Akin to card providers American Express and Discover, and fintechs like PayPal, Klarna is a payment network, card provider, and bank (licensed bank in the EU, powered by WebBank in the US).

Unlike traditional consumer cards, however, Klarna’s rails transmit “Level 3” data, which contains exact product details for past purchases. Traditional consumer cards only transmit Level 1 data (price, date, and merchant) while Level 2 & 3 are traditionally reserved for business cards. Klarna also provides order tracking and return initiation through their app, features traditional card providers don't offer, and that reinforce Klarna’s goal of being a better interface for managing spend.

Gaining SKU-level data from merchants was a hard-won effort, serving as a symbol of trust in Klarna and a testament to the usefulness of detailed data—improving product recommendations, spending insights, and advertising. In this sense, Klarna has an advantage over traditional payment providers, connecting purchases across users and knowing exactly what they buy and how they shop.

Competitive Advantage

Klarna has clearly built a compelling product that draws in over 100 million users, but is this position defensible?

Banking in general has some structural advantages. High regulation limits new entrants, scale drives better investment opportunities, and switching costs retain customers. But within the large bank tier, those advantages largely cancel out. Wells Fargo and BofA aren't winning customers from each other on moat strength. For fintechs and BNPLs, the dynamic is different again—lower regulatory barriers (for now) and easier account switching mean competitive advantages are harder to sustain.

Many operate outside the realm of traditional banking regulations and the costs to enter and grow aren’t that high. Switching costs are still a consideration, but differentiated offerings and a better customer experience motivate users to switch while modern payments technology has made moving money easier and cheaper. Jamie Dimon has even been vocal on the threat and level of competition fintechs are posing.

Within BNPL, Klarna’s main advantages are its network and European dominance. Klarna already has one of the largest merchant and customer bases for BNPL and a brand identity directly associated with BNPL. For instance, although PayPal has 3.7x as many active accounts and millions of merchants, its BNPL volume and number of users are a third of Klarna’s. Meaning, there is a disconnect between PayPal's brand and its ability to convert users to BNPL.

European dominance provides a structural funding advantage. Interest rates on savings accounts in Europe are nearly half those in the US. European consumers are also more likely to view Klarna as a banking option than US consumers are to view BNPL that way. With 85% population penetration in Sweden and deep roots across the Nordics and Germany, Klarna has a funding base that US-centric BNPLs like Affirm simply can't replicate.

On the merchant side, Klarna's scale is equally difficult to replicate. With Klarna’s merchant base crossing 1M and their wide array of payment options, it’s a no-brainer to choose Klarna over a random BNPL upstart. Klarna may lose occasionally to Affirm or PayPal, but that is still a duopoly/oligopoly type market.

On the consumer side, Klarna’s switching costs are worth questioning. BNPLs tout low-friction sign-up as a feature, but that same ease undercuts the network’s stickiness. When linking a debit card and answering a few onboarding questions takes only minutes, are there any switching costs?

I’d say so. A few-minute sign-up and a lower spending limit are still enough friction to prevent many purchases. Most users settle into a preferred BNPL and stay there. Merchants recognize this too. A customer base as large as Klarna’s is not worth giving up for a few extra basis points.

Some also believe the moat is weakened with merchants integrating multiple BNPLs. I disagree. Offering multiple BNPL options was inevitable and is beneficial, same as offering Visa and Mastercard. While it will heighten competition for new customers, the more merchants Klarna can add, even if they already offer Affirm and Afterpay, will still strengthen Klarna’s image as being available everywhere and will grow GMV from existing customers.

As more users adopt Klarna Balance and the Klarna Card, building up transaction history in the process, I expect stickiness to increase and strengthen Klarna's brand value.

Klarna is also continuously leveraging its network in new ways. Instant peer-to-peer transfer across Europe puts Klarna on par with traditional banks, backed by deposit accounts customers use for real transactions. To offer alternatives to traditional banking rails altogether, they've also announced a dollar-pegged stablecoin (KUSD) for cheaper and faster cross-border commerce and plans to develop a crypto wallet with Privy.

But they remain behind leading neobanks on several fronts. Competitors offer stock trading, currency conversion, and are (or plan to be) US licensed banks. Management hasn’t yet indicated whether this is on their roadmap or if their strategy is different.

Revenue Generation

As mentioned earlier, Klarna’s blended take-rate sits well below competitors. It also differs from what they report to charge merchants.

Klarna's merchant take-rate is just 1.4% of GMV versus Affirm’s 2.4%, and even including consumer fees and advertising, total revenue is only 2.7% of GMV. This clearly differs from Klarna's stated merchant fee of 3.29%-5.99%, so what's dragging the rate down?

While we discussed EU funding costs being an advantage, the Klarna Card faces a structural disadvantage. Since it runs on Visa rails, EU regulations cap debit interchange at 0.2% and credit at 0.3%, meaning the card take-rate in the EU is substantially lower than in the US.

For Klarna’s integrated checkout products like Pay Later, the dynamic is different. These run on Klarna’s own rails and aren’t subject to the same caps. Still, Klarna likely charges EU merchants fees significantly less than what they charge US merchants to keep rates competitive. Pay-in-Full carries a lower take-rate globally as well, and Klarna likely offers volume discounts to large merchants, trading take-rate for ubiquity. Put it all together, and the low blended rate pencils out.

Still, Klarna commands higher merchant fees than Visa/Mastercard because brands believe it drives sales that wouldn’t otherwise be made.

For most Klarna users, the revenue model is pretty consumer-friendly. 97% of transactions are interest-free, and consumer late fees make up only 11% of total revenue.

For FY25, merchant fees made up 53% of revenue, Fair Financing interest 18%, consumer service revenue 11%, and advertising 7%. In contrast, Discover generated 85% of gross revenue from interest, and even Amex generates 22% from card fees and 32% from interest.

Revenue distribution varies among BNPLs as well. Affirm extracts more from consumers in the form of interest (42% of revenue) and Sezzle earns at least 68% of its revenue from customers, while Klarna earns only 33% of revenue from consumers. But this is also why Klarna has the lowest take-rate among the three and remains the only one yet to reach profitability.

For Klarna’s size, it’s also undeniable the growth rate is impressive. Klarna is the largest BNPL provider by GMV, facilitating $127B in 2025 (+22% YoY) versus $43B for Affirm (+39% YoY). In Q4 revenue grew 32% YoY with US GMV growing an astonishing 43% YoY and US revenue growing 58%—signaling outsized growth in high take-rate products.

Their payment volume is also larger than many emerging neobanks and fintechs. SoFi’s 4Q25 annualized card GMV was $22B, and Block’s GPV, including Square and Afterpay, was only double Klarna’s at $250B. But Revolut is a standout with £1T in transaction volume and just 52.5M customers for FY24, with volume growing 52%. Though their business model is different, focused heavily on foreign exchange.

Part 3 will focus more on neobanks, but it’s important to note the clear difference in business models and margins. Revolut facilitates 10x the transaction volume of Klarna with half as many customers, generating 2x the average revenue per active consumer (ARPAC) but only a 0.31% take-rate.1

In 2024, Revolut’s revenue grew an astonishing 72% to £3.1B with a 35% EBIT margin and 3x as many employees as Klarna. Klarna’s foray into banking hasn’t seen this level of success yet, but its bank account users already generate 4x as much ARPAC as Klarna checkout users.

The question is how much of Klarna’s user base wants to bank with them—especially in the US—and if Revolut and Nubank’s entrance into the US will pose a threat or support neobank adoption overall.

Although Klarna has done incredibly well in the BNPL space, it dropped the ball on checkout and is noticeably behind as a full-suite neobank.

Profits (or lack thereof)

Klarna has been a hot topic for its lack of profitability, with many questioning BNPL economics. Siemiatkowski has also been infamously outspoken on using AI to replace employees and do more with less—decreasing employee count by 49% and increasing revenue per employee 3.6x in three years. From 2022 to 2025, revenue grew 108% while adjusted OPEX decreased 8%. Despite this, Klarna is still unprofitable.

Management argues they’re strategically prioritizing growth over reported income, while analysts are concerned about whether Klarna is a bad business model. Although Klarna was profitable in the past, some believe the US is too risky and costly to be profitable. After entering the US, operating margin fell to -47% and -51% in 2021 and 2022, respectively. Worries over the ‘K economy’ and views of BNPL as a product for subprime borrowers have only exacerbated this concern.

One reason for the stock collapsing recently is the operating margin backtracking from -4.3% in 2024 to -6.6% in 2025—though some of this is distorted by revenues from Klarna Checkout (KCO) and the $73M gain from selling KCO in 2024. Klarna also missed 4Q25 guidance on transaction margin.

Management attributes the reported losses to the growth of their Fair Financing product and immature margins in the US. Klarna’s reported profits are harmed by accounting rules, requiring the entire provision for loan losses to be recognized upfront—different than most accounting rules, which aim to match revenues with expenses.

For Klarna’s Pay-in-4 and Pay-in-30, this wasn’t an issue since losses and revenue were generally recognized in the same quarter. But Fair Financing, with revenue spread across 6-24 months, hurts Klarna's quarterly and annual reported profits the faster they grow.

In the quarter loans are originated, the provision for credit losses outweighs the revenue recognized, resulting in a net loss for the quarter of origination.

Even though that’s not how losses are actually incurred:

Klarna’s push into Fair Financing and the end of ZIRP have also increased funding costs, from 8% of revenue in 2022 to 19% in 2025.

But Klarna still boasts relatively low funding costs. Through their EU banking license and quick loan turnover, the $13B in customer deposits supplied roughly 90% of funding requirements. As a result, Klarna’s funding costs averaged 2.4% of balance sheet liabilities vs xIBOR of 2.5% in Europe. However, off-balance sheet funding has driven up total funding costs to 5.1% of deposits in 2025 from 3.1% in 2023.

Fair Financing is the primary driver of this increase. Unlike Pay Later, which turns over in weeks, Fair Financing loans sit on the balance sheet for months, requiring more external funding. I'll compare Klarna's funding costs directly to Affirm in the next post, but the short-term nature of Pay Later has kept Klarna's costs structurally lower. Now, Fair Financing is shaking up its funding model before revenue take-rate has caught up.

To mitigate capital constraints, Klarna has entered into several forward flow agreements to sell books of Pay Later and Fair Financing loans to third parties. In August 2025, Klarna agreed to sell up to $26B in US Pay Later loans to Nelnet. In March 2026, they doubled their Elliott Management deal to $17B in Fair Financing loans and partnered with Varde Partners to sell $1.7B in Euro loans, supporting $40B+ in total lending capacity.

The acceleration of Pay Later loan sales was the main cause of increased funding costs in 2025, making up $133M of the $164M YoY increase. Most of the Pay Later economics are in the merchant fee, which is collected upfront. By the time the loan is sold, the remaining receivable carries little yield, so it must trade below book value. The fair value adjustment for the loan sales is recognized within funding costs, accounting for a quarter of total funding costs in 2025 and weighing on transaction margin.

“Every additional billion in loans we add in a single quarter will reduce our transaction margin dollars by roughly $25 million in that same quarter due to funding and provisioning costs, but it will increase margin by $60 million in the following quarters.”

- Sebastian Siemiatkowski, 4Q25 Earnings Call

Beyond loan sales, processing fees as a % of GMV also trended upwards from 2024. The US has higher payment processing fees, along with US users more likely to link debit cards to Klarna rather than bank accounts, introducing extra fees for Klarna on every payment. If Klarna succeeds with the neobank strategy, then these costs should be structurally lower as well, with installment payments held in Klarna Balance already.

Return on Equity

For a loan portfolio growing over 100% YoY, it tracks that IFRS accounting makes the growth look unprofitable. And while Klarna's arguments have merit, the ROE math is still worth examining.

If we take Klarna’s 2025 average equity balance of $2.5B, to even achieve a 10% ROE, Klarna would need a net margin of roughly 7%. The current net margin is -8%, and next year’s projections for adjusted operating margin are >6.9%, which after other OPEX and taxes, would still be a loss or break-even at best. Management expects net income profitability within a few years as they move from high-growth to earnings harvesting. But that’s still a couple of years of uncertainty.

Sacrificing short-term profitability for growth also isn’t necessarily a problem, though it means we can’t evaluate the return profile of Klarna for at least a couple of years—something that will probably loom over the stock price.

To reach IFRS profitability, it’s clear OPEX, funding costs, and credit loss provisions as a % of revenue have to come down by several percent. But Klarna’s current lack of profitability and public market pressure puts them at a disadvantage to competitors like Revolut, generating over $1B in profit every year while employing 3x as many people to work on growing the business.

Valuation

Management has laid out long-term targets that, if achieved, would change the picture significantly—a transaction margin of 50% and adjusted operating margin of 25%, compared to FY26 guidance of ~37% and >6.9%, respectively.

If we take these estimates, after other OpEx and taxes, the net margin could be roughly estimated at 18%. Applying 18% to LTM revenue would imply $631M of profits on mature margins.

With a current share price of $12.97 and a diluted market cap of $4.8B, Klarna trades at 7.6x hypothetical mature earnings, 2.8x current gross profit, and 1.8x book value—a book value premium compared to most banks and credit card issuers but lower than JPM. Even if we cut the net margin in half to 7%, it’s trading at 19x hypothetical mature earnings.

At current prices, Klarna is cheap if management's margin targets are achievable and growth sustains. But both assumptions require things to go right that haven't gone right yet. For now, the more useful exercise is identifying what would have to be true: stable or improving US credit losses, Fair Financing scaling without margin deterioration, and card and banking adoption accelerating beyond the current 4% and 13%2 of users.

While these outcomes are uncertain, I believe Klarna has a unique position in fintech and BNPL with strong upside if they get profitable and continue to grow double-digits.

Share

Is BNPL A Better Form of Credit?

Of course, BNPL providers like to say they’re more consumer-friendly and better for overall debt levels than traditional credit cards. But whether that's true depends entirely on which customer you're talking about.

Klarna’s argument does have some support. Credit cards provide revolving credit lines that can easily snowball. Banks, particularly those that target lower-income groups (e.g., Capital One and Discover), are incentivized to approve credit that the consumers can’t pay off in full. Pushing cardholders to make minimum payments and pay interest increases profits. As a result, cardholders are allowed to continue making only minimum payments while spending and accumulating more debt. The opposite is true for Klarna.

For Klarna Pay Later, each purchase creates an individual loan with fixed payments on scheduled dates. Since the loans are interest-free, unpaid loans won’t grow, becoming so large they’re unpayable and wreck the consumer’s credit score. And Klarna isn’t incentivized to make loans that can’t be paid off since the upside is limited to a late fee, not compounding interest.

Underwriting is also better as each transaction is underwritten separately. While credit card users are given a lump-sum balance that rarely changes (besides going up), Klarna’s lending is dynamic. If a user misses a single payment, Klarna turns off future lending until balances are paid in full. Meaning, credit from Klarna can’t snowball.

Consequently, Klarna boasts charge-offs of only 0.4% (1% for the US) compared to 4%+ for credit cards, and users hold an average balance of only $88 versus $6,500.

While the argument of promoting less debt has merit, some of management’s talking points need to be examined:

Addressable Customer Bases

$88 vs $6,500 of credit balance: Klarna loves to show this statistic, but the argument is disingenuous. The highest spenders likely don’t use Klarna. Most users aren’t even approved for $6,500 of credit with Pay Later, and Fair Financing maxes out at $10,000.

Transaction frequency is also much lower than credit cards, making the balances less comparable. But it still does point out the lower debt levels and lack of snowballing. It's incredibly easy to accumulate debt when you can charge everything to a credit card and only face the bill up to two months later.

Pay Later integrated into checkouts has a strong argument for being more equitable to the lowest wealth classes. But the argument of being a better credit option is weak when considering healthy credit users and Klarna’s other products.

Credit card companies have done a fantastic job in the US of positioning credit cards as an easy way to defer payments and earn cash back. For those with enough cash to pay off their cards every month, they likely use credit cards for every transaction and never pay interest. These consumers benefit from cashback while lower-income consumers pay for it.

For Klarna, SKU-level data, in-app delivery tracking, and seamless checkout help offset some of that friction — though as Apple Pay and Link have made low-friction checkout increasingly standard, Klarna's convenience advantage for this segment has narrowed.

The Klarna card trades away many of these benefits, making credit transactions more effortful and eliminating SKU-level data. For cash-tight consumers, though, it offers something more useful.

By combining debit and credit in one card with a 12-hour conversion window, consumers can easily manage cash flow. It promotes less debt accumulation, making credit a conscious choice for each transaction.

Unfortunately, this proposition is again less attractive to wealthier consumers. For these customers, the main argument for convenience is that it’s easier to have one card for both debit and credit than multiple. So, unless Klarna converts reliable credit card users into debit-first spenders, roughly 53% of the US population, that entire segment remains out of reach.

That means their target customers are two categories: people who like paying in debit and only using credit occasionally and those who can’t handle credit/don’t qualify.

For these customer bases, even the Klarna card can be a lifesaver. At least 30% of Americans live paycheck to paycheck. They don’t care about cashback. They want to avoid growing credit balances and be able to put food on the table when cash is tight. Rather than taking out predatory payday loans, they can use Klarna’s card to buy groceries, paying off the loan even a few days later.

Targeting this customer segment isn’t necessarily an issue; CapitalOne and Discover have built large businesses on the lower to middle classes. But these models arguably aren’t as strong compared to premium brands like Amex. It also limits the level of monetization per user across transaction volume and wealth management since the users have less wealth.

There's one more caveat worth flagging on the Klarna card specifically. When you factor in the service fee, the APR for a Pay in 4 transaction could be up to 36%—same as the highest APR credit cards.3

Fair Financing: Klarna’s foray into Fair Financing destroys many of their previous arguments, providing credit card-like loans at similar or higher interest rates with the extra inconvenience of needing to apply for a new loan for each transaction. Fiscally, Fair Financing makes little sense. But mentally, it helps consumers selectively use credit and view each purchase as a separate structured loan.

Similar to Pay Later, this appeal mainly applies to lower-income classes and those who eschew credit cards, resulting in a large barrier for Klarna to reach higher spending, more creditworthy customer segments.

While all these points would suggest quite a low-income specific customer base, Klarna’s customers seem surprisingly diverse, spreading across education levels, location, and stages of life.

Credit Risk

Many critics suggest Klarna’s underwriting is very lax. The main narrative is that people use Klarna if their credit card is maxed out or they aren’t even eligible for one.

But what does Klarna’s underwriting actually look like?

Klarna analyzes over 100 first and third-party data points to assess identity and creditworthiness, making decisions in less than a second. Although they perform traditional soft credit checks, first-party historical payment data is the main driver.

Despite public criticisms, little in Klarna’s lending data indicates lax or loosening standards. Even if some customers have high credit card balances, over 99% of them pay back their Klarna balance. With an average balance of $88, it’s much easier to pay off than a $6,500 credit card. And since Klarna can be a critical source of cash flow support, users are motivated to pay their balance off so they don’t get cut off. This data matters more to Klarna than traditional credit scores.

Machine learning, credit issuance flexibility, and the vast amount of consumer data have led to Klarna having better underwriting, reaching much higher predictability scores versus traditional credit models:

Klarna also benefits from quick loan turnover. With an average loan life of 39 days, they can adapt underwriting models to delinquency trends and interest rates better than traditional credit card and loan providers.

This model is particularly favorable for users who were burned by the US credit scoring system and now struggle to borrow money.

Even in the US, less than 4% of Pay-in-4 users incurred late fees in 1Q24. In stark contrast, ~50% of Americans don’t pay off their credit card balance each month. The notion of subprime seems to be false as well, with the loan-weighted average credit score being near-prime in the US and prime in the UK (2025 20-F, p. 107).

A core metric to watch is provision for credit losses as a % of GMV—showing what percent of transaction dollars Klarna processes that they expect to not get paid back.

Klarna likes to report this metric, but it paints a slightly inaccurate picture since it includes Pay-in-Full and debit card transactions, which aren’t liabilities, and also the fraction of Pay Later that is paid immediately. Using Klarna’s metric, the ratio was just 0.63% for 2025.

To get a better picture, we can make a rough estimate of loaned GMV (based on the GMV splits by product). This shows the provisions for 2025 at closer to 0.91% rather than 0.63%. Still meaning that Klarna expects to lose less than 1% of all loaned GMV. They commonly compare this to an average consumer loan charge-off of 2.89%. Capital One’s charge-offs are even higher at 4.93% for domestic cards in 4Q25, 4.08% for personal loans, and 5.29% for international cards.

By all measures, Klarna is lending to supposedly risky consumers with much better success than traditional offerings.

With the increase in Fair Financing and higher growth in the US, it’s expected that the provisions will continue to go up. This isn’t inherently bad, as longer-term loans do have more risk but carry better yields.

We can see this trend with Pay Later vs Fair Financing charge-off rates. Group Pay-Later charge-offs are only 0.4% versus ~1% for the US, while Group Financing is 2% versus 4% for the US. Even with US fair financing, that’s still no worse than Capital One’s consumer loan charge-off rates and better than their credit card charge-offs. The question is how much of the higher US charge-offs are due to the supposed greater credit risk of US customers and how much is from Klarna intentionally taking on more risk to grow US adoption and improve underwriting.

Beyond product-level comparisons, the headline charge-off ratio itself warrants scrutiny. Klarna focuses on cumulative charge-off ratios and provisions to GMV, while not mentioning standard annual charge-off metrics. This is for good reason.

Most lenders report charge-off ratios by dividing net charge-offs by average loans outstanding for the year. For a lender with long-term or revolving credit, this metric makes sense—it conveys how much of the outstanding loan base the lender charges off each year.

For Klarna's model, the metric isn't meaningful in the same way. Dividing annual losses by average loan balance effectively compares a full year of charge-offs to roughly one month of loan volume, which is why the 6.4% net charge-off ratio reported in their 20-F is misleading at face value. It simply can't be compared directly to credit card charge-off rates.

Growth Prospects and Neobank Strategy

Klarna has inked some landmark deals recently, including integrating with Google and Apple Pay in several large markets, an exclusive partnership to handle the backend for Walmart’s OnePay, partnerships with several leading PSPs, and becoming a payment option for Airbnb and Uber. Surprisingly, even with enterprise deals that likely have lower take rates, Klarna’s GMV take-rate has continued to increase from 2.29% in 2022 to 2.76% in 2025.

The pace of large deals is remarkable and shows Klarna is becoming a respected and important part of the payments ecosystem. Partnering with many of the leading PSPs alone adds over $9T in processing volume for Klarna to tap into as the preferred BNPL. But expanding beyond BNPL means competing on new turf.

As Klarna adds fair financing, debit cards, and works to become a financial assistant, its path converges with neobanks like Revolut, SoFi, and Nubank. Competition is heating up as these companies begin to fight for a share of the US market. Incumbents like JPM, Visa, and BofA are also vying to maintain and take share in the new fintech era, and at the same time, undoubtedly lobbying for the neobanks to be more heavily regulated.

Klarna has already demonstrably grown adoption of its banking product, as shown below. These users purchase on Klarna more often, generate 3.5x more revenue, deposit 7.4x as much, but also carry higher charge-off rates and outstanding balances. Although they might be riskier overall, the revenue and funding provided more than offset that.

With Klarna Banking growing at 3x the rate of total users, the only question is whether it becomes a mainstream offering or remains a niche product for a subset of power users.

Beyond banking, there is also still room for Klarna's core payment products to grow:

• Many merchants still don’t offer all of Klarna’s payment options, namely Fair Financing, at 20% penetration.

  • It likely won’t near 100%, however, since Klarna is over-indexed toward low-value items like apparel. But, Fair Financing should help Klarna expand further into high-ticket item retailers.

• Google Pay and Apple Pay continue to roll out for more countries, including the UK announced in February 2026.

  • As more people become aware of this option, I expect BNPL usage to grow significantly.

• Rapid expansion of active users means there’s runway for newer cohorts to increase spend.

• GMV take rate will likely increase as high take rate countries (e.g., UK and US) and products continue to grow faster than the overall company.

  • But the US also has higher processing and servicing costs, actually leading to a lower transaction margin. Eventually, this should level out as Klarna scales further in the US.

• Fair Financing growth will have a disproportionate impact on revenue growth due to its better economics. Fair Financing is 9% of branded GMV, but Fair Financing interest is 18% of revenue.

• Penetration is still low in newer geographies. While it won’t likely get to Klarna’s 85% penetration in Sweden, it is reasonable to think US penetration can grow from 11% to maybe 20%.

• The instant activation on Stripe and other PSPs provides a strong tailwind, making Klarna the preferred choice for the millions of stores powered by Stripe.

Ownership

One of the remaining questions is ownership. As mentioned, Siemiatkowski still runs Klarna, while all his co-founders have departed. But what’s interesting is who holds significant stakes in the company.

Siemiatkowski and co-founder Victor Jacobsson control a combined 13% of voting rights, Sequoia still controls an astonishing 23%, a family-owned holding company Heartland A/S holds 9%, and Commonwealth Bank of Australia owns 5%.4 With ~50% of voting rights held by long-term shareholders, Klarna has the backing to prioritize long-term growth over short-term profits.

Further signaling board confidence, Michael Moritz, a famed venture capitalist and chairman of Klarna, acquired 3.47M shares for $50M in March 2026 at roughly $14-$15 per share. Moritz has been a longstanding supporter of Klarna, and while insider transactions shouldn't be over-interpreted, his purchase alongside Sequoia's continued large stake signals conviction in Klarna's long-term trajectory.

Klarna’s Potential

Klarna’s addressable market is enormous—$20 trillion in GMV with $9 trillion in the US alone, 157x Klarna’s current global volume. Just as there were many winners in the previous era of banking, the same will likely be true again. The question becomes, of course, which companies will thrive.

If Klarna’s track record of growing population and wallet penetration while decreasing loss provisions continues to hold, then it’s quite feasible Klarna could be several times its current size. PSP partnerships continue to ramp, contributing hundreds of thousands of new merchants already, with many more to come.

Winning over traditional credit users in the US remains the hard part, though. Without that, it will be difficult for Klarna to become larger than a small subset of the payments industry. But their path to changing that runs through holding customer balances and facilitating debit transactions at any merchant that accepts Visa.

Klarna also doesn’t have to be a leading bank for it to be a good growth story. Most third parties project BNPL to continue growing at double-digit rates for the next decade, and Klarna has already started to prove its banking foray is attractive to some.

Klarna's ceiling is ultimately a monetization question. Their products resonate most with consumers who dislike credit cards or can't qualify. It's a large segment, but one Klarna has struggled to monetize as effectively as Sezzle, let alone the premium customers Affirm has built its model around (as we'll see in Part 2). Klarna's strategy is to close that gap by increasing spend per user. Whether that's enough to move the economics upmarket is the central question for the next few years.

Now that we have some insight into how Klarna makes money and where their model has limits, the next post will compare Klarna to Affirm, their largest competitor, and Sezzle, a much smaller BNPL with a controversial past and a very different answer to the monetization problem.

Share

Interested in more?

Read my most popular posts:

• Topicus: Falling Knife or Undervalued?

• Dino Polska Update

My thoughts on Uber and Waymo

Understand my approach to investing (though it’s arguably evolved since): Types of Companies I Invest In

We’ll see how the stock actually trades tomorrow, but it is down 21% after-hours to low $90s per share.

After listening to the earnings call, I believe the following quarters will be rough as a shareholder, but I am still optimistic over the long term.

The Ugly

Guidance is obviously below what many investors and sell-side analysts expected. Although it was communicated in Q3 that management was going to focus on DAU growth over monetization, it was arguably undercommunicated how much growth could slow and margins contract. The most telling figure is 10-12% bookings growth guidance, meaning growth is expected to slow further.

From the earnings call today, management also did not expect this level of deceleration when they spoke last quarter. They had already seen a negative trend in DAU growth, sparking the shift to focusing on DAUs over monetization, and that negative trend continued through Q4.

Shifting towards DAU growth means improving the free tier features, lowering free tier friction (which lowers conversion), and spending more on R&D and marketing. Consequently, margins are expected to contract. Expected FY26 Adj. EBITDA margin is 25% versus 29.5% in FY25 and 25.7% in FY24.

What happened?

Analysts pointed out, and CEO Luis von Ahn agreed, that Duolingo had pushed too hard on monetization. Von Ahn was forthcoming that a large amount of paid conversion was driven by increasing friction (i.e., ad-load). Unfortunately, the conversion through friction strategy can only be taken so far. While friction converts as many current DAUs into paid subscribers as possible, it worsens the experience for new free users. So, future DAU growth slows and monetization eventually follows. Their new strategy is to try to flip the equation back, where DAU growth accelerates and monetization eventually follows.

The Bad

Firstly, I don’t think the outlook is positive for the stock in 2026. Although I like the long-term strategy and management’s honesty, weaker growth and margins are unlikely to lead to favorable results for investors who held before this latest drawdown. Growth will have to decelerate throughout the year to reach the 15-18% revenue growth numbers. Company projections are modeling 25% growth in Q1, an expected slowdown in Q2 and Q3, and then a moderation in Q4.

Von Ahn and the new CFO, Gillian Munson, said that early results of their DAU efforts may show up in the revenue numbers by late 2026 and more so in 2027 and 2028. As a result, I expect bearish sentiment until the results begin to materialize.

The other “bad” was my strategy and investment decision related to Duolingo. Although I am still confident in the company, its differentiation, and especially the management team, the way I entered the investment could have been better. It resulted in lost time and a higher cost basis than necessary.

It’s always hard to balance taking advantage of what you believe is mispricing and taking a moderate approach to adding a position. When I wrote about the company, sentiment was dropping quickly, and I got overly excited about the valuation and long-term prospects without considering the near term implications of momentum and the shift towards DAU growth over monetization.

It would have been prudent to wait and see where the stock price settled, and even consider waiting until Q4 results. Of course, it’s easier to say this in hindsight, but it was definitely at odds with my typical investment strategy, and it provided an important learning experience that I can build off of for future investments.

Despite the misjudgement, Duolingo is one of my smallest positions, which softens the blow and leaves room for additions in the future if the strategy succeeds.

The Good

News is not all bad. As always, management was quite honest about the situation and answered investor questions thoroughly, without dancing around it. They provided good direction on their strategy in the next twelve months and set quite an ambitious 2028 target (though I’m skeptical).

Historical Context

With the current situation, it’s important to remember that no company’s growth trajectory is smooth sailing, and Duolingo’s hasn’t been either. Although public market investors are used to the 40%+ DAU growth of recent years, this isn’t the first time Duolingo has experienced significant deceleration.

In 2018, DAU growth had slowed to single-digits YoY. But through the creation of a dedicated gamification team and continuous innovation on features like streaks and leaderboards, they increased DAUs by 4.5x in the following four years. And then another 4x in the 3.5 years from 2Q22 to 4Q25. That’s incredible reacceleration from single-digit growth in 2018.

The FY26 guide of 20% DAU growth, while a step down, provides wiggle room for continued deceleration and experiments around monetization and friction.

Language Learning Growth

On the DAU end, management plans to test lower friction (i.e., ad load) while continuing to invest in lesson types and getting the top 9 languages to a Duolingo score of 129 (the level needed for a job). Some of the lesson format and content changes are already appearing in users’ accounts, with a significant reworking of the lesson pathway.

Some of the monetization team is now dedicated to top-of-funnel user growth, and the company is looking for ways to diversify their monetization drivers, for instance, better features in paid tiers over friction. Recently, they’ve been testing a feature that lets users lock other apps until they complete a certain number of lessons that day.

They have also finally decided to put more effort into the ads they display. Von Ahn discussed a shift away from network ads, which are generic, low quality, and low revenue per impression. They’ve begun direct partnerships to display ads for specific companies (which generate much higher revenue) and are also experimenting with showing ads in the language that the user is learning.

Within the FY26 EBITDA target is room for introducing the AI features into lower tiers. ‘Explain my answer’ has already been added for all users, and the costs of Video Call have come down enough to test adding it into Super. Depending on the A/B test results, this could mean metered use on the Super tier or even full access and somewhat higher Super pricing if incremental revenue is high enough.

Saturation and Competition

One analyst also asked about the potential for market saturation, and here’s what Luis had to say:

“For example, in the United States, 2% of all internet users on a given day use Duolingo, so 2% penetration in the U.S. for daily users. In the U.K., that’s 3%, in Germany, that’s 4%, Germany is not the highest country we have. We actually think we can get much higher than that. Even if we only assume that every country got to 2%, which is the U.S. penetration, we would more than double our daily active users. I just don’t think we’re near saturation at all.”

Besides saturation, Von Ahn argued competition wasn’t a large concern. Duolingo not only dominates language learning apps in terms of DAUs (~85% of the total market), but also benefits from the fact most new entrants fail to compete. A freemium strategy is incredibly hard to scale profitably with Duolingo already existing, resulting in most either failing to reach scale or retreating behind a paywall, which inherently caps their DAU growth.

Other Courses

Chess is now at over 7 million DAUs (many of which are also language learners), making it the second-largest chess app/website behind Chess.com (which is estimated at around 10-11 million DAUs). Most of this adoption has been through word of mouth, as Duolingo doesn’t even show up as a chess app on the App Store.

Von Ahn also believes that Duolingo will be the best tutor for math this year. Math is attractive because of the higher willingness to pay from the parents of the K-12 kids who use the app. It fills a similar role as after-school math services, but is cheaper and more flexible.

Growth Targets

What’s really fascinating is the DAU target for 2028 of 100 million, implying a 23% CAGR from 4Q25 and an acceleration to 25%+ annual growth after 4Q26. Duolingo has so far proven an ability to manage towards metrics pretty well, and I wouldn’t be surprised if they do reach this goal. It won’t be easy, though.

Management is framing this as: no matter how you cut it, the business will be much bigger if they reach this goal. Munson provided hypotheticals that the business could reach $2.5B of revenue and $700M+ of EBITDA in a few years with the pivot, while keeping previous margins and monetization would yield a business closer to $1.5B in revenue and $400M+ in EBITDA. Of course, nothing to hang your hat on, but it shows how they view the opportunity (or at least what they want to tell investors).

On top of this, the board approved a share repurchase program of $400 million. I don’t like how some investors get overly excited, believing share repurchase programs are signals of management thinking their stock is undervalued. Regardless, if you believe the company will do well on its growth initiatives, then the share repurchase program will be a good allocation of capital.

Lastly, this was new CFO Gillian Munson's first earnings call. Having started just days before, she came across as well-prepared and aligned with the strategy ahead. I’m interested to see how this growth strategy plays out and plan to hold my shares for now.

Disclaimer: This is not investment advice. Any statements made are the opinion of the writer and are produced for informational and entertainment purposes only. The writer may buy, hold, or sell positions in any of the companies discussed.

Read the full disclaimer here.

Subscribe now

Autonomous rideshares (aka robotaxis) have catapulted to the spotlight. Waymo is quickly expanding, offering rides in ten cities with plans for eighteen more, and Tesla and Amazon Zoox are now running in two cities each.

For anyone who’s had the chance to ride in a Waymo, it’s truly incredible. It drives surprisingly smooth and can detect people and objects that a human could never. For riders, there’s more privacy and control over the experience. You choose the music, have private conversations, and the driving is pretty consistent.

Similar to the early days of ridesharing, numerous companies are competing with different form factors and business models, all trying to win early customer adoption. Beyond the US, other rideshare and AV companies are treading a similar path, mainly launching across China. Although Waymo is ahead in the US, it’s clearly too early to decide the winners.

On average, prices for Waymos are still higher than Uber/Lyft, but this could change. Hardware costs are continuing to drop, and scale will help cover Waymo’s R&D costs. As autonomous rideshare becomes more prevalent, it’s not hard to imagine ten years down the line, most would prefer a robotaxi over a human driver.

Given this backdrop, it’s unsurprising investors are worried about AVs displacing Uber. Uber has long recognized the importance of AVs for ridesharing, starting an internal research group in 2015. However, with an unsuccessful strategy and desire to refocus the company, they sold the division in 2020. Uber has since been cementing AV partnerships across the globe, though much of it feels like playing catch-up. They arguably also dropped the ball on AV data collection, only launching an effort this past month.

So far, the three robotaxi companies that have launched in the US all used independent apps rather than Uber, leaving many to question if Uber even fits into the equation. Interest in robotaxis is already strong and growing quickly, with competitive pricing and a better experience. But can AVs actually take over the rideshare market?

Does Waymo Need Uber?

Waymo (and AVs in general) are at an interesting point strategically. These companies want to scale, are constrained in how fast they can scale, and have established networks they can scale on (e.g., Uber). The technology is even compelling enough to pull users onto entirely new platforms, like Waymo One.

But the already intense competition and established networks question whether owning the network should be the goal. Do these companies focus on AV technology across use cases, or do they seek to become the new rideshare winner? Or both?

I believe Uber, for several reasons, is positioned to take a great share of the robotaxi market, and Waymo recognizes the need to partner with them.

Scale and Network

Market fragmentation among robotaxis raises an important question of convenience. Just as it would be impractical to have five or ten rideshare apps, having that many robotaxi apps wouldn’t work either. Most people willing to pay for rideshares won’t go through the inconvenience of checking five apps for prices and ETA. I’m very frugal, and even I could only imagine checking 2-3 apps at most. Unsurprisingly, this is how human-driven rideshare played out: the top 2 rideshare apps hold majority share in each market.

This is Uber’s value proposition as a network. Uber serves as a supply-demand aggregation tool, connecting riders with drivers. But networks only work if there is ample supply on both ends to make it readily available and profitable. Although the economics are different for autonomous rideshare, reaching profitability in a competitive rideshare market is hard. It took Uber fourteen years and $37 billion in revenue to generate positive free cash flow—and they’re the largest rideshare app in the world!

It’s reasonable to assume a similar dynamic will play out in autonomous rideshare—one or two winning apps per city. But that doesn’t mean the AV industry itself is a winner-take-all market, as I’ll discuss later.

The question becomes who will own the networks and how many AV companies can succeed?

“The goal is to…realize the positive impact of [this technology] and get it to as many people and as many places as possible. There’s some things we know we want to do internally, which is building the core of the Waymo driver. Then other things, like building the Waymo car or how we partner in terms of running the service, that we are open and very practical, all in service of getting it out there and to as many people as possible. Uber has been a great partner in that vein.”

—Dmitri Dolgov, Waymo co-CEO, (podcast)

Waymo has recognized the network constraints laid out and what’s needed to scale. In Phoenix, Waymos have been available on both the Uber and Waymo apps. In 2025, they launched in Atlanta and Austin solely on Uber. In those markets, Uber riders can turn on the AV preference and get randomly matched with a Waymo if one is available.

According to Uber, this test has gone well. Waymo utilization and ETAs in Atlanta and Austin outperformed markets where it is offered solely on the Waymo app. This is unsurprising. Uber’s demand and ride matching far outstrips Waymo’s supply, basically guaranteeing a steady supply of rides for Waymo.

It also makes sense given the competitive environment and Waymo’s supply constraints. When Waymo launches on its own app, most users are early adopters who actively seek the service. This works in San Francisco, where Waymo has a lot of supply, and most people are tech-forward early adopters. But this strategy is reputationally harmful if they launch with a few dozen cars in cities like Atlanta or Dallas.

Launching on the Waymo app is inefficient for reaching the average user. Early adopters with a higher willingness-to-pay suck up supply, drive up prices, and limit the average rider from experiencing a Waymo. With competitors like Tesla and Zoox entering the market, Waymo wants as many people as possible to experience their vehicle first. For many markets, offering on Uber is the way to do so.

Another factor in reaching widespread awareness is price. In Atlanta and Austin, Waymo's supply is folded into Uber’s overall supply-demand equation, meaning Uber offers Waymos at the same price as regular Ubers. As a result, there’s no premium pricing barrier that would drive the cost-conscious AV-wary users away.

Instead, riders will randomly get matched with a Waymo on the Uber app and say, “Why not?” Much less friction. For robotaxis to become mainstream, embedding them in Uber seems like the fastest path.

Beyond customer awareness, there are capital considerations. Waymo scaling isn’t just reaching customers; it’s deploying vehicles, managing and maintaining the fleet, and providing customer support, several of which Uber is helping with.

The Uber strategy does have drawbacks, though. In Atlanta and Austin, those eager to ride in a Waymo were frustrated by not matching with one. People even ordered and cancelled rides—trying to match with a Waymo—so much so, their Uber account got banned.

But it seems the supply frustrations would have occurred anyway, as Waymo first launched with as few as 30 vehicles in those markets. In Atlanta, the fleet is still only 100 Waymos, versus 1,000 in SF and 700 in LA. As Waymo continues to announce new cities, despite low supply in current cities, their strategy is clearly to go wide and shallow. Getting a foothold in as many markets as possible and building up city-specific knowledge and regulatory approval, rather than trying to take a majority share immediately. Launching on Uber makes much more sense in this light.

Utilization

Deploying on Uber also recognizes the massive variability in demand.

I believe that unless autonomous vehicle costs come down by many multiples, it is unrealistic that AVs could fully supply the rideshare market. And even if costs did come down, Waymo clearly can’t deploy vehicles quick enough yet. This makes it unlikely Waymo will take over the rideshare market as an independent app.

Here’s the issue with utilization:

In the latest quarter, Uber provided an example of demand variability. Demand on Saturday is more than double Monday, and daily troughs are just 5% of peaks. Uber’s human driver model works because supply is also variable, leading to fair prices and availability day and night.

Waymo, in contrast, is supply-constrained. The only way to scale supply is to build more Waymos. As seen below, a Waymo fleet works well for providing a steady supply base, but doesn’t make sense for anything above the weekday average.

So, if multi-app is inconvenient and low supply drives up prices and wait times, then deploying on independent apps will hinder mass-market adoption. To reach mass adoption, I believe AV companies either have to fully supply a market or partner with Uber to supplement it.

Let’s imagine a robotaxi company decides they want control; they don’t want to pay Uber’s 20% take rate. The company will aim to serve all the demand—since no one wants a rideshare service that doesn’t work well when demand is high.

After all, companies like Waymo and Tesla are technologically advanced and own the autonomous technology; why wouldn’t they own the app?

Here are two approaches they could take:

Option 1: Build and deploy enough vehicles to serve the highest peaks (e.g., the Taylor Swift concert on a Saturday night). To achieve this, you could estimate 55% of the vehicles would be sitting idle most of the week (or all the vehicles are running at much lower utilization).1 To be profitable, the combined ROI of weekday demand and higher weekend demand must compensate for sub-50% utilization (even the vehicles to serve Monday’s peaks are idle part of the time).

Waymo’s main operating costs are fixed, meaning profitability relies on utilization. Since utilization drops in this scenario, ride prices must increase to maintain ROI. But the issue with higher prices is that demand would shrink since most consumers don’t want to pay much more than the price of an Uber. As a result, the robotaxi offering gets less adoption and has poor economics, so most people will still take Uber.

The AV company would also have to employ enough people to handle maintenance and emergencies on the entire fleet, especially at peak times, driving costs up further.

Option 2: The AV company could onboard human drivers to service the peaks when demand outstrips supply—basically becoming Uber + AVs but building it from scratch to own the platform. This would mean the robotaxi company—whose core competencies are AV tech and now matching riders to those AVs—would need to develop a whole separate arm to handle wages and take rates for drivers, insurance, bad driver complaints, and all the lawsuits that come along with it. Now, a company that began as an autonomous technology company has to become an application developer and a gig-work company, going through the turmoil Uber has experienced in the last sixteen years.

These two scenarios are clearly antithetical to Waymo’s mission of making the technology as widely adopted as possible.

Option 3: The robotaxi company realizes that neither of these scenarios is best for autonomous adoption, the AV company’s scale, or the riders. Instead, it makes economic sense for robotaxis to be just part of the supply side, letting a company like Uber handle human driver relations, fulfilling demand at peak times, and interacting with customers. Vehicles reach even higher utilization than on the Waymo app, and more people use the product.

Once you factor in Waymo’s struggle in rain, snow, and power outages, it’s unthinkable that Waymo could control a rideshare market. Any harsh weather drops their utilization significantly, and the costs are only exacerbated the greater their vehicle base is in a city. This could change one day, but that will take many years and billions of dollars more of capex before that happens.

To summarize: Deploying AVs on Uber is optimal as it matches the AV’s fixed costs with the relatively fixed demand levels, and matches variable cost human drivers to the highly variable portion of demand.

The Uber Tradeoff in Numbers

To illustrate the tradeoffs, let’s consider a hypothetical rollout on Uber vs satisfying full demand independently.

Let’s say Waymo has the capacity to produce 400,000 vehicles to cover all the cities in America up to the average weekly demand (roughly illustrated with the red box below). This means they service the relatively stable levels of demand, but not the excess weekend demand. Since they can’t cover peak demand, assume Waymo deploys on Uber and pays Uber a 20% take rate.

Using a conservative $15/trip, each Waymo currently generates $140k in bookings per year. Since the 400k vehicle count could reach similar utilization, 400k vehicles would generate $56 billion in bookings. After Uber’s take rate, Waymo would generate ~$45 billion in annual revenue.

Now, let’s assume Waymo took the same 400k cars and deployed them in 45% of the cities to reach peak demand and fully supply the market. Those vehicles would be utilized maybe 35% of the time compared to Waymos at full utilization. But since these vehicles only run at peak times, assume they generate 1.5x the average price per trip of full-utilization cars. While the full utilization cars generate $140k per car, the weekend peak cars reach $75k each, averaging out to $105k between the two car groups.

Since Waymo only had the capacity to deploy 400k in both scenarios, gross bookings drop from $56 billion in the Uber example to $42 billion in the peak demand scenario. This is lower than the net revenue of $45 billion that Waymo would generate after Uber’s take rate. If you take into account the extra costs of running the platform that Uber normally covers, say 15% of revenue, comparable earnings drop to $36 billion.2

As a consequence, Waymo is present in half as many cities and generates 20% less revenue after platform costs to allocate to R&D, operations, and profits.

Of course, this oversimplifies reality, but it should help portray the opportunity cost of Waymo trying to take over a rideshare market versus partnering to grow in more markets.

The R&D that goes into Uber’s network is also underrated in the era of AVs. Since AVs don’t care about the time, distance, or endpoint of a ride, algorithms can be significantly reworked to optimize ride matching for these vehicles. They could eventually be doing long-hauls at night when demand is low, and be more efficiently spread across the city at peak times.

The Uber Tradeoff is Like Android (kinda)

AV companies face a similar choice to the one phone manufacturers did in the late 2000s. When Apple launched the iPhone, it sparked a consumer shift towards well-designed UIs and a developer ecosystem. In response, competitors like Samsung shifted from an internally developed operating system (OS) to Google’s Android in 2009. Others, like Xiaomi, started on Android from the beginning. Nowadays, nearly all leading phones run on Android (e.g., Samsung, Xiaomi, Oppo, Vivo).

By choosing Android, phone manufacturers gave up some control over the OS design, app store, app revenue, and exclusive data collection—but they saved millions of dollars and years of development time. More importantly, the choice prevented network fragmentation.

Imagine if every phone brand had its own OS. Developing Apple and Android apps is already two separate efforts, so adding specialized Samsung, Xiaomi, Oppo, and Google Pixel versions would have been untenable. Most app developers simply wouldn’t bother. The phones without strong app ecosystems would have either failed outright or pushed even more consumers toward iPhones, which were differentiated enough in hardware and software to justify a walled garden.

Deploying on Uber can be seen as the Android of ridesharing. Just as Google provided Android for free—or even paid manufacturers to adopt it—Uber is subsidizing Waymo's deployment, providing fleet management and currently operating rides at a loss. The parallel runs deeper than subsidies, though. Android's real value wasn't that it was free; it was that it unified a fragmented market.

By rallying manufacturers around a single platform, Android gave app developers one ecosystem to deploy on, which attracted more users, which attracted more developers. Uber offers robotaxi companies the same flywheel: a single platform with massive existing demand, so AV companies don't have to build their own user base from scratch while competing against five other standalone apps for the same riders.

Waymo gives up some platform control but gains the fastest path to scale while staying focused on its core competency: building the best autonomous driving technology, not managing a rideshare network.

Waymo’s co-CEOs have been explicit about this preference, talking about their desire to apply autonomous technology across industries rather than becoming a fleet and ride-hailing manager. Uber envisions a similar reality where, just as it doesn’t own the cars human drivers use, third parties will own and manage robotaxi fleets—making the value of a single platform even greater.

The 3P structure also sets up robotaxis to be a multi-winner market. If Uber directly owned and deployed autonomous vehicles, there'd be little reason to partner with more than one or two AV companies. But with third-party fleet operators on the platform, it starts to resemble the car market itself—different brands at different price points serving different preferences.

AV companies could operate as technology licensors, where manufacturers combine their autonomous software with a vehicle, sell it to a fleet operator who pays recurring fees for the algorithms, and then deploys on Uber. One fleet manager might run Waymos, another Lucids, a third Zoox. And similar to Google releasing the Pixel phone alongside its Android partners, Uber is already launching its own robotaxi with Nuro and Lucid this year.

Amazon has proven out this tradeoff in e-commerce. While they initially went hard on 1P, they’ve realized the choice and capital efficiency from a 3P network is optimal.

The result is a market where a rising tide lifts all boats—many AV companies sharing in the profits and focusing on their core competency, rather than one winner taking all, or too many fragmented offerings leaving everyone worse off.

The Bear Case (for Uber)

The autonomous rideshare market is still so nascent that predicting its structure is inherently uncertain. Hardware costs will change, regulations will evolve, and competitors will emerge and die. Much of the bull case for Uber rests on assumptions about cost structures and competitive dynamics that could look very different in five years.

The main bear argument is that Waymo—or another AV company—simply doesn’t need Uber. Especially once supply catches up. Today, Waymo’s constraint is vehicles, not demand. But Alphabet has the balance sheet to change that. If Waymo scales its fleet to hundreds of thousands of vehicles and vehicle costs fall meaningfully, the utilization math shifts. A world where AVs cost $30,000 instead of $200,000 makes idle vehicles on off-peak days far more tolerable. Suddenly Uber’s 20% take rate looks like an expensive tax on a company that could capture that margin itself.

Then there’s Tesla, which may pose a direct structural threat to Uber’s model. The utilization argument for Uber assumes that AV fleets are a fixed supply—you build vehicles, and they either sit idle or they don’t. But Musk’s strategy flips this entirely. With millions of vehicles already on the road and owners eventually able to lease them out as robotaxis when not in personal use, Tesla could replicate the variable supply model that makes Uber work today—except the “drivers” are parked Teslas earning money for their owners.

This would give Tesla’s network the same elasticity as Uber’s human driver pool; more supply comes online during peak demand because more owners opt in when prices rise, and it recedes when demand falls. If FSD reaches sufficient reliability, Tesla wouldn’t need to choose between Option 1 and Option 2 from the utilization section—they’d have a third path that sidesteps the fixed-cost problem altogether, built on an installed base of vehicles that no competitor could match overnight.

More broadly, Tesla represents the integrated stack argument. Apple proved that owning hardware, software, and distribution creates a defensible moat that platform-dependent competitors struggle to match. Tesla builds the vehicle, develops the autonomy software, and already has a consumer brand and app. If FSD reaches Waymo-level reliability, they could undercut everyone on cost, given their manufacturing scale, and Musk has already indicated no interest in partnering with Uber.

Leadership intentions can also change. Waymo’s co-CEOs say they want to focus on the technology layer, but that’s easy to say when you have 100 cars in Atlanta and need Uber’s demand. With 10,000 cars and a proven product, the calculus is different. Waymo could decide that owning the customer relationship—and the data that comes with it—is worth more than the incremental utilization Uber provides.

Finally, the product could be so compelling that people go to Waymo first, even if wait times are worse. Waymo could service average demand on its own app, and riders would be fine switching to Uber only at peak times when Waymo prices spike. Eventually, the scale would grow large enough and costs fall enough that Waymo could target the peaks too.

Rebuttal

Although the bear argument is feasible, I believe it is overly optimistic and ignores competition and Uber’s competitive position. Besides competition among robotaxi companies with standalone apps, there are also autonomous vehicle programs targeted at supplying existing rideshare companies, including the Uber partnership with Lucid and Nuro, and DiDi’s recently launched vehicle in China.

The vehicle below will be exclusively launched on Uber in late 2026, owned by Uber and third-party fleet operators. Assuming the technology works well, this sets Uber in a very advantageous position, capitalizing on the world's largest ridesharing user base: 202 million monthly active users, of which 46 million are UberOne members.

Those UberOne subscribers are particularly sticky. They're already saving on rides and Uber Eats deliveries, giving them little incentive to switch to a standalone robotaxi app for one slice of the value they get from Uber. At least some AV companies will recognize this and choose the Uber deployment route, weakening the position of standalone AV apps.

If Waymo wants to get ahead, they need to establish the greatest scale and widest user awareness, which arguably can’t happen solely on its standalone app. I don’t believe the rest of the market is far enough behind for Waymo to build enough scale to be the one winning independent app.

Now, for the Tesla thesis. Although Musk has espoused ambitious goals and strong claims of Tesla’s autonomous capabilities, reality has fallen quite short. From Robotaxitracker.com, Tesla’s fleet in Austin is still under 100 vehicles, and a very small fraction are unsupervised. Service availability is also incredibly spotty. Compared to Waymo’s near 100% availability, Tesla Robotaxi is often below 50%. Early anecdotes also suggest the Robotaxi experience is noticeably worse than Waymo's. Nonetheless, it is still early days for AVs, and Tesla Robotaxi is just beginning.

The claimed strategy of leveraging existing owners' vehicles, while compelling in theory, faces real practical hurdles. Current Tesla owners skew toward higher household incomes than general vehicle buyers. This demographic is unlikely to be motivated by generating robotaxi income. Beyond finances, there's the friction of unsupervised strangers, wear, cleaning, maintenance, and simply not having your car when you may need it.

There are also open questions about whether Tesla's camera-only approach can match the safety and reliability of lidar systems like Waymo's, which would affect both regulatory approval and owner willingness to participate. And even if all of these issues are resolved, timing works against Tesla. By the time owner-deployed vehicles become feasible, multiple competing AV taxi options will likely be available—recreating the same fragmentation problem that favors a unified platform like Uber.

Lastly, weather limitations, occasional outages, and a lack of penetration in lower-income areas will prevent an AV-only model from being feasible for years to come.

The Future of Ridesharing

If hardware costs continue to decline and AVs achieve the utilization levels discussed above, ridesharing will become significantly cheaper and more accessible. Even ignoring robotaxis, Uber seemingly hasn’t reached market saturation, growing MAUs by 18% YoY and trips by 22% YoY. With AVs promising more consistent service, lower prices (eventually), and better reliability/ETAs, it’s reasonable that the ridesharing market could be multiples of its current size.

In parking-constrained cities like Los Angeles and San Francisco—where you’re sometimes paying $10-20 to park—more people will choose to take a robotaxi over driving. This creates a flywheel: as AV volume grows, costs decline. Cheaper robotaxis get used more widely, lowering the overall cost of going out, which increases how often people go out, driving trip volume even higher.

Lower prices growing the overall pie isn't theoretical. Sam Walton proved it with Walmart, RyanAir and Southwest did with flights, and AWS did with cloud computing. Ridesharing is likely next.

Uber has already observed how AVs impact demand in Austin and Atlanta, where new and current users have increased trip frequency faster than in most markets. Interestingly, Uber noted that its trip volume actually accelerated in San Francisco, a market where Uber doesn't even offer robotaxis and Waymo operates independently. The presence of autonomous rideshare may be expanding the overall market, not just redistributing it.

As I see it, autonomous vehicles will have a neutral-to-positive impact on non-autonomous ridesharing in the next five years. Many cities will likely not have robotaxis or be severely undersupplied. Scaling to sparse geographies and cities with varied layouts and demographics will take longer than realized. By the time robotaxis become profitable, affordable, and deployed in most cities, that will provide a second boost to Uber’s growth as regular rideshare growth slows down.

Robotaxis could actually be lengthening Uber’s opportunity for above-GDP growth, far different than eliminating their existence like some bears would argue.

Stay tuned, as a second post will likely come out on Uber soon, focusing on the non-autonomous growth runway and valuation.

Disclaimer: This is not investment advice. Any statements made are the opinion of the writer and produced for informational and entertainment purposes only. The writer may buy, hold, or sell positions in any of the companies discussed.

Read the full disclaimer here.

Thank you for reading my work; every subscription, like, and comment helps me continue to produce these reports.

Join Investor's Odyssey

Share

Interested in more?

Read my most popular posts:

• Topicus

• Dino Polska Update

Read my deep-dive on Amazon: Amazon Part 1, Amazon Part 2

Understand my approach to investing (though it’s arguably evolved since): Types of Companies I Invest In

Since I gave a brief value update on Topicus last November, the share price has continued to drop, down 45% from the peak in August 2025 and 23% from November. This is part of a broader carnage across public SaaS equities.

It is undeniable that there is now more risk in software companies in light of AI, which reasonably pressures multiples. But the market deciding that Topicus and Constellation are worth 50% less in just six months marks a significant shift.

I believe this is driven by several factors:

1. Concerns that AI will significantly disrupt VMS.

2. Higher M&A competition.

3. Valuation multiples of CSU and TOI were very high, pricing in perfect execution.

4. Mark Leonard abruptly resigned (health-related), and shareholders aren’t happy with the general lack of communication.

But do the risks warrant the current valuation?

I believe not.

Let’s highlight and then break down the main arguments:

• AI could weaken software moats by making development easier for competitors and customers.

• AI could reduce customer headcount, lowering revenue under per-seat pricing models.

• AI could shift SaaS revenue from per-seat to usage-based, making cash flows more volatile.

• More money is flowing into software M&A with Constellation copycats and software PE firms. This could push acquisition multiples higher, reducing the ROI of new investments.

• Since Constellation and Topicus are already relatively large, the number of investable VMS companies could be smaller. Going forward, more acquisitions and larger companies are needed to maintain high growth rates. With current risks and a limited growth runway, Constellation and Topicus could deserve lower multiples.

• Mark Leonard’s resignation could negatively impact Constellation or be a sign of his lack of confidence in future performance.

I’ll address these concerns in three sections: AI and moats, M&A competition, and key-man risk—before revisiting valuation.

Context

For very brief context, Constellation was founded in 1995 by Mark Leonard to systematically acquire vertical market software businesses (VMS). VMS businesses focus on specific industries and use cases, such as scheduling software for public transportation or guest relationship management for hotels.

VMS are attractive acquisition targets because they often have small addressable markets, which attracts little competition. The customer base is sticky because all their data and workflows are stored on the software, and it works well enough for them. It works great for Constellation because VMS produces large, consistent cash flows, which can be used to buy more VMS businesses. Owners like selling to Constellation because they know it won’t be scrapped or loaded with debt and potentially bankrupted.

In 2013, Constellation acquired Total Specific Solutions (TSS), a VMS acquirer based in the Netherlands with €174 million of revenue. It became an operating group within Constellation and continued to be run by the Founder/CEO, Robin van Poelje. In 2021, Constellation spun out TSS to merge with Topicus, another Netherlands-based software conglomerate. Topicus is now a public company still partially owned by Constellation and previous Topicus shareholders, with nearly €1.5 billion in revenue.

In the 2018 shareholder letter, Mark Leonard provided a realistic but unsettling prediction:

Has VMS reached the period of decline, or does it still have room to run?

AI Weakening VMS Moats

The main concern investors have about horizontal and vertical SaaS companies is that AI will significantly disrupt SaaS as we know it. It could displace the need for many SaaS solutions altogether, change pricing, and increase competition. AI agents are becoming better at administrative tasks and can code software to help. If agents start managing outreach, scheduling, and email drafting, how much use is a legacy CRM?

While this is theoretically exciting, it lacks nuance.

(1) VMS moats and AI capabilities are more complementary than competitive.

(2) AI agents are efficiency drivers, not fully autonomous decision-making employees. Business owners and managers still value insight, reliability, and access to critical information.

(3) Trusting an AI agent will be slow for most business owners.

(4) Security risks are significant if you haphazardly implement AI tools and AI-created software.

(5) VMS TAMs will likely increase due to AI.

To frame this discussion, it’s helpful to consider what disrupting an industry actually means.

Disruption Theory

Disruption theory posits that an industry can be disrupted when the incumbents target higher value customers, ignoring the lower end of the market. Disruptors then enter through the lower end with a different business model and a “good enough” product, eventually producing a technological advantage and enough capital to start moving upstream. Good examples are laptops disrupting desktops and mobile phones disrupting laptops.

This theory doesn’t apply to VMS because many already target the price-sensitive SMB that wants “good enough.” Instead, in this context, AI mainly leads to “sustaining innovations” by improving existing products. Sustaining innovations are generally a positive for incumbents, who are better positioned to develop and distribute these.

That’s not to say AI can’t replace SaaS, but this would more likely occur if the current platforms were systematically incapable of adopting AI. For instance, Uber overtook taxis because individual taxi companies couldn’t develop an app like Uber. This isn’t the case with software. If anything, AI is moving in the direction of being a commoditized application layer, where it’s easy for all companies to implement AI into their existing products.

With that framework in mind, let’s look at how this plays out in practice.

AI in VMS

AI agents help with a great many tasks, but they are still unreliable in how they execute. I don’t think most small businesses would currently trust AI to email suppliers, blindly manage inventory, or cover all administrative tasks. But a day may come when that will be true, and this will still benefit the VMS providers. It is becoming monumentally easier for software companies to create custom agents and implement them into their existing platforms. So, if existing VMS businesses do this, it should improve customer satisfaction and actually increase their stickiness and share of wallet.

The issue with the SaaS displacement argument is that most VMS isn’t used for cutting-edge technology and a sleek UI. Customers use a VMS product because it is well-suited for the industry’s processes, and the employees know how to use it. Customers stay because the solution is good enough, holds all their data, and changing workflows isn’t worth the minimal savings or improvements.

Investors who are overly concerned about VMS customers switching to vibe-coded alternatives underestimate how enterprises actually evaluate risk, reliability, and switching costs. The average SMB is wholly different from a unicorn tech company with a visionary founder. Most aren’t full of software engineers and focused on maximally improving efficiency and technical capabilities. They just want a product that helps file documents, drive sales, or log inventory.

Constellation/Topicus also earn significant revenue from “professional services”, meaning companies pay for software to be customized to their liking. If they paid for a tailored product and like it, I believe it would take extraordinary effort to get them to switch.

These factors make it sticky for SMBs and large enterprises. SMBs don’t have the desire, technical capability, or time to uproot their current software and processes, and large enterprises already have a solution that works for them. If these VMS platforms stayed completely stagnant and didn’t implement agentic AI, then sure, the customers might switch, but that’s not the case.

Why AI is good for incumbents

Back in September 2025, Leonard spoke on a CSU conference call about AI, mentioning 27% of business units (BUs) were developing an AI product for customers. Operating group leaders discussed their expectations for this to increase. Adoption rates were even higher for internal use, estimated at 50%+ for sales, marketing, R&D, and customer service.

At the same time, only 3% of BUs reported replacing employees with AI tools; they instead used the extra time to work on more projects. If this is the case with the BUs, it is likely the same for their customers.

Constellation and its BUs survived and grew during the previous technological shifts of cloud, mobile, and fintech; why wouldn’t they profit from this one as well?

So far, AI features are typically priced as add-ons—either per seat or usage-based. Some worry AI could pressure margins by reducing headcount or increasing revenue variability. I see pricing as a non-issue: if AI drives productivity gains, the software providers will capture a fair share of that value. Customers will either replace labor with AI spend or use the efficiency to grow. Either way, the VMS benefits.

It is difficult to see a scenario where the software enabling productivity improvements captures less value over time. I believe the main concern would be the shift towards a higher percentage of variable costs due to inference costs. But this should decrease over time, similar to how CPU-based compute did.

Whether it’s usage-based or per-seat, better productivity should be a positive for VMS.

Being a system of record means that it is where the company’s data is stored. In the age of AI, this provides an advantage for incumbents, able to develop AI-powered technologies to leverage company-specific data and use cases. Just as cloud, mobile, and fintech drove greater TAMs for VMS, it is reasonable that AI will do the same.

AI should expand the TAM for these businesses by several multiples as it augments sales, marketing, receptionists, and finance functions. Through these functions, AI-powered VMS could help customers decrease their CAC, leading to better margins or reaching more customers.

A counterargument could be made that larger TAMs will increase competition for some vertical markets. But this should benefit both existing and new companies. Many vertical markets will still be too small to attract much competition anyway. Increasing a TAM by 5x from $5 million to $25 million isn’t enough to entice high competition, but it is fantastic for existing providers.

Another counterargument would be that Constellation doesn’t reinvest heavily into R&D and improving its products. Newer entrants might focus more on reinvestment and making the product as strong and well-designed as possible. And this could be much faster to do with AI. Larger software companies with more resources could also better customize software for different verticals.

But this is unlikely to happen for many verticals because of the small addressable markets. These aren’t areas where billions of VC dollars and the most talented founders are going to go. Second, CSU isn’t averse to reinvestment; they just are if it’s low ROI. I think management understands the importance and value of developing new AI-powered applications and won’t sacrifice short-term cash flows for long-term durability. As mentioned earlier, AI-driven efficiencies have actually enabled Constellation BUs to produce more with the same amount of R&D spending, so it may not have a significant effect on cash flow or returns anyway.

To summarize, after researching and heavily considering the AI risks, I believe they are materially overstated and, in several cases, directionally incorrect.

Another aspect bears tend to overlook is the reality of AI coding versus software engineering.

Security, Scaling, and Sustainability

AI coding tools are incredibly useful, but it’s not like you can just prompt an LLM and out pops a VMS competitor. That notion completely contradicts the job market narrative that entry-level engineers are struggling to find jobs, but experienced ones are in high demand. As quoted above, software engineering is not dead, just coding in the current form. This isn’t the first time programming has changed; the format and ease of coding have continuously shifted from the early days of computers. This is just the next step (see here).

Could I create a vertical-specific CRM on Lovable? Probably. Would I feel comfortable that it is robust, can scale, and is actually secure? Not at all. It may sound nice at first, but it will cause chaos when it breaks for the first time, and the designers realize all the edge cases they missed.

The issue with AI-developed programs, especially for non-experienced engineers, is that AI is not perfect at creating easily maintainable code, isn’t good at maintaining that code itself, and isn’t good at debugging. Proper AI coding necessitates a hands-on approach and in-depth knowledge to leverage these tools well enough for an enterprise-quality solution. So, why would a business trust a random, new, vibe-coded product over something that has proven to be reliable?

The same argument goes for the idea of internally developed software. Depending on the industry, the customers can be small and/or lacking technical talent. Even if AI-supported coding can help create software, it still takes substantial time and knowledge to turn that into secure, scalable software.

There are also many instances where it makes sense to have a third party as the software provider rather than the company itself, providing a unified platform and handling sensitive information.

To put my arguments into context, let’s consider a few examples from Topicus BUs.

Topicus Examples

First, a TSS company, Alfa Pro. It provides ERP software for the horticulture industry. A subscription can include a POS, webshop, inventory management, logistics, and bundled hardware. It’s used by companies across the horticulture value chain and is better suited for handling and managing their specific type of inventory than a general ERP or POS. Many of the customers are individual horticulture retailers with minimal to zero software engineering capabilities, meaning they rely on Alfa Pro to help set up the software and continuously provide updates and support. With all of their inventory and business processes on Alfa Pro and the employees trained on that system, why would they move to some new upstart?

Alfa Pro has existed for over 30 years and has ~160 clients. I don’t think taking their customers is a very worthwhile pursuit of someone’s time. This solution is also based on an understanding of the processes and value chain in the horticulture industry. Selling software to these companies is more likely to be successful if the VMS clearly knows the industry and is trusted by others.

To have a good competitor emerge, they would need to be knowledgeable in computer science and the horticulture industry, as well as have the interest, drive, and ability to compete with Alfa Pro. Instead, this is the kind of niche trust-based market where AI will help Alfa Pro drive business process improvements for its customers with AI tools.

Consider another TSS company, CreAim, simplifying Standard Business Reporting (SBR) for accounting/payroll/tax firms in the Netherlands. SBR is required annually, varies by industry, and can change with shifting regulatory requirements. CreAim also provides a system for legally valid digital signatures and a secure portal to communicate with clients. In a highly regulated industry such as accounting and tax, it makes obvious sense for these functions to be offloaded to a third-party software provider.

CreAim keeps up with all the filing requirements, is trusted to act as a third-party system of record for filings and client communications, and connects to several other accounting and tax software.

Large companies like PWC Netherlands even see value in the CreAim portal, paying CreAim extra for customized solutions. They are also ramping up spending as new SBR requirements are introduced:

“After the requirement becomes effective (in 2025), 80% of PwC’s services will be provided through the portal. PwC considers this collaboration a co-production, unique because of the short lines of communication with both the CreAim contact person and the developers themselves.”

This is not something tax professionals would outsource to AI, nor quickly switch to a new competitor.

AI making software development easier doesn’t eliminate the value of VMS; it increases the ability to adapt to improve the offerings and expand the scope. It would be tough to argue that upstarts will use AI to produce software good enough and cheap enough for customers to replace existing solutions. If AI is that useful, then shouldn’t the existing solutions be able to advance, too? Existing VMS companies are better positioned to communicate with customers and figure out what new capabilities to develop. They can also detect early what products customers may switch to and develop similar offerings.

The Downside Is Also Limited

Investors forget that Constellation uses very high discount rates in its models, which minimizes the terminal value. On a hypothetical deal, if you assume 3% organic growth and a 25% required rate of return, all the cash flows after year ten are less than 15% of the discounted value. As a result, even if you paid for the terminal value and the company disappeared after year 10, the IRR is still 20%.

If AI is unlikely to destroy VMS economics, the next major concern is the acquisition landscape.

VMS M&A Competition

AI Impact

If the argument is that AI makes it easier for VMS competitors to pop up, then shouldn’t it mean there are more companies to acquire?

The concern of competition, though valid to an extent, seems not to acknowledge how Constellation and its BUs function. Constellation is heavily decentralized, meaning competition between BUs occurs, which leads all BUs to push harder on development, sales, and retention. To an extent, it doesn’t matter which wins; Constellation still benefits. Why should this be any different with the emergence of AI? VMS businesses still target the same customer bases and have similar stickiness. If there are more VMS companies out there, it just supplies more for Constellation to acquire, and then they compete internally.

Even though Constellation is heavily decentralized, it also benefits from shared knowledge. Managers and operating group heads share learnings and struggles with their peers. How in the world could a random VMS startup have a knowledge advantage over this?

PE and Copycats

PE can be an issue, but many don’t compete in the same areas as Constellation. Many of the companies Constellation buys are too small or too ugly for most PE firms and don’t necessarily serve as platforms for continued growth. On the copycats’ side, some do it well, but many either pursue different strategies or fail at Constellation’s approach to investing. It takes serious accountability and discipline to have deal teams underwrite to a 20-30% IRR—especially in a competitive environment—without manipulating assumptions. Internal pressures from cash deployment requirements and bonus hurdles make this even harder.

Additionally, if Constellation’s space is supposedly so competitive, why would more competitors enter? Wouldn’t they carve out a different expertise, such as targeting fast growers, turnarounds, or platforms to build off?

Lastly, if you believe Constellation consists of skilled teams of capital allocators, then they truly could become several times larger. Look at Berkshire Hathaway; it has become the 8th largest company (in terms of profits) just by being a skilled capital allocator. It would likely be harder for Constellation to shift strategies due to the decentralized model, but finding reliable businesses and modeling IRRs is the core of what Constellation does.

As pointed out in my Topicus Update, the growth runway for Topicus is less of a concern. Each deal moves the needle more; they’re focused primarily on Europe, and Topicus is 1/7th the size of Constellation in terms of revenue.

Mark Leonard’s Resignation

The main focus of this post was AI concerns, but I will quickly address this.

Was Leonard instrumental to the creation and continuous development of Constellation? Yes.

Does this mean they will fail without him? Not at all.

One of the virtues of extreme decentralization is the lack of key-man risk. No individual manager will make or break the strategy; what matters is the collective sum of their skills and efforts. Leonard had little involvement with the majority of Constellation’s transactions; he focused on long-term strategy and large, specialized deals. If you believe he is smart and visionary—as he has demonstrated to be—don’t you think he would’ve prepared the other managers and board members to do his role? Few public company CEOs are as intensely shareholder-focused as Leonard. I think he has well prepared the organization to exist without his daily involvement, and the resignation is a manageable concern.

Topicus Valuation

While the above applies to both companies, I will focus on Topicus here.

As I showed in the update, here is a sensitivity table based on the current share price and various FCF growth rates and terminal multiples. Refer to the update if you want to understand more about the valuation methodology.

As can be seen below, even at a 16x terminal multiple and 8% 10-year FCF growth, the outcome would still be a 10%+ IRR. If Topicus compounds FCF anywhere near historical levels, the current price implies an unusually attractive return.

As a reminder, Topicus’ growth is basically the reinvestment rate x ROIC + organic growth. So, if Topicus even achieves a 10% ROIC from new acquisitions and 2% organic growth, you could easily reach a mid-teens return.

That is all for this post. Please let me know if you have any input, feedback, or questions.

Disclaimer: This is not investment advice. Any statements made are the opinion of the writer and produced for informational and entertainment purposes only. The writer may buy, hold, or sell positions in any of the companies discussed.

Read the full disclaimer here.

Thank you for reading my work; every subscription, like, and comment helps me continue to produce these reports.

Join Investor's Odyssey

Share

If you are interested in another SaaS stock that I think is undervalued, view my Duolingo post.

Read my deep-dive on Amazon:

• Amazon Part 1, Amazon Part 2

Or some other popular posts:

• Dino Polska Update

• Finnish Serial Acquirer at 9x EBITA: Part 1, Part 2

Understand my approach to investing: Types of Companies I Invest In

Consider joining to support my work.

Join Investor's Odyssey

Hi investors,

This post covers a new position I’ve taken following a ~71% drawdown from all-time highs. It focuses on the challenges currently facing the business and why I believe it still has room to grow. The company has a relatively simple business model, and with the depressed stock price, I believe a focused overview is more helpful than a multi-part deep dive.

Contents:

• Investor Concerns (AI, Growth, CFO)

• Why I Like the Company

• Growth Runway

• Valuation

A price drop alone doesn’t indicate value, but it can be a good place to look. In balancing my portfolio between high-growth and value, this company stands out as an attractive addition: it’s easy to understand, mission-driven, founder-led, and run by management that knows how to leverage data to grow users and revenue.

The company is Duolingo, the largest language-learning app in the world. Duolingo provides language-learning through a mobile app, with a robust free tier and two paid tiers: Super and Max. Duolingo is known for gamification, short lessons, and social media marketing with its green owl Duo.

Duolingo’s main focus has been language learning, but they have branched out into math, music, and most recently, chess. Monthly active users (MAUs) reached 135 million in 3Q25, daily active users (DAUs) are now ~53 million (4Q25), and paid subscribers are at least 11.5 million.

Since 2019, paid subscribers have grown ~12x, DAUs ~9x, and revenue ~13x. DAUs now make up 37% of MAUs, and 20% of DAUs are now paid subscribers. Duolingo has clearly learned how to drive revenue through subscriptions while maintaining high user growth.

But the stock has tumbled in the past year from a peak of $530 to ~$150. Duolingo now trades at one of the lowest sales and gross profit multiples since going public.

So, what happened?

Investor sentiment has been battered from several sides in the past year:

1. General software selloff due to AI disruption concerns.

   1. Exacerbated by the nature of language learning and Duo’s platform.

2. Lower near-term growth expectations, signaled by low 20% bookings growth.

3. CFO transition announced on January 12, 2026.

Let’s break each of these down.

Concern 1: AI

The claim has three components: (1) AI can teach languages as well or better than Duolingo. (2) AI-assisted coding makes it easier to develop competitive apps. (3) AI has decreased Duolingo’s content quality.

Investors have further fixated on complaints over efficacy and recent content quality.

These criticisms completely misunderstand Duolingo’s value proposition and even contradict each other.

It is known that Duolingo is not the most efficient way to become fluent in a language. But the entire point is misconstrued. Firstly, no single method will get you fluent in a language; you need to use several. Second, Duolingo’s value proposition isn’t efficient fluency; it’s engagement that feels better than TikTok or Instagram.

The majority of Duolingo’s users likely would not go out of their way to chat with an LLM to learn a language. It’s not remotely the same experience; learning with an LLM takes more time to set up and is less engaging. Translation technology also doesn’t present a threat since it’s a different value proposition, and travel isn’t a main driver of Duolingo use.

“We have a lot of learners that have different motivations, but the two big buckets are not travel. One of them is a hobby, and that’s the biggest bucket in the US.”

“The other huge group is people learning English. For them, it’s not about travel. They just need to learn English either for educational opportunities or for job opportunities. Those two big buckets account for 90–95 percent of our users.”

Luis von Ahn on the Decoder podcast

AI-Developed Content Quality

Another side of the AI criticism is the quality of content Duolingo has been releasing. Duolingo has been very public about its use of AI to speed up education content creation—creating lesson content twelve times faster. Occasional posts have surfaced of lessons with incorrect grammar or answer keys.

While the point has some merit, it completely contradicts the “AI will overtake Duolingo’s role” narrative.

If Duolingo’s AI-created lessons occasionally contain errors despite human review, then raw, unguided LLM output is unlikely to be more reliable. The actual quantity of inaccuracies on Duolingo is also minimal and not impactful. Out of tens of billions of lessons per year, a few incorrect phrases won’t ruin language acquisition.

Also, what Duolingo holds over any AI application isn’t purely the content; it’s the interface it comes in. Duolingo has spent fifteen years, acquired two animation studios, and employs 800 people to improve animation, gamification, and lesson plans. A person using an LLM can’t simply recreate this, nor can a developer using Claude Code.

AI is actually a positive for Duolingo

Some lesson ideas were previously axed because they needed too much content. AI has now enabled these types of lessons to be created. LLMs can rapidly create thousands of unique sentences and potential interactions for any learning level.

It has even led to their higher subscription tier, Duolingo Max, which lets you have live calls with their AI avatars—closing the gap between Duolingo and live conversation services.

On the lower tiers, there are short adventure activities and a mock radio show with pretend callers. All users can now use AI-powered “explain my answer” to better understand why an answer was wrong or right. The point is, the app is still getting better.

Concern 2: Growth Expectations

Subscription bookings growth slowed from 40%+ in previous quarters to ~34% in 3Q25. On the earnings call, management guided towards continued deceleration (~21-23% bookings growth), causing the stock to drop 26% in a day to $193 per share.

Duolingo recently announced reaching the top end of bookings guidance, meaning ~23% y/y growth. DAUs grew ~30% y/y, which is strong, but slower than previous quarters in percentage terms and in absolute user additions (~12.2 million added y/y in 4Q25). The stock has continued to drop, now at ~$155 per share.

Management communicated that slower bookings growth was expected as they have decided to prioritize “platform LTV”, focusing on DAU growth over monetization. DAUs are strongly correlated to conversion to paid subscribers, since the subscription makes daily use much more practical. So, Duo management believes driving more DAUs on the platform now and focusing on monetization later is the correct strategy.

Founder/CEO Luis von Ahn believes the opportunity ahead is so significant that they want to prioritize user growth, just as they did in the early days.

“What Luis is describing is a change of small proportion right now that is going to help us grow users for a long time, get users to do more lessons, learn better, spend more time on the app, and in general, be more engaged Duolingo users. And we think that if we do that and we do that effectively, we'll both grow users for a long time and we'll increase platform LTV for a long time.”

CFO Matthew Skaruppa, Duolingo 3Q25 Earnings Call

The concern with fourth-quarter numbers is that they don’t support the DAU prioritization narrative (at least yet) since both DAUs and bookings growth were lower.

For investors, the questions are: how much room does Duo have to grow MAUs and DAUs, will they win this new age of learning, and will these new DAUs be as likely to subscribe? This problem is related to the first concern of AI, but harder to judge.

“As we said in our third quarter 2025 shareholder letter, we’re prioritizing teaching better and user growth, and we executed on that strategy in the fourth quarter. While I’m delighted with our preliminary fourth quarter metrics, we plan to continue to invest meaningfully in the product, even when it involves near-term tradeoffs. We are doing this because our long-term opportunity is to teach billions of people, and while we’ve made incredible progress, we remain early in our journey.”

Luis von Ahn, Duolingo Founder and CEO

This could be a bumpy road for the next few quarters, but if the reinvestments pay off, long-term shareholders will be rewarded.

Concern 3: CFO Transition

On January 12, 2026, the CFO since before the IPO, Matt Skaruppa, announced he is stepping down. Unsurprisingly, the stock tumbled further. The transition was unexpected based on recent earnings calls; both the CFO and CEO were discussing their enthusiasm for the future of the company. Skaruppa has also been selling a significant portion of his shares, and the announcement gave no reason for his departure. But CFOs leave companies all the time, and extrapolating this as a signal of diminished confidence is likely excessive.

I don’t think the change of CFO provides any material execution risk, and the new CFO was a long-time board member, Gillian Munson, who has previous roles as a CFO at Vimeo and a VC at Union Square Ventures. From everything I’ve heard, Skaruppa did play a significant role in Duolingo and investor relations, but I think Duolingo will be fine.

The founding CEO and CTO, Luis von Ahn and Severin Hacker, are still with the company and hold a combined 13.5% of shares outstanding and a 76% voting stake in the company. It’s safe to say their hearts are still in it, and they have the voting power to make non-consensus decisions.

My Opinion

The information is mixed. AI-related concerns appear overblown, and the CFO transition is unlikely to have a material impact, but concerns over growth slowing are real. If DAU growth reaccelerates, investor confidence should follow; if not, near-term volatility is justified.

What investors should keep in mind is why founder-led, controlled companies can be powerful long-term compounders. They have both the cultural influence and voting control to prioritize durable value creation over short-term optimization. I don’t see a competitor as large or as singularly focused on delivering an engaging, improving learning experience as Duolingo. With that context, the question becomes whether Duolingo’s advantages and long-term growth potential justify the current valuation.

Why I Like Duolingo

Data and Scale

Data advantages and scale advantages are some of my favorite sources of a competitive advantage. It often enables better product and retention, higher margins, and better monetization. Since competitive advantages can drive greater scale, it then leads to more data, helping drive even more scale.

Duolingo is a data-driven organization by nature. They understand how users interact with the platform, how often they return, and the best way to keep them. Duolingo teams run thousands of A/B tests a year, trying out new features, designs, and ads to convert users to the premium subscription.

“We had just built a company around—our entire product organization that’s not just product managers, engineers, designers—shipping A/B tests as fast as possible.”

Luis von Ahn on the Acquired Podcast

Duolingo’s teams have become very good at managing towards a metric and improving it. This is how the whole organization is structured. Teams each own a metric rather than a function. A long history of tests and outcome data means management can clearly and realistically prioritize DAUs and platform LTV over revenue if desired.

Despite being data and A/B test-focused, management also knows where to draw the line. A small committee (including the CEO) reviews all the proposed changes and says no if they believe it’s bad for the long-term health of the app.

Scale is incredibly important as well. Duolingo can design more courses, better animation, and have stronger insight into user interaction because of its scale. It would be difficult for a competitor to replicate Duolingo’s UI and gamification—especially at a similar cost per user—due to how many engineering and design hours Duolingo has put into the platform. Even with Duolingo’s higher user base, they were barely cash flow positive in 2021 and weren’t net income positive until 2023. This is accelerated in the age of AI, where there are incremental API costs. Competitors could not offer Duolingo’s same AI features at a similar cost and earn a strong margin on them.

Another reason I like Duolingo is that they’re mission-driven:

Mission-Driven

Luis von Ahn was a professor, and Severin Hacker a PhD student, who became united on the idea of making education more accessible. Von Ahn grew up in a country where learning English was very economically valuable, but learning and testing it were costly and unattainable for most.

This bled into the company’s culture, where they went the first four years without monetizing the platform. So much so, that many employees were concerned about introducing monetization as they felt it conflicted with the mission. Over time, they realized that charging the people who could pay let them make the platform better for everyone and continue to serve those who couldn’t pay.

Those four years of no monetization taught them to prioritize learning metrics and showing users value before attempting conversion. Management recognizes that same user focus will help them win the AI era. The cofounders still run the company, are board members, and controlling shareholders. Of anyone, they understand Duolingo’s mission and how to build value over the long-term. The rest of the management team is similarly aligned on this mission.

“Generally what we do is we go for users, as in if something decreases the number of active users that we have, unless it’s going to make us $500 million, we’re really not going to launch it.”

Luis von Ahn, Acquired Podcast

The third reason I like Duolingo is cash flow:

Cash Flow

Similar to SaaS companies in general, cash flows are much stronger than GAAP income due to deferred revenue. Excluding SBC, deferred revenue from annual plans makes up half of their operating cash flows, $131 million. What this means is that Duolingo already earns a 23% FCF margin despite an 11% operating margin. This has driven Duolingo’s ability to reinvest and hold a strong cash balance, with >$1 billion in cash and zero debt.

The question remains, will these cash flows grow?

Growth Runway

Duolingo’s continued popularity is prevalent. Despite claims that competing apps are better or gaining popularity, data says otherwise. Duolingo had 5.6x more downloads than the next most downloaded language app, and 20x more than Babbel and Busuu.

Duolingo's long-standing dominance and early entry mean it's creating the market rather than just taking share. People may have used online programs, YouTube, textbooks, traditional schooling, or weren’t learning languages at all. This means you can’t look at the language learning app market and assume they take a greater share. Instead, you have to consider total spend on language learning and how that market may grow over time. We can start at the top and work our way down.

Research firms estimate the global online language learning market at ~$20-30 billion in 2024, and expected to grow 15-20% through at least 2030.1 Meaning, Duolingo at most holds a 5% market share in online language learning revenue. Other sources suggest closer to 1% share. If they keep or even slightly lose share, they should grow strong double digits. As mentioned, benefits from operating leverage will improve their offering versus competitors. Use of AI functionalities accelerates this, as larger companies can better optimize cloud and AI spend.

Now let’s consider Duolingo-level numbers. From 2021-2025, MAUs have grown 220% to 135 million. At the same time, DAUs grew 400% to 50 million. Faster DAU growth was driven by better conversion of MAUs to DAUs—up from 24% to 37%—as the app became more engaging. Paid subscribers grew faster than MAUs, too, at 360%. Since ~8.5% of MAUs are paid subscribers, whereas ~23% of DAUs are paid subscribers, Duolingo has at least two strings to pull: higher MAU growth and higher DAU conversion.

Ten-year growth could be mapped out like this:

• Global language learners are 1.5 billion, estimate this number grows 5% per year

• MAU growth slows to 12% average

• DAU/MAU is 37%, estimate that retention efforts help this reach 50%

• Paid Subscribers to DAU could decrease since newer, incremental users in lower-income markets may convert at lower rates, estimate 23% => 18%

• Modest pricing growth of 3% per year

Outcome:

• Paid subscribers compound at 12.6% and bookings compound at 16%

• Subscriber penetration of global learners only needs to increase from 0.8% to 1.5%

Feel free to plug in whatever estimates you want, but this should help illustrate how levers can be pulled to grow DAUs and bookings:

This doesn’t even consider future operating leverage potential or greater pricing power.

With that being said, where will growth come from?

Monetization and user penetration still have room to grow

Management has communicated that there are no clear signs of market saturation. They have seen little correlation between the user saturation in a country and growth rates. For instance, Latin America was their most mature region, but was growing DAUs at 80% y/y as of 4Q24.

“If you look at our DAU growth, some of our most penetrated markets are actually the ones that are growing fastest. We don't see that like, oh, we've reached a level of penetration that makes it so that we're going to grow slow everywhere. We don't see that.”

Luis von Ahn, 2Q25 Earnings Call

Higher conversion is a clear area of growth. Subscription revenue is over-indexed in the United States and among iPhone users in general. Even for people who can afford subscriptions, the culture in less wealthy countries makes them less likely to subscribe. This means a large runway remains if Duolingo can understand how to better incentivize subscriptions in these countries.

Subscriptions should also grow through Duolingo Max, which is double the price of Super in the US. As of 3Q25, only 9% of subscribers were on the Max tier. This should continue to grow as it becomes more widely available and cheaper.

For instance, they started testing Max in China during 3Q25 since they finally got approval on AI use from the government. China presents a strong opportunity due to the large population, and it only makes up ~6% of subscribers.

Max should also grow as AI API costs lower. Duolingo Max offers unique AI functionalities, which are still too expensive for countries like India. In 2024, Max in India was priced at $70 USD per year due to AI costs, causing a huge gap between Super and Max. As these costs continue to dramatically decrease, the number of potential Max users should grow.

Efforts in other education areas are important too. While math, music, and chess aren’t separately monetized (yet at least), it strengthens the value proposition of the subscription tier and widens the potential TAM. Content amounts are continuing to grow, such as math, which didn’t have full K-12 content previously.

Valuation

This section has been updated to better reflect the correct valuation by subtracting interest income from the cash balance. The valuation is updated for the price as of 2/3/2025, which is $121 per share, down from $156 at the original time of posting.

All the growth potential and competitive analysis only matter in the context of valuation, and I believe the price is attractive.

TTM 3Q25 free cash flow was $224 million excluding SBC, a 23.2% margin on revenue. The midpoint of FY25 guidance is $1.03 billion in revenue, implying approximately $239 million in free cash flow and a 30x P/FCF multiple.

Once you subtract the $44.3 million of interest income for TTM 3Q25, FCF is an 18.6% margin, implying an estimated $192 million in FCF for FY25 and an EV/FCF of 24x.

It is important to note that long-term revenue and free cash flow growth won’t be linear, as cash taxes will increase, and slowing growth could shrink the gap between net income and cash flow. However, operating leverage should help drive margins higher over time.

For valuation, assume you want a 10% return and believe growth will slow to 3% after year ten:

Currently, the diluted market value is $5.78 billion, and Duolingo holds $1.1 billion in net cash. This implies an enterprise value of $4.65 billion, or $97 per diluted share.

To achieve a ten-percent return, Duolingo would only need to grow FCF ~10% for the next ten years.

Instead, if you expect a 16% FCF growth rate, Duolingo is undervalued by an estimated 37%.

Under the original post, the price was $156 per share, which would have resulted in a 14% required growth rate for a 10% return if you adjust for the new FCF estimate.

We’ve already touched on whether double-digit growth can continue, but could growth decelerate faster?

A fatal flaw investing in growth stocks is underestimating how quickly growth can slow. Many growth stocks see unexpected growth deceleration, causing multiples to re-rate and investors to get burned. Some have proven to be falling knives, while others have rebounded. Popular growth stocks Text S.A. and e.l.f. Beauty have seen large drops like this recently.

But growth deceleration isn’t always permanent. Meta and Netflix were overly sold off and skyrocketed when growth returned, and e.l.f. has already started to rebound.

The question is, which will Duolingo be?

Analyzing the respective cases should provide insight:

Recent decelerators:

• Text S.A. (enterprise chatbot) - FY23 growth was 33%, FY24 was 23%, and the latest quarter was -7.4%.

  • Text S.A. was popular due to the AI and chatbot buzz, but their offering wasn’t differentiated or very sticky.

  • There were numerous competitors, SMB budgets tightened, and there wasn’t a clear growth story.

• e.l.f Beauty - Grew anywhere from 26% to 85% every quarter from June 2022 to December 2024. Growth slowed to 3.6% in March 2025 but climbed back up to 14% for September.

  • Deceleration was driven by tariffs raising prices (75% of the product comes from China), poor shipment timing, high debt from the $1 billion Rhode acquisition, and a lack of recent popular products.

  • Beauty is also a very competitive industry where sentiment shifts year-to-year based on popular, trendy releases. E.l.f. was priced at extremely high expectations, but management execution and tariff policy disrupted that.

Reaccelerators

• Netflix - Growth slowed to low single digits in 2022 due to the COVID shutdown ending, heightened competition, and consumer budget tightening.

  • Growth re-accelerated to double-digits because these trends were short-term. Consumer spending rebounded, and Netflix still had a stronger brand and content when the hype of new platforms died down.

  • Netflix still had growth levers to pull on: password sharing and an ad-tier.

• Meta - Negative revenue growth in late 2022-early 2023. DAUs went practically flat or negative for the US and Europe, and ad revenue declined even more as the ad market softened. Meta was also spending heavily on Reality Labs, hurting margins.

  • Growth re-accelerated to strong double-digits as Meta continued to improve recommendation algorithms and ad-targeting, and significantly cut overhead costs.

  • The Facebook and Instagram ecosystem strength and the role of data were underestimated by investors.

While all these situations aren’t perfectly comparable, the general trend of deceleration came from macro headwinds, supply risks, and a lack of sustainable competitive advantages. Re-acceleration came from underestimated brand stickiness, content, data, and growth levers.

Duolingo is more likely to fall into the latter category. Their products aren’t as easily replicable as Text S.A. and not as fad-driven as e.l.f. Beauty. Instead, they have data-driven growth levers to pull and are differentiated in the combination of brand and content + interface. Just like Meta and Netflix, Duolingo’s management will do everything in its power to increase the app’s popularity and profitability.

At today’s price, investors are paying for a materially slower Duolingo than the one management believes they are building.

If you want to learn more about Duolingo’s business, approach to content and design, and how Luis von Ahn manages the company, these podcast episodes with him are great:

Decoder, Acquired

Disclaimer: This is not investment advice. Any statements made are the opinion of the writer and are produced for informational and entertainment purposes only. The writer may buy, hold, or sell positions in any of the companies discussed.

Read full disclaimer here.

Thank you for reading my work; every subscription, like, and comment helps me continue to produce these reports.

Join Investor's Odyssey

Share

Read my latest deep-dive on Amazon:

• Amazon Part 1, Amazon Part 2

Or some other popular posts:

• Dino Polska Update

• Finnish Serial Acquirer at 9x EBITA: Part 1, Part 2

Understand my approach to portfolio construction: Types of Companies I Invest In

Consider joining to support my work.

Join Investor's Odyssey

Part 1 Recap

Part 1 detailed my analysis of the outlook for the AWS & e-commerce businesses. On the AWS side, they are continuing to execute on their AI strategy and Anthropic partnership, and are poised for continued high-teens to low twenty percent growth. For e-commerce, margins are expanding through fulfillment center automation and reorganizing distribution networks. Growth will continue as e-commerce takes a larger share of the retail and grocery market, driven by faster shipping speeds.

With these two covered, advertising represents the third major growth driver and perhaps the most underappreciated opportunity.

Part 2 covers:

• Advertising - data advantage, Prime Video, DSP, growth levers

• Emerging businesses - healthcare, Kuiper, and Zoox

• Valuation - assumptions and expected returns

If you’re only interested in some sections, the sidebar provides easy navigation to my subheadings.

Amazon Ads

“Similar to the way I think about Amazon retail, for all practical purposes, I believe AWS is market-size unconstrained.”

- Jeff Bezos, 2014 Amazon Shareholders Letter

Amazon, beginning with e-commerce and expanding into AWS, has produced businesses with market sizes that investors initially underestimated, and they continue to generate great prospects decades later. I believe Amazon Ads fall into this category as well.

Amazon Ads has quickly become the third-largest digital advertising platform behind Google and Meta, yet its long-term potential remains underappreciated. Unlike traditional digital ads, Amazon’s offering is tightly integrated with commerce, allowing advertisers to directly link ad spend to purchases.

Amazon Ads has rapidly scaled from $12.6 billion in 2019 to $64.6 billion for LTM 3Q25, a 32.8% CAGR. Amazon has unlocked an incredible revenue and profit stream with a differentiated ad offering compared to Google and Facebook.

Advertising revenue comes from a few sources:

• Sponsored Ads - Search & Shopping

• Video Advertising - Prime Video and Twitch

• DSP (Demand-Side Platform) - a platform to buy ad-inventory

All represent strong opportunities and serve as catalysts for an inestimably large market size.

Sponsored Ads is their longest-standing ad business, generating revenue from sellers paying for sponsored placement on the search page. It likely generates $40-$50 billion worth of the $64 billion 3Q25 LTM ad revenue. As the number of customers, products, and purchases per year continues to grow, search ads likely will as well.

Video Advertising is growing in importance, with Prime Video gaining users and expanding its ad tier. Launching ads in 2024, there is a long road ahead. Conversion has been increasing as they have introduced interactive advertisements, which make purchasing easier. Ad supply has also grown as Prime Video started to increase its ad-load from 2-3 minutes per hour to 4-6 minutes in June 2025, which is still much lower than cable’s average of 14-18 minutes.

Demand Side Platform (DSP) is lesser known and has high potential going forward. It utilizes Amazon’s detailed customer data (i.e., purchases, searches, and content engagement) to sell ads not only on Amazon’s properties, but also third party platforms. In 4Q25, they launched a close partnership with Netflix, allowing advertisers to purchase ads for Netflix’s ad tier through Amazon DSP. The partnership is beneficial for Netflix in two ways: (1) advertisers seeking to place ads on Netflix can now leverage Amazon’s first-party e-commerce data, and (2) use Amazon’s purchase attribution tracking, getting better insight into ROI.

How DSP works

Amazon can track customers across platforms by essentially assigning a customer ID to devices, emails, phone numbers, and IP addresses. Partnerships allow them to link the customers in a clean room environment across platforms and sell ads with more detailed specifications.

For instance, Keurig could say “target customers who have put a coffee machine in their cart but not bought in the last two weeks,” then choose to show Keurig machine ads on platforms like Hulu, Spotify, and Netflix. If that customer returns to Amazon and buys a Keurig, the advertiser has clear data of an advertisement leading to a purchase. By understanding a customer’s entire shopping history and reaction to ads, Amazon can further improve the return on ad spend and conversion rate.

Amazon maintains structural advantages over Google and Meta

Google’s intent data is more vague; they understand the interests of consumers, and maybe if they’re actively shopping for something, but once the customer goes onto a brand’s page, Google has much less insight and tracking ability. It is not as easy for an advertiser to connect advertising across platforms to a customer making a purchase. It is the same for Meta to a larger degree, meaning most brands use social media for discovery over clear purchase conversion.

Surveys consistently rank Amazon as the place where over 50% of people begin their shopping journey, outranking search engines entirely.1

Another strength of Amazon DSP is that brands can manage their entire digital advertising strategy. For example, an advertiser can set a frequency cap that is counted across platforms, so a customer doesn’t see an ad ten times on Twitch, then another ten on Netflix. Amazon’s platform allows for optimized ad-spend, less friction, and greater data for potential advertisers, already signing similar deals with Disney, NBC Universal (Peacock), and Roku as well. Due to their granular intent data, there is a benefit to running a significant portion of your advertising on Amazon DSP, even outside Amazon’s properties.

DSP also connects to dozens of external ad exchanges, meaning an advertiser can direct Amazon DSP to purchase ad spots across the web based on specific customer metrics and ad-campaign goals.

To be clear, Google and Meta have some of these targeting capabilities as well and are strong in top-of-funnel advertising—introducing people to a brand. But Amazon is encroaching on this too with Prime Video, Twitch, and the numerous DSP partnerships.

Prime Video Dominance

Potential within Amazon’s own streaming ecosystem is great, expanding their ad offering and releasing numerous successful series, including Fallout, The Boys, and Reacher.

Originally, I expected Prime Video would be popular in the US but minuscule internationally, and that was far from the truth. They’re competing with Netflix in several large markets; 41% of their commissioned content in 2024 targeted non-US markets like India.

Another growing source is live sports, one of the few reasons people still keep cable television. Prime now displays 66 NBA and WNBA games a year and all of Thursday Night Football, as well as the UK Premier League. With the ad-tier included in the Prime Membership and ad-free only a few dollars more, the value of Prime is still growing.

“The NFL drives such big reach, but how do you continue telling that story and interact with those fans throughout the day? And it’s proving better performance when we had multiple sports just in that one day with advertisers.”

- Danielle Carney, Amazon Ads head of live sports and video2

Sports on Amazon have seen greater engagement and better ad performance versus the historical average. Amazon is continuing to build its live sports strategy, linking different sports throughout the day and selling advertisers day-long packages.

Advertising generates strong margins

Many analysts estimate a 35-40% operating margin on the ad business; if you apply that, then ads contribute $22-26 billion of Amazon’s operating income, which is nearly a third of Amazon’s total operating income.

On top of AWS becoming an increasing share of revenue and e-commerce margins growing, this is yet another factor causing the long-term earnings growth for Amazon to likely outpace revenue growth.

How ad revenue has room to grow:

If Amazon’s ad revenues are to grow, those dollars have to come from somewhere.

“When you layer on top of it the combination of the audience curation and development we can do along with the advantage measurement, it just all leads to a return on advertising spend that’s very unusual. I think there are multiple places where we can expect to continue to grow. One is in our stores business. I still think if you look at the worldwide market segment share of retail, still 80% - 85% of it lives in physical stores.”

- Andy Jassy, Amazon 3Q25 Earnings Call

In 2025, the digital ad market is estimated at $767 billion, and the total advertising market at $1.1 trillion, implying digital takes a 69% share of the total advertising market3, other sources estimate closer to 60%.

For LTM 3Q25, Google takes a substantial chunk with $216 billion for search and $39 billion for YouTube. Meta takes another $184 billion with Facebook and Instagram. Amazon takes third place with $64 billion in ad revenue, the majority coming from Amazon.com. In total, these comprise nearly half of the total ad-spend and 66% of digital ad-spend worldwide.

Analysts expect the share to continue towards digital, reaching 80-90% by 2040 as digital format permeates every area of life (McKinsey). Digital permeates out of the home as well, with electronic display screens for advertisements likely to continue spreading.

For Amazon ads to continue growing at double digits, three things will have to happen: digital will continue to take a larger share of ad spend, ad spend will continue growing at mid-single digits, and Amazon will take a larger share of the ad market.

Here is how growth will continue in each area:

Prime Video is in the fast-growing advertising video on-demand (AVOD) and overall connected TV (CTV) market, which is estimated at $45-55 billion, growing at double-digit rates. For Amazon, it will grow as the number of subscribers and viewing time grows, the ad load increases, and conversion rates for their ads increase.

When Prime Video introduced ads, it converted all regular Prime users to ad-supported users with the option to upgrade for no ads. As a result, Prime now captures the largest quantity of ad-supported users, 315 million monthly active viewers globally and 130 million in the US, as of November 2025 (Amazon). In contrast, Netflix reported global ad-supported MAVs of 190 million that same month. AVOD is another place where Amazon’s first-party high-intent data and closed-loop attribution system give it an advantage. Amazon can provide the lowest friction experience for interactive shoppable ads as the only US retailer with a streaming platform.

AVOD will inevitably continue taking share from linear television. Despite Linear TV accounting for an estimated 85-90% of ad impressions, CTV takes 35-40% of TV advertising dollars, a virtue of better data driving higher ad prices.4 As the number of AVOD subscribers grows, it is bound to continue taking share from linear TV and growing the pie. AVOD is better than linear TV for advertisers in practically every way, knowing who they are advertising to, how they engage, and if it leads to a purchase. Linear TV’s lingering benefit is greater reach for lower cost-per-thousand (CPM) views, but much weaker targeting. It is still early days for advertising on CTV, but the future looks promising.

The risk... is that a greater supply will decrease CPMs for streaming ads, which already occurred in 2024 when Prime Video introduced ads. Some estimates believe average CPMs will remain at this range for the next couple of years, which could slow short-term growth. But, as Prime Video came in at a lower-than-average CPM initially, the effect will likely be greater on streamers like Netflix, which had the highest CPMs. Regardless, connected TV (CTV) is one of the highest growth areas going forward, as the possibilities and ROI improve and continue to transform.

Amazon Search Ads can also garner more dollars as premium brands join the platform and the algorithms continue to improve. By each advertising dollar being more effective, it will drive more brands to the platform and could increase their ad-spend. Retail media is the fastest-growing digital ad medium, growing over 20% in 2024, with many advertisers indicating they will increase retail ad spend the most (Statista). If you believe Amazon’s share of retail sales will continue growing, then its share of the currently $160-190 billion retail media market should grow as well.

Additionally, as grocery moves further towards e-commerce, more of the grocery advertising spend will shift towards Amazon search. Grocery ad spend would also likely increase due to the improved attribution and ROI compared to traditional advertising sources.

Retail Ad Service was announced in January 2025, with the idea to provide retailers with Amazon’s ad technology. In the two-decades Amazon has been selling ads online, they have developed sophisticated machine learning models and front-end ad technology to help advertisers optimize ad campaigns and maximize revenue for Amazon.

Currently in beta, this AWS-based service provides simple API integration and enables advertisers to manage ad campaigns, even if they’re not Amazon sellers. This has the potential to take a large share of ad-tech since it is valuable for retailers and advertisers to have the same interface across retailers. This initiative is just beginning, but it is worth watching.

Amazon DSP, though first established in 2018, is still nascent. Based on third-party estimates, DSP accounted for roughly 21% of Amazon’s net ad revenue in 2022, with off-site ads contributing ~5%.5 For simplicity, you can think of DSP as programmatic targeting and conversion outside of search. Sellers turn to search to place higher on search keywords, but brands use DSP to extend awareness. As ad spend outside of search grows, DSP will take a share, meaning it will likely also grow as a share of Amazon’s total ad revenue.

Amazon DSP has begun to solidify as a leading ad network and exchange with fantastic conversion and ROI. This will likely continue as more dollars are directed at retail media and AVOD, along with proving when advertisements are leading to sales. Since many of these partnerships are just beginning, it’s clear there is room to expand.

DSP serves as a flywheel for advertising dollars. As advertisers use Amazon DSP across third-party platforms, it provides a more diverse dataset on each of their customers. They can understand engagement and conversion rates across platforms and how each customer responds to ads. This is just the beginning for Amazon Ads, and I’m excited to see where it can go.

Emerging Businesses

These are businesses that aren’t reported as clearly separate line items and likely have a long growth path ahead. The few I will briefly discuss are healthcare, Leo, and Zoox.

Healthcare

Healthcare is Amazon’s effort to attack a $1.5 trillion+ TAM across primary care and pharmaceuticals. Amazon Pharmacy was launched in late 2020, providing access to pharmacists and delivery to your door. Pharmacy improves Amazon’s positioning as an everything store and makes the process of getting prescription medication significantly easier. OneMedical is the other side of their healthcare push. Amazon acquired OneMedical for $3.9 billion in 2022, and customer testimonials demonstrate how OneMedical is changing the game.

Traditionally, it can take weeks or even months for the average person to get an appointment with their doctor. Whether you’re feeling sick or there’s something you want looked at, traditional medicine can be a hassle. Those who can afford it pay thousands of dollars per year for private practice doctors, where you have more appointment availability and a direct line to the doctor.

One Medical bridges the gap between the two, with an affordable, fast option. Through One Medical, patients can easily book telehealth and same-day or next-day in-person appointments—reflecting the same time-sensitive customer focus Amazon aspires to. One Medical is just $99 per year, which is much cheaper than private care plans. From reviews online, they are more accessible than traditional doctors’ offices, actually listen to patients’ problems, and even address chronic conditions well.

The subscription can be linked with your Amazon account for convenience, and integration is expected to improve over time.

If there’s a market where people have been wanting a change, it’s healthcare.

Leo

Leo is later to the satellite internet game compared to Starlink, but it still presents a great opportunity. Early reviews indicate Leo provides a strong alternative to Starlink in terms of hardware and internet capabilities.6

Currently, most internet access is through broadband, cell networks (2G-5G), and some geostationary orbital satellites (GEO), but 31% of the global population still lacks internet access according to the World Economic Forum.

Each source has tradeoffs; GEO internet offers the best coverage but lacks portability, resulting in low network speeds and high latency. Fixed broadband and cellular networks generally offer higher speeds, but aren’t accessible for much of the global rural population. The effort to install fixed broadband or cellular can be costly and ineffective in some currently uncovered areas. It is also low ROI in less densely populated regions, and poor countries’ governments can’t or won’t necessarily fund it.

Low Earth orbit (LEO) satellites provide a good alternative, being small portable terminals that directly connect to smartphones with better latency and speeds than GEO satellites. Goldman Sachs estimates the low-earth orbit satellite market could be worth $100 billion by 2035.

Increasing internet access will help all of Amazon’s other businesses. More people online means more e-commerce sales, more people to show digital ads to, and more computing and storage needed.

But Amazon Leo’s success isn’t a sure thing. SpaceX has over 9,500 active satellites with a long-term target of 42,000, compared to Leo with about 150 in the sky and a target of 3,200. Economics are also different since Starlink is a division of SpaceX, whereas Leo contracts launch services from third parties, primarily Blue Origin and SpaceX.

Goldman Sachs also estimates that a majority of the launches will be from China. Since areas lacking internet access include many countries with which China has a history of partnering (e.g., through the Belt and Road Initiative), this could help cement China’s position in the global market.

Zoox

Zoox was acquired by Amazon in 2020 for $1.2 billion, aiming to tackle the $54 billion market size in the US and $166 billion globally. It’s differentiated from other robo taxi options like Tesla and Waymo by fully rethinking the design of a taxi without a driver. There’s no steering wheel, no pedals, and bench seating on both sides. It’s already available in San Francisco and Las Vegas, with plans to scale production and launch in Miami and Austin soon.

It’s no surprise that robotaxis will become a very large market as share shifts from non-autonomous rideshare, prices decrease, and more people use autonomous rideshare overall. The winners in this market are still unknown; we will just have to wait and see how Zoox and its competitors are received and how they execute.

Valuation

Amazon is notoriously hard to value. The diverse businesses, lack of reporting transparency, and embedded R&D and growth costs make it difficult to evaluate the company in its entirety, or even on its vague reported segments (NA, INTL, & AWS). I tested a few methods and came to the belief that Amazon is valued for modest double-digit-growth, but has potential to surprise to the upside. They will have to execute on several of their business areas for investors to earn a strong return, but have several business lines and profit initiatives going for them.

The method I’ll show uses a sensitivity table based on various growth assumptions. Rather than splitting between AWS, NA, and INTL, as is reported by Amazon, I decided to run it on AWS, Advertising, and the rest of the businesses as the third segment (which are mainly e-commerce, in-store grocery, and Prime subscriptions). I’m sure some will disagree with this method, but I believe it best separates the growth/profit levers, and at Amazon’s current size, their subscription revenue will likely track close to their overall e-commerce growth.

Growth Cases

Since many of Amazon’s businesses will likely slow in growth in the coming years, I mapped out several growth cases across three 5-year periods, meaning a 15-year explicit growth projection. Those estimates can be seen below, which were applied to the Advertising and AWS businesses.

These estimates aren’t perfect, but rather just meant to give insight into various levels of starting growth and rates of deceleration.

To put into perspective whether these growth cases are reasonable for the market, I provide hypothetical scenarios of market share, the imputed market size, and CAGRs for the different growth cases.

For both businesses, I believe the lower growth cases would be from a slower-growing TAM and the businesses holding or taking less market share.

Margin Estimates

From what I could determine, many estimate a 40%+ margin for the advertising business. To be conservative (and so it fits reasonably with the NA and INTL profit numbers), I assumed 35% operating margins. I didn’t assume any margin growth, as this was a rough assumption anyway. In reality, it’s probably difficult to disentangle advertising margins from e-commerce, but this is the best method I could determine with the limited information. For AWS, I assumed a modest increase over the 15-years from 35.9% to 37%—though in the end, it had a minuscule effect on expected returns.

From AWS’ LTM margin and the estimated margin on the ad business, that left roughly $14 billion in estimated profit from retail, subscriptions, and other sources for FY25 and a 2.7-2.8% margin. Implied within this are likely all the additional R&D costs for new businesses and the costs associated with e-commerce growth. Leo alone is expected to have $1-2 billion in operating losses.

I assumed margins for this entire segment would grow to ~5.5% by 2040 as unprofitable ventures materialize, e-commerce margins grow, and higher margin businesses take a larger share (e.g., 3P and subscription). I would not be surprised if margins end up higher, as robotics-led cost-saving alone could drive this margin increase, but that can’t be reliably factored into margins. Assumptions for the e-commerce growth vary by revenue segment, but average out to a 8.1% revenue CAGR and 13.1% EBIT CAGR due to the margin expansion mentioned.

Here are the overall results of my calculations:

If you take the case 2 assumptions (13.6% 15-year CAGR) for both AWS and Ads, then the model shows you should expect ~13% return. Of course, stock investing never works out as simply as this, but it shows what kind of expected returns are embedded currently.

Amazon is currently priced in a position where the long-term returns will be similar to the underlying earnings growth. So, depending on what growth rates you expect for Amazon and how confident you are in those estimates, you have a sense of whether Amazon is fairly valued.

And for any critics of the advertising margin assumptions, while they have an impact on returns, it is frankly less than expected:

Yet another example of how growth rates are most important for returns, and Amazon is downturn-resistant due to the diversity of its cash flows.

Takeaways

To summarize, I like Amazon as an investment for several reasons:

• E-commerce has clear reasons for continued growth and margin expansion.

• I believe the downside risk for the e-commerce business is rather low compared to the upside potential.

• AWS has competitive risks, but has been executing well and holds outsized growth opportunities.

• Advertising presents significant untapped potential through improved data analytics, advertising on third-party platforms, and strong first-party content.

• Amazon’s advancements in automation, delivery, grocery, and advertising all create a fantastic flywheel that’s supercharged by ever-increasing data and analytical models.

• Over the very long-term, Amazon has initiatives that could lead Amazon to surprise to the upside through new revenue opportunities and cost-savings.

  • It is becoming less far-fetched that packages could be delivered (whether through self-driving delivery vans or drones) much more efficiently, autonomously, and at a variable lower cost.

  • Fulfillment center automation technology could be sold to other fulfillment companies.

Some of the risks present are:

• Poor execution on AWS’s AI and foreign strategy.

• E-commerce share of retail and grocery grows less than expected, or other retailers like Walmart take more share.

• Automation cost-savings turn out to be less than expected.

• We enter a period of an advertising glut where growing supply shrinks CPMs materially.

From the research I conducted, my conclusion on Amazon’s prospects is that they have shown an ability to systematically disrupt and take a large share of markets as traditional formats shift to technology-based ones. Due to the various businesses Amazon operates in and the clear runway for further growth, it serves as downside protection in a sense. While several businesses have to do great to earn a strong return, to earn a poor return, several businesses would have to underperform.

I continue to like Amazon as a position in my portfolio, but don’t see it as severely undervalued today.

I will leave it up to you to judge what outcome is most likely.

Read full disclaimer here.

Thank you for reading my work; every subscription, like, and comment helps me continue to produce these reports.

Join Investor's Odyssey

Share

See my most recent posts:

• Dino Polska Update

• When Quality Fails

Read the two-part write-up on a Finnish serial acquirer: Part 1, Part 2

Understand my approach to portfolio construction: Types of Companies I Invest In

Consider joining to support my work.

Join Investor's Odyssey

I’ve held Amazon for over five years and have added opportunistically, including during the 2022 lows at $89 per share. As the business has grown more complex, the position has taken a backseat while I focused on other investments. Its relatively small weight in my portfolio—and the need to refresh my thesis—ultimately led to this piece.

While Amazon’s headline growth has slowed, margins have improved materially, and the long-term outlook across its core and emerging businesses remains attractive. Much of this potential is already reflected in the stock, but I continue to view Amazon as a durable, long-term compounder at reasonably discounted prices.

What began as a brief update to my Amazon thesis evolved into a more comprehensive, two-part analysis examining the outlook for Amazon’s core businesses and how that translates into valuation.

Part 1 covers:

• Business Overview

• AWS - Bedrock, Trainium, secular growth

• E-commerce - robotics, margin expansion, shipping speed, e-commerce penetration

Part 2 will cover:

• Advertising - data advantage, DSP, increasing supply

• Emerging businesses - healthcare, Kuiper, and Zoox

• Valuation - assumptions and expected returns

Overview

Amazon has evolved into one of the world's most operationally complex businesses. Beyond its core e-commerce and AWS segments, it now generates meaningful revenue from advertising while continuing to invest in longer-term opportunities such as Project Kuiper, Zoox, and One Medical.

Growth
At nearly $700 billion in annual revenue, scale creates tradeoffs. It strengthens Amazon’s innovation engine by providing capital, talent, and the ability to absorb heavy R&D spending, but it also makes incremental top-line growth increasingly difficult. Generally, long-term equity returns are driven by revenue growth; for Amazon, however, a larger share of future returns is likely to come from earnings growth driven by e-commerce margin expansion and the continued growth of higher-margin businesses such as AWS and advertising.

Because e-commerce operates at very low margins, higher-margin businesses like AWS and advertising disproportionately contribute to earnings growth even when total company revenue grows more modestly.

Innovation
Amazon’s innovation broadly follows two paths: strengthening existing moats and incubating new businesses. Initiatives like Trainium and Prime Video enhance existing platforms while generating billions of dollars in incremental revenue. Trainium improves AWS’s unit economics, while Prime Video supports both subscriptions and advertising.

Meanwhile, healthcare, Kuiper, and Zoox represent bets that expand Amazon’s total addressable market while slightly reinforcing the broader ecosystem—for example, through Prime benefits tied to One Medical. Many of these investments depress current margins, but successful execution would meaningfully expand long-term earnings power.

Outlook

Amazon's investor perception has been shaped by several headwinds. Both investors and management extrapolated COVID-era e-commerce growth too far forward, and the subsequent slowdown exposed excess fulfillment capacity, pressuring margins. AWS also faced growth headwinds as enterprise spending moderated.

Some investors point towards Amazon’s stock underperforming, but this is only true in selective timeframes. Investors who bought near the COVID peak have earned sub-10% annualized returns, but this remains a relatively short window. Since year-end 2019, Amazon shares have compounded at roughly 17%, even as its EV/EBITDA multiple declined meaningfully—a natural outcome for a maturing business.

Investors also argue that Amazon’s declining EV/EBIT is evidence of value, but this metric is misleading. Historical multiples reflected expectations of sustained 20%+ growth and temporarily depressed EBIT following heavy post-COVID investment. EV/EBITDA and Price/Sales likewise fail to fully capture Amazon’s evolving earnings profile, as margins are still normalizing and higher-margin businesses are becoming more important contributors.

As a result, traditional multiples provide limited insight today. Even segment-level valuation is challenging due to limited profit disclosures outside AWS. For that reason, this piece focuses on understanding the underlying earnings power of each business before addressing valuation explicitly in Part 2.

AWS

AWS is hailed as an incredible business but faces continuous scrutiny among investors, including concerns over pressure from new entrants, Azure gaining market share, rapidly decelerating growth in 2022-2023, and their AI strategy. Despite this, AWS is still the market leader at ~29-30% share. From FY2022 through 3Q25, AWS revenue compounded at 16.5% and operating income by 26.7%.

Concerns intensified in 2022-2023 when growth decelerated sharply from 29% y/y in 2022 to 13% in 2023. Amidst macro-economic concerns, many businesses became focused on cloud cost-optimization rather than allocating new dollars to the cloud.

AWS took another reputational hit among investors in 2025. According to SemiAnalysis, Azure's dollar growth again outpaced AWS in 1H25, bolstered by its OpenAI partnership. While GCP is still growing much more slowly, the competition between AWS and Azure has been close.

But in 2024, AWS made a strategic partnership that will shift its fate. They became the “primary training partner of Anthropic,” working jointly on chip and data center development.

Results of AWS’s re-acceleration strategy speak for themselves: LTM AWS revenue reached $122 billion (an annualized run-rate of $132 billion) with RPOs of $200 billion (+22% y/y).

For 3Q25, y/y growth hit 20.2% and q/q growth was 7%—especially impressive given the business is much larger than two years ago.

Now, AWS’s AI strategy and its Anthropic partnership have begun to bear fruit with further growth re-acceleration expected in the coming quarters. Project Rainier—a 2.2 GW cluster built specifically to train and run Anthropic’s models—started operations in October 2025. Nearly 500k Trainium2 chips went online; 1 million chips and 1 GW expected by the year-end of 2025. Seven buildings are finished, with thirty buildings and 2.2 GW expected upon completion at the Indiana location.

Project Rainier is a key example of Amazon’s ability to broker deals and scale infrastructure quickly. In anticipation of data center demand, reports suggest they started scouting sites in early 2023, announced the $11 billion expected investment in April 2024, broke ground in September 2024, and Rainier was officially operational by October 2025—bringing initial capacity online in roughly 13 months. Wells Fargo’s analyst estimates Rainier could generate $14 billion in revenue at full capacity.

In contrast, Fairwater 1 (300 MW), the first part of Microsoft’s largest AI factory, has taken much longer. The Mount Pleasant investment was first announced in March 2023, broke ground in September 2023, and as of January 2026 (28 months later) was seemingly still not online. Other phases of the Mount Pleasant site are further behind as they paused in early 2025 to “evaluate scope and recent changes in technology.”

Fairwater 2 was faster at ~18 months from breaking ground to starting operations, but still noticeably slower than AWS. Both Fairwater facilities are also powered by Nvidia chips, in contrast to Project Rainier, which only contains Amazon silicon.

For Amazon, Project Rainier and Anthropic’s partnership is a signal of Anthropic’s long-term confidence in AWS, a mutually beneficial arrangement backed by an $8 billion investment from Amazon in Anthropic. Expansion will likely continue as Dario Amodei, Anthropic’s CEO, believes scaling is still the path forward (Dealbook Interview).

AWS has recently announced another investment in Indiana, this time $15 billion, though not specifically earmarked for Anthropic.

Why did Anthropic choose AWS?

In addition to the substantial equity investment and ability to execute on infrastructure expansion, it is their silicon program.

AWS’s Trainium 2 and the recently announced Trainium3 were codesigned in a close partnership between Anthropic and AWS’s Annapurna Labs. SemiAnalysis highlights the noticeably better total cost of ownership (TCO) versus comparable Nvidia chips. Tailored for Anthropic’s training strategy, Trainium has 25-30% better TCO per memory bandwidth. Amazon’s investment and this partnership are key reasons for Anthropic going heavy on AWS silicon with Project Rainier.

Amazon’s bet seems to be correct, Anthropic’s performance has been impressive this year, establishing itself as a leading enterprise AI provider, growing revenue by 10x, and expecting to be cash flow positive by 2027.

On top of the Anthropic partnership, AWS’s AI growth has and will come from several angles. Strands (agentic AI), Connect (AI customer service), Trainium, and Bedrock are each now billion-dollar businesses. Jassy believes Bedrock could eventually be as large as EC2, which has been estimated at 40-60% of AWS revenue, meaning $48-72 billion.

How Bedrock could become as large as EC2:

EC2 became AWS’s foundational service by abstracting away infrastructure complexity. Instead of owning and managing servers, enterprises paid for compute as a utility, allowing them to focus on making better applications. This model scaled with usage and became embedded in nearly every AWS customer.

For the AI era, AWS’s thesis is similar. Just as every application requires compute, every AI-powered application requires a reasoning and inference layer, provided through Bedrock. From coding agents to AI-powered customer service, these applications depend on scalable inference infrastructure. As model capabilities improve and costs decline, inference demand should scale rapidly. With its existing customer base of large enterprises and fast-growing startups, AWS is well-positioned to serve this demand.

Bedrock’s multi-model marketplace is equally strategic. Customers can compare costs across models and switch providers with fewer code changes. Since much of the world’s data already resides on AWS, it’s the logical platform for AI implementation—reducing friction and migration costs.

Finally, vertical integration creates a powerful flywheel. The majority of tokens processed on Bedrock run on AWS’s proprietary Trainium and Inferentia chips, delivering better economics for customers while preserving AWS’s structurally high margins. This mirrors their successful Graviton CPU strategy for EC2, which has proven that custom silicon can simultaneously improve customer TCO and AWS profitability.

Historically, EC2 relied on Intel x86, with many assuming ecosystem and compatibility constraints would limit migration to ARM-based systems. AWS’s Graviton ARM-based CPUs, however, now account for more than half of new CPU capacity added at AWS, and 98% of the top 1,000 EC2 customers use Graviton (Amazon News).

A similar narrative exists around GPUs: many analysts point to Nvidia's CUDA ecosystem as a form of lock-in for AI workloads. AWS's custom AI chips like Trainium are being positioned as lower-cost alternatives. While Microsoft is also developing custom silicon, its Maia chip appears further behind, potentially giving AWS and GCP early advantages in cost-efficient AI compute and associated developer ecosystems.

AWS still has room to grow:

“85%-90% of worldwide IT spend is still on-premises versus in the cloud. In the next 10 years-15 years, that equation is going to flip, further accelerated by companies’ excitement for leveraging AI.

AWS’s significantly broader functionality, stronger security and operational performance, and much deeper experience helping enterprises modernize their infrastructure bodes well for the AWS business moving forward.”
—Andy Jassy, Amazon 2Q25 Earnings Call

“Infrastructure modernization is much less sexy to talk about than AI but fundamental to any company’s technology and invention capabilities, developer productivity, speed and cost structure.

For companies to realize the full potential of AI, they’re going to need their infrastructure and data in the cloud.”
— Andy Jassy, Amazon 1Q25 Earnings Call

With these dynamics in play, a re-acceleration of cloud growth is plausible. Non-AI cloud services continue to expand even as enterprises rationalize spend, while AI introduces a new source of incremental compute demand.

Enterprises will scale AI usage both directly through AWS services like Bedrock and indirectly through third-party model providers. The indirect channel is particularly important: when customers use Anthropic’s direct API, inference runs on Anthropic’s underlying cloud infrastructure—AWS and GCP—generating incremental demand even when customers are not using Bedrock. This makes AWS’s partnership with Anthropic strategically critical, as Anthropic’s enterprise adoption drives underlying compute growth for AWS, albeit with different economics than first-party AI services.

All of these factors are further driven by the continuous scaling of computing used in our daily lives, the quantity of data produced, and more people gaining access to the internet.

Beyond AI, substantial opportunities remain for capturing on-premises IT spend:

Regulated industries like finance, healthcare, and government have migrated workloads cautiously, keeping most core systems on-premises due to data sovereignty concerns. On January 15th, AWS launched its European Sovereign Cloud based in Germany. AWS’s Sovereign Cloud offering directly addresses these data concerns, unlocking significant enterprise spending that has remained off-limits. Through 2040, AWS plans to invest €7.8 billion in Germany. Assuming demand is strong, AWS is expecting to continue this buildout across Europe.

Sovereign Cloud offers several layers of security with physical and logical separation from non-EU infrastructure. It will enable AWS to capture many workloads that haven’t migrated yet due to data residency and sovereignty concerns. Fortune Business Insights estimates the global sovereign cloud market at $155 billion in 2025 and estimated to reach $824 billion by 2032, a 27% CAGR.

Modernizing infrastructure represents another unexpected tailwind. Even as companies trim cloud bloat, those savings are often allocated to new migrations or growth initiatives. Cloud users recognize the value that the cloud provides and want more of it, not less.

Legacy data migration is a significant aspect of this, as companies are specifically moving legacy datasets to the cloud as a foundation for future AI capabilities. These data migrations can take a long time, ranging from months to even years, for large and complex setups, so companies want to be ready.

The question is, how much more can the cloud grow?

As Jassy mentioned, roughly 85% of global IT spend remains off the cloud. For 2025, global enterprise IT spending is estimated at $5.5 trillion (Gartner), while total spend on cloud services is ~$723 billion, and cloud computing is at ~$421 billion.

As shown below, the cloud computing market size by 2040 could range from at least $2 trillion to $7+ trillion, depending on the rate of IT spending growth and the share of cloud services. Even if cloud spending only takes a 35% share of IT (compared to Jassy’s 85%) and IT spend grows at just 5%, cloud compute will still grow double digits for the foreseeable future. For AWS, this implies $500 billion to $2+ trillion in potential revenue by 2040.

As the security, sovereignty, simplicity, and AI use-cases improve, the cloud is bound to continue taking share.

Despite the optimistic outlook, there are still risks

Market trends aren’t completely black and white. Some sources reference cloud repatriation, with companies and governments relocating workloads back to private clouds (Broadcom Paper). Customers have become concerned that cloud spend is becoming too high and much of it is wasted. For large tech companies, this can be millions of dollars per year (Duolingo Case). But most of the research suggesting major repatriation is sponsored or produced by private cloud suppliers like Broadcom.

Edge computing presents another trend against the cloud. As processors become smaller and more efficient, more computing can be done on the device rather than in centralized data centers. It is expected that a greater amount of AI inference will move to the edge, which can improve model response time. This isn’t guaranteed, however, as edge devices can have significantly lower processing power and memory available.

Customers also report seeing rising AI costs under AWS and are considering alternative cloud solutions like Neoclouds. Business Insider reported on a memo leak of discussions between Capital One and Nvidia about concerns over AI costs under AWS and alternatives to control costs. Specialized providers like Coreweave have grown rapidly as they present lower costs by focusing on AI infrastructure, don’t have massive overhead in-terms of legacy CPU infrastructure, and don’t offer countless non-AI related services.

While these risks warrant monitoring, several factors suggest they’re manageable. AI costs continue declining, AWS’s custom silicon provides cost advantages, and scale enables competitive pricing. AWS has also been offering these AI services while maintaining overall mid-30% operating margins. Although undesirable, AWS could lower AI compute prices to maintain or gain market share if needed.

Nonetheless, the general cloud trends seem to be towards further cloud adoption, especially for enterprises wanting to leverage AI. Once customers make the effort of migrating systems and data onto the cloud, moving on-prem or even to another cloud provider becomes unlikely. AWS’s AI powered Transform, which helps migration onto the cloud even more easily, will only speed up cloud adoption.

The market is growing, and AWS has held the position as the market leader. Azure is the closest competitor and is empowered by their generous free credits and horizontal solution bundles, but there is a place in the market for both. It appears the AI compute market will be more fragmented than traditional cloud, but AWS’s competitive advantages continue to be important.

With AWS discussed, let’s now turn to e-commerce, another area I believe has a strong outlook.

Subscribe now

E-Commerce

The namesake and origin of Amazon lies in their e-commerce business. Beyond revenue from goods sold, it has served as an incredible platform for new revenue streams: Prime subscriptions, Prime Video, Fulfillment by Amazon, and advertising, to name a few.

Amazon’s e-commerce business has run on razor-thin margins for decades as they continue to expand infrastructure, add new product lines, and always aim to offer lower prices. Due to the size of this business, understanding the true margin profile and what is embedded in those thin margins is very hard to comprehend. It’s hard to tell how much margin growth or shrinkage is driven by e-commerce economics versus advertising, grocery, healthcare, and Prime Video.

Investors have little insight into distribution network investments and whether they’re operating at full capacity. Distribution investments have occurred in large spurts, such as during COVID, and subsequently have been made more cost-efficient.

Emerging technological developments are also embedded in the low margins. Amazon likely recognizes billions in operating expenses from currently unprofitable ventures, like Zoox and Leo. Tens of billions of growth expenses are also recognized from growth and cost-saving initiatives, including robotics and drones for e-commerce and AI services for AWS.

These can cause a large drag on earnings before they become successful enough for the revenue or cost savings to outpace development costs. Once some of these bets start to pay off, however, they produce a fantastic case for earnings growth.

The e-commerce margin bull case is robotics

In 2012, Amazon acquired Kiva Systems for $775 million to continue innovation on automated fulfillment technology. While innovation and new robotics have continuously developed since then, the largest cost savings are just on the horizon.

Amazon has now deployed over 1 million robots of 10+ different types across operations, many of which have been launched in the last two years. It ranges from robots to retrieve bins of inventory to ones that can transport 2,500 pounds, carefully pick up small items, and create custom-sized paper packaging.

Many of these robots work in confined areas of the warehouse, separate from human workers. Autonomous technology advancements will enable robots to work in the same spaces as humans, improving efficiency and safety. Amazon’s new robotics AI model, DeepFleet, has already improved travel time by 10%.

With these advancements, Amazon’s first “next-generation” fulfillment center was opened in Shreveport, Louisiana, in September 2024. The robots have been designed to be installed in current distribution centers and new specialized ones, meaning this can be done somewhat efficiently. Amazon’s mid-term target is to automate 75% of shipments. In the near term, they plan to roll out 40 more next-gen warehouses by the end of 2027; compared to Amazon’s current global network of >1,300 logistics facilities, of which ~350 are large fulfillment centers, there is a long way to go.

Amazon is already seeing fulfillment processing times reduced by up to 25% in these facilities and increased capacity for same-day and next-day delivery. As a result, Amazon estimates it can also reduce the cost to serve during peak seasons by 25% in the next-gen facilities. Robots are great for fulfillment margins as it turns a variable cost (labor) into a fixed cost. Rather than closely scaling with unit growth, the robots become cheaper per unit the more demand they have.

Saving 25% of fulfillment costs is incredibly valuable for customers and 3P sellers through lower prices and greater margins for Amazon. Fulfillment expenses were $106 billion for LTM 3Q25, and worldwide shipping costs were $100 billion. Costs will be improved by quicker fulfillment speed, better accuracy, and trading variable costs for fixed costs. Shipping expenses will be slightly reduced as sorting becomes more efficient, but the greatest gains will likely be in fulfillment labor. It’s estimated that labor makes up roughly 60% of fulfillment costs. Based on today’s numbers, that would mean automation alone could drive savings of ~$16 billion per year. Although these estimates are imprecise, they give strong insight into the magnitude of potential savings.

If even half of these savings accrued to Amazon, that would boost operating profit by $8 billion per year, boosting the combined NA and International margin from 6.4% to 7.8%. Even more optimistic, Morgan Stanley analyst Brian Nowak projects that even 40% automation could drive savings of $10 billion per year. This is significant considering non-AWS operating income was just $36 billion for LTM 3Q25.

Amazon has established a unique position in the AI and automation trend, focusing on the physical application of these advancing technologies and becoming the leading manufacturing automation enterprise. While this could even theoretically become a revenue-generating business, the cost-savings alone are great for investors. Amazon Robotics is yet another example of Amazon’s continued ability to drive innovation through acquisitions and internal development.

Amazon’s e-commerce growth is positioned to continue as well

Intuitively, Amazon’s e-commerce business appears massive and would seemingly have already saturated the US market, but this is far from the truth. Compared to other large countries, the US e-commerce share of retail sales is relatively low. Excluding automotive and groceries, South Korea’s e-commerce share is estimated at 54%, the UK is 41%, Japan is 27%, and the US is 26%.

Factoring in groceries pulls the share down further, as groceries are less penetrated. I estimate the US at 22%, the UK 28%, China 30%, and South Korea 50%.

These figures may even understate the gap in urban e-commerce adoption. China’s 30% national figure includes vast rural areas with limited delivery infrastructure; in cities comparable to US metros, e-commerce penetration likely reaches 40-50%

*Note these numbers aren’t perfectly comparable, as numbers and categories are reported differently by country, but the general relationship is correct, and the data is directly sourced from the government’s statistical office websites.

Several structural reasons led to the regional disparity. Population density is much higher in countries like South Korea, China, and the UK compared to the US. Greater population density means last-mile delivery is easier and cheaper. The number of distribution centers needed in the U.S. to fulfill quick delivery is substantially greater, making inventory planning and logistics much more difficult. Even within most cities, housing, grocery stores, and restaurants are more spread out than you would see in China or much of Europe.

Cultural factors are also at play, i.e., big-box stores, suburbs, driving culture, and in-store shopping being an activity. E-commerce share has continued to grow faster than total retail, however, increasing from 9.6% of retail sales in 2018 to the 16% in 2024. The pace has slowed, but will likely continue as e-commerce speed, quality, and price improve.

Fulfillment improvements have begun to tackle this geographic hurdle

Amazon’s distribution system used to be a national system, which meant that if the local FC was out of stock, packages might be shipped from across the country. As the scale and number of FCs grew, this created incredible complexity. Shipping costs (for Amazon) were unnecessarily high, and delivery speeds were slower.

After doubling their fulfillment network during COVID, Amazon recognized large inefficiencies, electing to shift to a regionalized model by 2023, organizing into 8 geographic regions that can largely supply themselves. This reorganization reduced cross-country shipping dramatically.

Since each FC now has to send goods to fewer locations, the trucks are fuller, meaning better cost efficiency. Progressions in machine learning have also made this much easier, improving inventory model accuracy so they can better stock distribution centers to optimize delivery speed and stockout costs

Amazon is continuing to innovate on its distribution network, increasing speeds for perishable grocery, rural areas, and same-day delivery. Customers can now order perishable groceries alongside regular goods in the same Amazon cart and have them delivered for free the same day with Prime. Since launching in January 2025, it has shown strong potential, doubling the frequency with which customers visit Amazon.com versus non-perishable shoppers. As of Q3, this option was available in 1,000 cities/towns and expected to be in 2,300 by year end 2025. Amazon is also doubling the number of rural areas that have access to same-day and next-day delivery. Grocery should provide a strong growth lever as it only makes up an estimated 12-14% of U.S. grocery sales.

Amazon is also putting more effort into establishing same-day delivery centers, carrying the most frequently ordered goods, so customers go to Amazon for things like diapers and toothpaste rather than having to drive to a store. Same-day delivery centers are also less expensive than traditional FCs, meaning the trend is great for investors and customers.

Another barrier to e-commerce growth is fragmentation, where many branded goods historically couldn’t be found on e-commerce sites like Amazon. It causes a problem for e-commerce sales overall, as ordering from direct retailers often leads to higher shipping costs for the consumer and seller. Shipping costs can deter customers from purchasing online or from purchasing the goods at all. Fragmentation also means Amazon’s moat has room to improve, moving further towards being the everything store and making the Prime subscription more valuable.

After previous pushback, more and more brands are realizing the value of selling on Amazon. In 2019, Nike halted selling products on Amazon to focus on company-owned channels, in mid-2025 Nike announced they are returning to directly selling goods on Amazon US. As noted in the Q3 earnings call, selection grew 14% q/q, including premier brands like North Face and Charlotte Tilbury, and Amazon’s everyday essentials category continues to grow fast as customers realize the value of ordering necessities through Amazon. I expect this to continue as shipping speeds improve and selection continues to grow.

Projecting specific growth rates is difficult, but I expect Amazon can maintain ~9-11% annual growth for the e-commerce business (1P + 3P) in the coming years. 3P will also likely continue to grow faster than 1P, which is a positive for margins.

Many of these trends apply internationally as well. While competition is fiercer abroad, Amazon’s automation capabilities should provide a competitive edge in shipping speeds and pricing. International success matters, but it isn’t core to my investment thesis.

Recap

AWS’s re-acceleration is driven by AI platform expansion (particularly Bedrock), custom silicon advantages, and the secular shift of IT spend to the cloud. With ~85% of global IT spend still on-premises and AI creating new compute demand, AWS has a long runway for double-digit growth.

E-commerce margins will benefit from robotics automation (potentially $8-16 billion in annual savings) while revenue continues growing as Amazon captures a larger share of retail, from everyday essentials to branded products and groceries.

With the outlook for AWS and Amazon.com covered, Part 2 will examine their advertising business, which has multiple growth drivers and competitive advantages that became even more compelling the deeper I researched. After covering advertising and emerging businesses, I’ll discuss how I value Amazon and what expected returns look like.

Part 2:

• Advertising - data advantage, DSP, increasing supply

• Emerging businesses - healthcare, Kuiper, and Zoox

• Valuation - assumptions and expected returns

Read full disclaimer here.

Thank you for reading my work; every subscription, like, and comment helps me continue to produce these reports.

Join Investor's Odyssey

Share

See my most recent posts:

• Dino Polska Update

• When Quality Fails

Read the two-part write-up on a Finnish serial acquirer: Part 1, Part 2

Understand my approach to portfolio construction: Types of Companies I Invest In

Consider joining to support my work.

Join Investor's Odyssey

As I have reduced some of my other holdings (see: STRL Update), I’ve been reevaluating my current positions and searching for new ones to buy at favorable prices. As a result, I’ll be releasing a few shorter posts updating my holdings. The first is Dino Polska; posts on Amazon and Duolingo will likely follow.

Dino Polska

I have not posted a full write-up on this company—others have done great in-depth work on it, and I don’t believe adding another full write-up is necessary. Instead, I’m going to provide insight into why I still hold the company and why I’m adding on. If you wish a full report on Dino Polska, I highly recommend Sohra Peak Capital’s 2022 presentation, which goes in depth on the business model and competitive landscape.

This post will cover:

• Brief Overview

• Current Situation

• Valuation

• Metrics To Follow

Brief Overview

Dino Polska is one of the fastest growing grocery chains in Poland. Their model is based on a standard 400 sq meter store with 4,000 SKUs, primarily situated in small-to medium-sized towns. They provide low prices but aren’t as aggressive in discounting and bundling as traditional discounters (e.g. Biedronka and Lidl). Dino’s format would be hard for these competitors to replicate due to their existing distribution networks tailored for different SKU counts and product variations. The stores would also likely cannibalize existing larger format stores.

Among customers, Dino is known for its butchered meat counter. All of Dino’s fresh meat is supplied by their own processing plant (Agro Rydzyna), a contrast to the prepackaged meat at competitors. This is crucial as many Poles value the freshness of food, and meat is a large part of their diet.

Dino first expanded in Western Poland, with the highest saturation at 19.2 stores per 10,000 inhabitants as of Q3. Country-wide saturation of Dino stores is currently 7.8. Compared to large-format competitors, Dino can profit on a smaller customer base without cannibalizing other stores. They are still currently unknown to much of the eastern population, meaning brand recognition can still grow greatly from here. Based on national grocery spend, Dino’s current market share is likely ~5-7%.’

Dino also contrasts competitors by owning the land their stores are on, driving cost savings over the long-term.

The small-format and convenient location will likely work well in larger towns as well—Poles increasingly value convenience due to limited free time. Zabka has profited from a similar trend, growing rapidly like Dino but with convenience stores situated in cities. Zabka serves the higher-priced ready-to-eat category, while Dino is an affordable quick grocery store. Zabka is another fascinating company I’ve considered investing in, but the debt is rather high and the strategy somewhat fragmented.

Dino has another positive in that it is founder-owned, 51% by the founder and Chairman, Tomasz Bierznacki, now the third-wealthiest person in Poland.

Current Situation

This stock has moved relatively nowhere in the past two years as the market has re-rated the business. The last five years haven’t been much better, with only a 6.9% CAGR for the stock despite revenue and operating cash flow compounding over 26%.

This is a similar tale to Amazon’s performance over the past five years. Multiples have declined despite strong performance, as investors’ expectations have softened.

There are understandable reasons for the sentiment shift: Poland has faced rampant inflation, then disinflation, while minimum wage increased >19% in 2023 and 2024. Rising costs combined with a weary consumer led grocers to discount more than normal, and shoppers are continually cost-conscious. Dino’s margins have hit their lowest in a decade, slowing down earnings growth. The uncertain market also led Dino to build fewer stores in 2023 and 2024; store count growth has since re-accelerated in 2025. Nonetheless, Dino has fared better than competitors like Biedronka in terms of organic growth and margins. I believe the outlook from here will be more favorable for several reasons:

Depreciation was one of the fastest rising costs, outpacing revenue growth. This is unsurprising given the recent buildout: three distribution centers, 361 new stores, and the expansion of the meat processing plant finished construction in the last twelve months, bringing the total to eleven distribution centers and 2,933 stores as of Q3. A twelfth distribution center will be launched in Q4.

Dino distribution centers can service 350–400 stores, meaning eleven can serve 3,850-4,400, considerably more than the current 2,933. This excess capacity means mature margins will be considerably higher once stores and distribution centers reach full utilization. Additionally, depreciation overstates true maintenance costs, as upkeep of company-owned stores isn’t equal to the straight-line depreciation of constructing a new one. As Dino builds more stores, the discrepancy between depreciation and true maintenance costs will likely grow.

Increased wages have yet to be fully reflected in grocery spending. Minimum wage growth initially lagged inflation while Dino LFL sales outpaced it; the trend has now switched, causing downward pressure on margins. Consumers are still spending conservatively, and Dino had to offer greater discounts. However, I expect this to change. The real minimum wage is 20% higher than in 2021, meaning there should be greater buying power, especially in the towns where most Dino stores reside.

Now, as Dino’s revenue and wage growth have balanced out and minimum wage increases are expected to slow (only 3% increase in January 2026), Dino’s margin should be able to stabilize and grow again. After a decline in 2023, EBITDA margin has stayed at 7.8% YTD 2024 and 2025. While EBITDA margin is far from my preferred measure, it helps compare profitability, aside from potential D&A distortions.

Beyond infrastructure and wages, operational leverage is improving. YTD Revenue grew 14.9% while COGS grew only 14%, increasing gross profit by 18.1%. Dino is continuously gaining negotiating power with suppliers through a larger store base and more efficient operations. So, although operating expenses have been volatile, it’s clear Dino is still benefiting from scale economics.

The growth runway is still strong based on current store density. From 2,933 stores, Dino can double or triple their store count in Poland. Assuming Dino opens 330 stores a year (similar to historical levels), that results in ~6,200 by year 10, which is still well below other estimates of 8,000-11,000. Dino also has the potential to grow further by expanding into neighboring countries and rolling out gas stations, but these aren’t likely to happen in the near future.

Debt levels are very low. Net debt is only PLN 192 million (excl. leases), and Net DEBT/EBITDA is 0.11x. This has two impacts: Dino is very conservatively capitalized, able to handle economic uncertainty, and their interest costs will be much lower compared to growing income.

Valuation

Dino is down 27% from the 52-week high and only up 5% from a year ago. The stock trades at ~PLN 41 per share, a price it could have been bought at two years ago. It is valued at 15-16x EV/EBITDA, the lowest valuation since going public in 2017 and well below the average of ~20-25x in recent years.

While this provides important context for relative valuation, it’s important to consider how much growth this is pricing in.

If you calculate free-cash-flow based on depreciation, FY25 cash flow will likely be PLN 2 billion, a 6.1% margin. Estimating true maintenance costs of the distribution centers and stores, FY25 FCF will likely be PLN 2.4 billion, a 7.3% margin.

First, let’s consider what the market is pricing in:

Even assuming the depreciation-based FCF is correct, the market is expecting Dino to grow just 10% annually over the next ten years to achieve an 11% return, then stop growing afterward. With my estimate of FCF, the market is expecting only 7.6% growth.

Now, if we take a look at various scenarios, you can see that an adequate return will be achieved even if growth is well below the historical average. If Dino adds around 300 stores per year and LFL sales grow 3-4%, you can expect a high teens return. Even at 0% LFL sales and 150 stores per year, the expected return would be over 10%.

Amidst the recent margin pressures, I figured it would be useful to see how long-term margins would impact returns. They end up having little effect. Even if the FCF margin dropped to 6% by 2035 (versus the current estimated 7.3% margin) and LFL sales growth stagnated, you could still achieve a low-teens return by virtue of store count growth.

Metrics to Follow

Store Openings — As discussed, this is the most important factor for long-term returns. If new store openings drop, it could signal greater competition, the TAM being smaller than expected, and struggles to find land and build new stores.

LFL Sales Growth — This is especially critical to compare against competitors such as Biedronka and Zabka (though Zabka is not as direct of a competitor). If Dino LFL growth is negative or noticeably lags others, that could signal customer sentiment shifting. Dino stores could decline in quality, competitors could produce similar offerings, or offerings like Zabka could grow in favor. While such a sentiment shift is unlikely, it’s important to monitor.

Conclusion

Dino Polska has fallen out of favor among investors due to declining margins and an uncertain macroeconomic environment. But fundamental business performance is stronger than reported metrics suggest. Depreciation and non-mature stores overstate expenses compared to their potential revenues. The economic situation is also likely to improve as inflation and wage increases have slowed, and Dino continues to execute on their expansion plans.

Current store density implies over ten years for Dino to continue growing, achieving higher margins and better buyer power in the process. The market is not pricing in this growth runway, instead implying Dino can only grow at <10% per year.

Thank you for reading my work; every subscription, like, and comment helps me continue to produce these reports.

Join Investor's Odyssey

Share

See our most recent post on the collapse of a Peter Lynch Favorite in the 1990s: When Quality Fails

Read the two-part write-up on a Finnish serial acquirer: Part 1, Part 2

Understand my approach to portfolio construction: Types of Companies I Invest In

Read full disclaimer here.

Investor’s Odyssey tracks my evolution as an investor, aiming to help others learn along the way while challenging my own ideas. I’m a value investor at heart, focused on quality companies that can be held for the long term and searching across geographies and market sizes to find them.

Consider joining to support my work and make this research sustainable.

Join Investor's Odyssey

Hi all, I know this is coming slightly late after earnings, but my main goal was to provide context into two companies I follow and their outlook amidst downturns in their stock price and/or business.

In this update you can expect:

1. Texas Roadhouse

   1. Quarter Update

   2. Business Discussion

   3. Valuation

2. Topicus

   1. Quarter Update

   2. Business Discussion

   3. Valuation

Texas Roadhouse (NASDAQ: TXRH)

TXRH is coming off a strong 2024 where EPS grew 37.5%1 and operating margin increased 200bps. The last few quarters have backtracked much of this expansion. As a result, the stock is down ~13% in the last year.

Heightened beef and wage inflation led to much of this deterioration, and they are forecasting similar inflation for next year.

The market continues to be short term focused, forgetting the decades of consistent growth they have achieved and strong brands behind them. Since 2019, TXRH recorded a relatively consistent ~14% CAGR in revenue, 16.7% CAGR in operating income, and a decrease in shares outstanding by ~7%.

Let’s take a look at the results.

Quarterly Results (vs prior year):

• Revenue grew 12.8% (13.5%)

• Operating income declined -5% (+38.1%)

• Restaurant margin decreased 168 bps to 14.3% (+140bps to 16%)

• Avg weekly sales of $157,235 ($149,176) per store with $21,409 ($18,914) from to-go sales

• Commodity inflation of 7.9% (1.3%) and wage/labor inflation of 3.9% (4.7%)

YTD (vs. prior year):

• Revenue grew 11.7% (13.5%)

• Operating income remained flat (+39.9%)

• Restaurant margin decreased 118 bps to 16% (+180bps to 17.2%)

• Commodity inflation of 5.1% (0.8%) and wage/labor inflation of 4.1% (4.5%)

As can be seen, on a quarterly and YTD basis, TXRH continued to perform on revenue growth but lagged on margin.

This profit stagnation is a result of their pricing policies amidst inflation. Management aims to only increase prices based on structural inflation, not cyclical inflation:

We’re not going to be able to price for every beef inflation as of right now. But we want to make sure that we protect the value side of our business and our menu and our perception.

What’s important to understand is beef prices are genuinely cyclical. Price per pound may increase for several years, then stagnate or drop:

The cyclical trend can be seen in their operating margin as well:

The craziest part is, their operating margin is still great! It’s the second highest in the past ten years, and YTD operating profit is 36% higher than two years ago!2

So what does this mean for TXRH?

Texas Roadhouse’s pricing choices make menu pricing more consistent, which is much easier to digest as a customer.

Compared to commodity inflation of 5.1% and double-digit beef inflation, LTM price increases were just 2.3%. FY25 increases will be similar to 2024 at 3.1%

Why do they do this?

If people saw steak prices at TXRH jump 14% this year, they’d be exclaiming the value is gone, and they’ve sold out to Wall Street. Instead, customers are recognizing the growing value versus buying steaks at the grocery store. Management commented that they’re actually seeing more people buy steaks! Importantly, steaks carry similar profit dollars to other entrees but a lower profit margin, contributing to margin depression in addition to inflation impacts.

Texas Roadhouse choosing to take the brunt of some of this price increase means they are strengthening their brand. These customers are legitimately getting an increasingly better offering for TXRH steaks versus buying one at the grocery store. So when beef costs do eventually stagnate or drop, Texas Roadhouse can capitalize on the sustained demand and profit boost.

Similar to companies like Amazon, Walmart, and Costco, the goal is to grow the value offered to consumers. They do this by sharing some of the operating leverage with their customers. This improves brand image over time and grows traffic. Famed investor Nick Sleep termed this model “scale economics shared,” and believed it was the strongest business model.3

For TXRH, as SG&A has shrunk as a percentage of revenue, gross margin was lowered, but operating margin was sustained. They aren’t just maintaining the image of a good value steakhouse, they’re improving it while keeping margins.

Rather than trying to prop up short term profits, the TXRH team has stuck to their guns and not compromised on value or quality.

Consistency and quality are paramount as you grow. Accept lower profits if you can ensure your product is always going to be as good as the original.

- Kent Taylor, Founder of Texas Roadhouse

Business Discussion

Overall guest traffic continues to grow, with a 4.3% increase compared to 3.8% in the prior year’s quarter. Per person average check increased 1.8%, but declined 0.5% when considering LTM price increases. This led to overall comparable restaurant sales growth of 6.1% which is fantastic for a 32 year old company.

The pace of store openings remains consistent, with a shift towards newer concepts in the upcoming year. Historically management has targeted opening 30 company owned stores per year across restaurant concepts, next year they’re guiding toward 35 restaurants. They expect 20 Roadhouses, 10 Bubba’s, and 5 Jaggers. Management said Roadhouse openings might be greater than twenty, however.

The company continues to buy back domestic franchises, with 17 bought back YTD. These are immediately accretive. An average acquisition price of $5.54 million per store — lower than the cost to build a new one — and average $1.38 million restaurant margin, implies an immediate ROI >20%.4

Unfortunately, this runway is limited, three more were acquired in Q4, and the last 5 in California will be in 1Q26, leaving only 31 domestic roadhouse franchises left. If they were to acquire the rest, it would yield an approximate $39 million of additional margin dollars.

Valuation

In aiming to estimate Texas Roadhouse’s current value, I took a few approaches. Below you can find a simple 15-year reverse DCF and also longer term sensitivity table that contains more assumptions.

As Texas Roadhouse is continuing to deploy much of their free cash flow into growth, an estimate for maintenance free cash flow was used. Thankfully, the company provides capex by type in their annual reports, so we can understand what percent comes from new builds versus refurbishment and existing store expansion. This resulted in an estimated $616 million of maintenance free cash flow for FY25, compared to $350 million of LTM FCF based on total capex spend.5

For a simplistic view of expected returns, see below for a 15-year reverse DCF:

Using a 10% discount rate and 3% terminal growth, the market is pricing in~5% growth, less than half their historical average.

If you were to estimate a more conservative FCF of $550 million, the market still expects only 6.5% growth.

Sensitivity Table

Out of curiosity, I decided to model TXRH based on fifty years of cash flows rather than assigning a terminal value or exit multiple. If anything, this gives more insight into the future of the company than just assuming the company grows into perpetuity. The tables shown below have assumptions for mature margins and same-store sales growth for the two main franchises, Bubba’s and TXRH. Jaggers is important for the future as well, but currently contributes a very small amount of revenue.

I assumed management maintained their historical 30-35 domestic store openings per year. One scenario assumes they stop at 900 Roadhouses (management communicated target), the other assumes they stop at 800 and only grow Bubba’s and Jaggers.

For context, the domestic company owned count for 3Q25 was 638 Roadhouses, 54 Bubba’s, and 10 Jaggers.

The 800 store scenario would be reached by 2033 and 900 stores by 2038. To be somewhat conservative, after 2038 I assumed zero store openings across all franchises and growth to drop to an inflation rate of 3% for all three franchises.

The outputs of the tables are implied returns from the cash flows from year one to fifty, no terminal value.

Result

For the 900 store scenario, the result was roughly a 12% implied return based on realistic assumptions.

The output conveys how margin has only a modest impact on long term returns. Even a 15% margin versus 17% margin has just a 1% impact on returns over the long run. Stopping at 800 Roadhouses also had a small impact of just 0.3% on implied returns. Even a most modest scenario of 2% LFL growth and 800 Roadhouses results in a 10% return.

This does not take into account additional growth from international franchises, domestic Jaggers franchises, or their grocery store products.

In fact, you could assume they don’t open any new stores after 2026 and still achieve a reasonable result:

While you won’t likely attain high double digit returns, one could conclude it is a relatively safe investment with a strong brand.

You could of course disagree with one or many of the assumptions I made, ranging from free cash flow to margins and terminal growth rate. But, hopefully this illustrates what I think of the company after following it the past few years.

Topicus

For those who don’t know, Topicus is a spin-off from Constellation Software, they both acquire vertical market software (VMS) companies to hold perpetually, with Topicus focusing on Europe.

Topicus and their parent company have been a topic of continuous debate amidst AI concerns, the resigning of Mark Leonard, and ever-increasing competition. Nonetheless, it is business as usual for Topicus, posting great Q3 results.

LTM 3Q25 Performance Metrics

• FCFA2S of €203.9 million, 20.4% y/y growth

• Revenue of €1.48 billion, 19.4% y/y growth

• €281 million of cash deployed in acquisitions

  • €198 million of which was from Cipal Schaubroeck completed in Q26

  • An additional €414.8 million cash used/committed for the Asseco investment

  • Another €30 million in cash was committed after the quarter

• Asseco Poland stake increased 14.48% to 24.83%

  • Earnings negatively impacted by choosing to recognize Asseco stake using the equity method. This has no real impact to Topicus’ economic earnings or FCFA2S.

As can be seen below, Topicus has deployed significantly more cash into acquisitions in the LTM than any year previously.

The visual above adjusts cash deployed to sum cash flows spent on acquisitions, investments in other equity securities, and changes in restricted cash for pending acquisitions. So, it reflects their recent investment in Asseco fully. For comparability I added the 2022 €67 million preferred dividend back into FCFA2S, but of course that wouldn’t have been investable cash in that year.7

You can also argue FCFA2S doesn’t fully reflect investible cash. Although they sometimes distribute cash to non-controlling interests, it is not consistent year to year, so their reinvestable cash is somewhere in-between FCFA2S and FCF.

You can see cash deployed versus overall FCF here:

As can be seen, cash deployment was much lower in previous years, and dividends have been paid out multiple times, including a preferred dividend in 2022 and common dividend in 2024.

This is unsurprising and characteristic of acquirers, some years several large investments come about, other years it’s more sparse. However, the dividend payments do reflect somewhat different cash retention principles than the parent company, CSU. Nonetheless, Topicus has continued to keep ~20%+ growth rates and reinvest much of their capital.

Business Discussion

As Topicus doesn’t provide any letters or quarterly earnings calls, there is less to comment on in this quarter particularly, but there’s some general trends to comment on.

Topicus is clearly different from its parent company.

They aren’t afraid to venture into different business models and growth avenues, similar to CSU’s recent style creep, but beginning at a smaller size.

The key to Topicus (and CSU’s strategy) is not just pure VMS anymore, it is where they can achieve high recurring revenues. This is similar to style creep I have recognized in SaaS private equity firms as well; if something is high-recurring revenue and mission critical, does it matter if it isn’t pure VMS or SaaS?

In the 2025 AGM in May, Robin also commented on the difference between Topicus and the two TSS segments.

[Topicus operating group] is doing a little bit more synergistic kind of acquisitions, which TSS historically hasn’t done, where they use that developer capability, where they use their market positions in existing verticals, where they try to build value chains in those verticals.

Topicus isn’t just focused on low growth independently run companies, they are actually developing software companies and considering how they can complement each other. You could argue this helps increase their potential value creation and market size as they aren’t only focused on one type of company. It may also cause worry as this strays from CSU’s time tested VMS independently run model.

Now, with their ~25% stake in Asseco, a public company, we have continued to see Topicus venture into other investment strategies.

Valuation

I previously preferred to calculate value based on the FCFA2S and undiluted share count, however that is becoming increasingly complex with the Asseco shareholding and Topicus’ slowly increasing ownership share versus non-controlling interests (64.18% in 3Q25 versus 63.07% in 4Q22).

To adjust, I added back NCI to FCFA2S to approximate the diluted company’s free cash flow including the shareholders of Topicus Coop. LTM FCFA2S was €204 million and FCFA2S + NCI was roughly €320 million. The NCI add back was also adjusted to exclude the estimated NCI share of Sygnity, Geoactive, and Geosoftware as those aren’t part of the diluted cash flow value. The calculations aren’t perfect and still don’t fully recognize recent acquisitions’ full year impact, but it serves as a rough conservative proxy for free cash flow to the diluted company.

To consider how different methods of appraising Topicus’ ownership of Asseco affect valuation, I calculated one scenario with the equity value added and the other with Topicus’ share of Asseco’s cash flow added. The results were roughly the same, so just the cash flow method is presented below.

I elected to use multiples for the terminal value as it would be more likely than not Topicus will continue to grow above inflation at year 10. It also makes it easier in a sensitivity table that displays a discount rate as the output since traditional terminal value calculations necessitate a discount rate and terminal growth rate.

In the first table you can see the year ten diluted market cap (in $CAD) for Topicus based on various growth rates and exit multiples. You can also find a comparison between CSU’s current market cap and P/FCFA2S multiple versus Topicus.

For comparison purposes against Topicus’ year ten outputs, CSU’s current market cap is presented below as well as an adjustment for ten years of 3% inflation to make the output more comparable to CSU’s size.

FCFA2S multiple was used for CSU as that was easiest. TOI’s undiluted market cap/FCFA2S is also not much different than the diluted/FCF calculation.

CSU’s trailing P/FCFA2S is 28x compared to 27x for Topicus, and this is after a drawdown in both stocks. If you assume Topicus grows at 20% and trades at 28x FCF on a diluted basis, they would be similar in market cap and FCF to CSU currently.

Currently, I think the market’s focus is with underperformance risk, so most of the scenarios presented below are if they surprise to the downside rather than the upside.

The table presents the weighted return of all cash flows the company produces. Essentially the long-term return implied based on the current diluted market cap, ten year growth, and prospects afterwards.

As a result, if the terminal multiple were to contract to 16x (versus 27x LTM) and FCF only grow at 12% per year, the business would produce a roughly 10% return. If they were to achieve a 20% growth rate and maintain a mid-to high 20s multiple, you would achieve a ~20% return.

Note, a 12% growth rate is still great for any company, so even the 12% growth and 16x FCF is pricing in above average performance. Nonetheless, CSU and its subsidiaries have proven the ability to maintain above average performance for abnormally long periods of time.

You may also take issue with the 8% growth of Asseco FCF. If you were to drop the growth rate to 4%, it would impact returns by roughly .3-.4%.

For those that are more interested in ten year returns based on the market-cap CAGR, those are presented below. As can be assumed, these are much lower as it doesn’t convey all the value a company creates, rather just the return based on current entry value and exit multiple.

I’m sure many will take issue with my valuation approach and calculation of cash flows and returns, but this is simply to present my view of the scenario and the value each of these companies create.

The free cash flow calculations are relatively rough estimates and don’t include the full-year impact of recent acquisitions. Make your own adjustments accordingly and choose what assumptions make sense for you.

Nonetheless, feel free to reach out with any questions, mistakes you believe I made, or just how you view the situation.

If you’re interested in learning more about terminal value and using multiples, I highly recommend The Footnotes Analyst newsletter: Exit Value Terminal Multiple

Thank you for reading my work; every subscription, like, and comment helps me continue to produce these reports.

Join Investor's Odyssey

Share

Read our two part write-up on a Finnish serial acquirer: Part 1, Part 2

Understand my approach to portfolio construction: Types of Companies I Invest In

Disclaimer: The writer currently holds positions in both companies discussed.

Read full disclaimer here.

Consider joining to support my work and make this research sustainable.

Join Investor's Odyssey

This is part two of a deep dive on Relais Group, a Finnish serial acquirer. Refer to part one if you haven’t read it yet.

In Part One, we covered:

1. Aftermarket Parts and Repair Industry

2. Business Segments

3. Competitive Advantages

4. Compensation and Ownership

Part Two will discuss:

1. Recap

2. Outlook

   1. Strategy

   2. Market

   3. Risks

3. Financial Considerations

4. Valuation

Recap

So far, we have discussed why the Nordic commercial aftermarket has been a ripe industry for consolidation and relatively low competition. We have highlighted how Relais operates in several steps of the value chain including parts wholesalers, repair shops, and specialty parts/equipment suppliers. Experience and vertical integration should improve their ability to operate these segments and drive efficiencies. This foundation supports their acquisition strategy. Management has continued to deploy capital in acquisitions at attractive prices, typically <8x EV/EBITA, then improve margins immediately afterwards.

The company is reaching an inflection point. Leadership is changing and reported earnings understate performance while overstating leverage. As Relais scales their M&A function, becomes a dominant player in the Nordics, and holds several strong platforms to build from, they are well positioned for future growth.

With the context established, let’s consider what lies ahead.

Outlook

Strategy

Relais has the foundation to continue their M&A strategy; the scope of acquisition candidates are numerous and Relais has put more focus on developing their M&A function. In March they restructured the management team to speed up decision making. The Strands MD now also heads the Products business area, the Scandinavia MD oversees Technical Wholesale, and Raskone MD oversees Repair and Maintenance. Ideally, it will empower MDs to take greater ownership over M&A and strategy for each area.

The new CEO, Christian Gebauer has already proven he can execute M&A at scale. As head of Construction & Infrastructure at PE firm Ratos, he managed EUR 1.4 billion in sales and 7,900 employees — roughly 4x Relais’ current scale. He grew the business area EBITA by 3.5x from EUR 36 million to EUR 127 million in 5-years, achieving a 19% ROCE in FY24. During his tenure, we estimate most acquisitions were conducted under 10x EV/EBITA on businesses growing between 8-20%.1

Gebauer’s incentives are also clearly aligned with Relais. He now has the chance to steer the entire ship, in this case a public company with strong historical performance. As a sign of alignment, he already purchased EUR 160k of shares in the open market.

With Relais growing their M&A efforts, I believe the funnel above can be scaled further. While 550 companies are actively monitored, there’s likely many potential targets they haven’t identified. This happens to even the best acquirers, Constellation Software has said they were unaware of roughly 70% of transactions that occurred in their space.

By Relais scaling M&A headcount and focus, they can maintain current relationships better and expand their database. Decentralization and headcount will also improve the bottom of the funnel. Each business area can engage in their own M&A discussions rather than relying on the head office.

The funnel will also scale naturally as they become a known market player. Relais has consistently acquired companies, made them better, and built upon them. Strong managers are retained and given more responsibility. If you spent your life building a small business, Relais is a good company to consider passing it off to.

Market

The Nordic commercial aftermarket is somewhat hard to gauge due to limited data and research on the commercial segment, but Europe level data and management provided numbers help provide some insight.

Relais’ current markets contain at least 1.9 million commercial vehicles with an average vehicle age >11 years, and 18 million passenger vehicles. Expanding into Denmark would add another 450k commercial vehicles, and the rest of the EU holds another 38 million.2 Analyzing estimated km driven, annual vehicle costs, and commercial vehicles by type, Relais’ claim of a EUR 20 billion market seems roughly correct.

In coming years, BCG expects fleet vehicles to become a larger share of total commercial vehicles:

We view higher fleet share of commercial vehicles as a positive trend for Relais. BCG estimates that fleet share by 2030 will grow to 41% of light-commercial vehicles (LCVs), 69% of MHDTs, and 79% of trailers.3 Even more than most commercial vehicle operators, fleet owners care about uptime, efficiency and price. Over 85% of fleets in Europe said price is a key criteria when selecting a workshop. As detailed in part one, Relais is strongly positioned to provide a better price, range of services, and coverage than most OEMs and IAMs in the Nordics.

While fleets may use OEM workshops under the warranty period, this is not a concern to Relais. Most of a vehicle’s lifecycle isn’t warrantied. Heavy commercial vehicle warranties generally only last 1-3 years and/or under 200,000 km depending on the part and vehicle model.4 As a result, over 80% of the vehicle’s useful life isn’t covered. Routine maintenance and wear and tear often isn’t covered during the warranty period as well.

While some suppliers offer extended warranty options, it rarely makes economic sense for the customer, as it costs more than IAM options and limits the workshops available. Moreover, Relais’ workshops maintain several OEM partnerships so they can presumably service the warranty portion as well.

Given short warrany periods and OEM partnerships, most commercial vehicles can be serviced at Relais shops using Relais-supplied parts.

As mentioned, Relais estimates the Nordic addressable market at EUR 20 billion compared to EUR 357 million of revenue, implying <2% market share. Specifically on the repair and maintenance side, we estimate the Nordic market at roughly EUR 8-11 billion, giving Relais only a 3% market-share in that segment compared to spare parts and workshop revenue.

Their addressable market would significantly grow if they expanded into adjacent industries or geographies.

Risks

Many of the risks were already mentioned in part one, but to quickly highlight them:

• Leadership Transition - Christian Gebauer has a strong track record, but will that carry over into his performance at Relais?

• Relatively High Leverage - Though lower than reported numbers would suggest, leverage is a concern for liquidity and capital for future acquisitions.

• Passenger Parts Wholesalers Entering Commercial - A relatively limited concern so far. The recent entrant, Meko, doesn’t have a strong track record of growth.

• Higher Competition in Broader Europe - Probably one of the largest concerns if you are considering Relais’ potential to enter other European markets. Many consolidators are present in the non-Nordic European countries and at a much larger scale. However, even in those countries, most seem to focus on the passenger market more than heavy commercial.

• Price/Service Competition - Primarily related to their parts wholesalers, the least glamorous of the business areas. It is important to consider trends related to Relais’ SKU count, price levels, ease of ordering, and frequency of delivery. The concern seems smaller than in the passenger spare parts space, but it’s important to monitor.

With these risks in mind, let’s discuss Relais’ financials and valuation.

Financial Considerations

Our estimate of real free cash flow required several steps, as is the case with many serial acquirers. We estimated true EBITA using management comments on organic growth and the full-year impact of recent acquisitions, then applied historical EBITA-to-FCF conversion rates (which averaged 84%). This yielded ~EUR 49 million of EBITA and ~EUR 40 million of free cash flow, which may still understate true earnings power.

Free cash flow was calculated after lease interest expenses and repayments, but before debt expenses. Intrinsic value will be compared to enterprise value without lease liabilities. The EUR 50 million hybrid bond was added to enterprise value as it is essentially debt.

Other considerations:

When evaluating Relais’ future potential, it’s important to be aware of the industry dynamics they operate within. Relais turns over their inventory two times per year and NWC four times, a drag on performance for an acquirer. Unlike asset-light industries like software, Relais must reinvest working capital to sustain growth. As a consequence, the speed at which they can grow is lower.

Valuation

Firstly, let’s consider what the market is pricing in using a reverse DCF. While I don’t personally use WACC, it’s applicable when gauging market expectations.

WACC is roughly 7.5% for Relais.5

Based on our five-year estimates, the market expects -3% annual growth and a terminal growth of 0%. Essentially free cash flow would shrink 14% then lag inflation for eternity.

Using our preferred discount rate of 11%, market expectations are still low at 4% annual growth for the next five years, then 1.5% into perpetuity.

I believe the rDCF is most useful for understanding what the company must do to achieve your desired rate of return.

However, for illustrative purposes, see below for tables potential returns and discounts to intrinsic value.

In terms of growth estimates:

Real return on capital metrics are hard to determine with a serial acquirer as the capital structure almost never reflects the true earnings of the businesses, so the best we will get is a rough estimate.

Based on purchase multiples, immediate ROIC from acquisitions ranges between 12-20% before accounting for margin growth and synergies. A rough average of 15% could be estimated. Traditional ROIC calculations using invested capital and NOPAT produce similar results of ~15%.6

Their historical FCF dividend payout ratio was ~30%, producing an estimated ~10-11% future annual growth from earnings deployed into acquisitions. Organic growth could be expected to add a few percentage points. It has been volatile but averaged mid-single digits since at least 2021 (~4%). Organic growth partly reflects margin expansion at recently acquired companies; other sources include market growth, increased market share, price increases, and cross-selling.7 Combining the two, annual growth could be expected to average 12-15%.

The tables below use only five-year growth of free cash flow due to the size of the company and uncertainty of the future.

If growth is sustained at double digit levels, at least a moderate re-rating of the multiple could be expected.

Using terminal value, you can see the company is trading at a significant discount at various growth rates:

I will leave it to you to conduct your own valuation work as well and draw your own conclusions.

Let me know if this analysis was helpful and what topics you’d like to see covered in future deep dives.

Thank you for reading my work; every subscription, like, and comment helps me continue to produce these reports.

Join Investor's Odyssey

Share

Read our post on multi-bagger Sterling Infrastructure and latest update: STRL History, 3Q25 Update.

Understand my approach to portfolio construction: Types of Companies I Invest In

Read full disclaimer here.

Consider joining to support my work and make this research sustainable.

Join Investor's Odyssey

Hi all, I’m excited to share this research report with you, it’s been a culmination of countless hours of research and writing.

This two part deep-dive will detail a Finnish serial acquirer in the automotive repair market. In the two parts, you will find information on:

Contents:

1. [Part 1] Executive Summary

2. Risks and Opportunities

3. The Aftermarket Parts and Repair Industry

4. Business Segments

   1. Vehicle Repair and Maintenance

   2. Technical and Wholesale Products

5. Competitive Advantages

6. Compensation and Ownership

7. [Part 2] Outlook

8. Financial Analysis

9. Valuation

Relais Group is the largest independent operator of commercial-vehicle workshops in the Nordics and a growing parts wholesaler/e-commerce platform.

Since their IPO six years ago, Relais Group stock achieved an Annual Total Return of 17.4% .1 Through organic growth and serial M&A, more than 20 companies have been acquired and Revenue and Free Cash Flow compounded at greater than 27% from 2018.

Executive Summary

Relais Group operates in two complementary, low-cyclicality businesses: Commercial Vehicle Repair and Maintenance and Technical and Wholesale Products.

I believe they face limited competition and strong growth prospects in the future, combined with a cheap valuation. Relatively low competition is present due to the slow growing markets, lack of attractiveness compared to the passenger vehicle market, barriers to success, and low greenfield opportunity for workshops and wholesalers. Suffice it to say, this isn’t a hyper-attractive industry. This market is far from the first choice for an established consolidator or PE firm to enter.

While one could argue there’s a low barrier to entry, the knowledge needed for success and strong return on investment is high. Relais has already built the know-how to get acquired companies fit fast and holds a leading position in commercial workshops across the Nordics.

In the future, most of Relais’ growth will come from acquisitions, with M&A becoming a core function since Relais Investment was formed in 2010. Limited organic growth is available through margin growth, new workshops, and expansion of their specialty products distribution.

Relais has proven scale advantages, improving margins on repair shops through operational efficiencies and private-label products. On the parts side, increasing private-label brands and bulk ordering can be expected to improve margin and wallet-share.

Relais has a long runway ahead of them. Management estimates a market of EUR 20 billion in the Nordics alone, compared to Relais’ LTM revenue of EUR 357 million. Although the underlying markets aren’t growing significantly, they also aren’t very cyclical. Over 90% of Relais’ revenue comes from the commercial segment, customers who need to keep vehicles running and want to extend the vehicle lifetime.

The independent aftermarket market (IAM) provides great value to companies, increasing vehicle uptime by ten hours per year, saving Europe’s commercial vehicle fleet an estimated 230 million hours per year.2 It is a critical industry that is not going anywhere.

So why are they reaching an inflection point?

Risks and Opportunities

The CEO Arni Ekholm had announced he will retire by March 2026. The board has appointed Christian Gebauer, an external candidate who is already transitioning into the role and starting no later than January 2026. He has previous experience with decentralized M&A and a long runway at only 45 years old. Although he has a strong track record, we have yet to see how he will approach capital allocation, leverage, and building out Relais’ M&A function further. One concern is that he will be based in Sweden whereas the head office is in Finland, a concern lessened by the fact Relais is a decentralized organization, only a few of the executives are based in Helsinki anyways. It was also just announced that he purchased Euro 160k worth of shares on the open market, a small but positive sign.

Several transformational acquisitions have occurred in the last twelve months, which could drive organic and acquisitive growth for years.

M&A frequency could increase with Relais’ recent focus on scaling their M&A team. In March, Relais restructured upper-management so operating managers are more involved in M&A and can make faster decisions, growing a decentralized model vaguely reminiscent of Constellation Software. The scale Relais has achieved also gives them greater credibility when pursuing acquisitions and financing, as evidenced by a 3x oversubscription in the hybrid bond issuance.

Leverage is high due to the transformational acquisitions in 2025, but reported numbers are misleading. At face value, Relais now holds Net Debt of EUR 200 million vs EUR 35 million of EBITA, (5.7x Net Debt/EBITA). Even after paying down debt with the EUR 50 million hybrid bond recently issued, debt/assets are 40% and reported EBITA/interest dropped to 3.9x. Leverage has undoubtedly contributed to returns as they consistently pay out at least 50% of reported EPS as dividends while debt remains high.

But, with serial acquirers you can’t take reported numbers at face value.

In our opinion (and management’s), to correctly evaluate their situation a few adjustments should be made. Leases should be excluded from net debt since lease expenses are already recognized in operating expenses. Furthermore, LTM earnings were impacted by several non-cash accounting charges related to recent acquisitions. Adjusting for this and the full-year EBITA gained from acquisitions, metrics are much stronger. Our estimate can be seen below, with Net Debt/EBITA of 2.1x (or 3.2x including the hybrid bond) and interest coverage of 5.4x.

Serial acquirers are continuously misunderstood because reported numbers don’t tell the true story. I believe this partially contributes to Relais’ low valuation.

However, leverage remains a concern as management said it’s too high to fund any large acquisitions in the near term without using equity.

Short-term acquisition effects dampening earnings also make growth, margins, and ROIC appear worse. Acquired assets and liabilities are fully recognized immediately, but it takes four quarters for the acquisitions to fully appear in earnings, and by then there are new acquisitions dampening metrics. Earnings are impacted by non-cash charges like acquisition cost allocations and amortization of put-call options. Thus, every metric appears worse than they are, which has been exacerbated by their high working capital in the past year. If you consider all the non-operational and one-time charges plus a full-year of earnings, growth will be quite strong and earnings much higher than financial data websites display.

Lastly, low trading volumes is a risk as with any micro-cap. Over the past year, average trading volumes were roughly 18k shares per day, excluding the eight largest trading days, the volume averaged 6.9k shares. In Relais’ 3Q25 report, they stated 3.6 million shares were traded between January to September, roughly 19.5% of total shares. This was a significant increase from the prior year period of 523k shares traded, only 2.9% of outstanding shares. While low trading volumes are a risk for entering and exiting a position, since we are long-term investors and expect the company to grow strongly, the concern is not as significant. Also, volume should continue to improve over time.

I’m cautious on the company’s prospects due to the size, leverage and leadership transition, but their growth potential and valuation are good reasons to be interested.

This two part deep-dive on Relais Group will cover:

Part one:

1. Aftermarket Parts and Repair Industry

2. Business Segments

3. Competitive Advantages

4. Compensation and Ownership

Part two:

1. Outlook

2. Financial Analysis

3. Valuation

Aftermarket Parts and Repair Industry

The market Relais operates in is large and stable, with opportunity for future growth and consolidation. Independent aftermarket companies, have also grown slightly faster than the overall industry. As of 2023, the EU independent aftermarket parts (IAM) segment was estimated at >EUR 73 billion (including passenger vehicles).

Relais primarily targets commercial vehicles in the Nordics, but has started to expand further into Europe. Management estimates the Nordic aftermarket size at EUR 20 billion, highlighted in blue below.

Relais’ focus on the aftermarket encompasses everything that goes into improving and maintaining a commercial vehicle after it is bought. Including spare parts, specialized equipment like lighting, workshops to repair and upkeep vehicles, and machinery for the trucks and repair shops.

You may recognize the value in the European Spare Parts market from popular stocks like Auto Partner SA and their outstanding growth. But it is key to distinguish the market Relais operates in from those of companies like Auto Partner and InterCars. These companies make almost all their revenue in the passenger vehicle business with some light commercial vehicle parts. Parts ordering is more random and last minute based on what part needs to be replaced. In the passenger market, there’s room for differentiation in same-day delivery and ease of ordering.

In contrast, roughly 90% of Relais’s revenue comes from commercial vehicles, much being heavy commercial (think semi-trucks and buses) and trailers. A fundamentally different market where sourcing is less volatile. So, Relais’ wholesale business is not at immediate disruption risk from players like Auto Partner or InterCars, as those companies have more growth to pursue in the passenger segment.

The virtue of the commercial market is the criticality vehicle fleets play to the operation of companies. Each hour and day that a semi-truck is down has ramifications across critical supply chains like food, healthcare, and discretionary goods. Consequently, companies ensure their vehicles are routinely maintained and don’t defer taking a vehicle in if it’s not working well or breaks down. Some cyclicality could theoretically occur if an overall economic slowdown results in less goods being purchased and thus less being transported, but that is a miniscule concern in my estimate. As mentioned earlier, the IAM market saves European fleet vehicles 230 million hours per year, a mission-critical business that isn’t going anywhere, which is why it’s a great platform for consolidation. As the market has grown, so has the age of vehicles, reaching record highs throughout the EU, meaning IAM products and services are becoming ever more critical.3

Some readers may be thinking about potential disruption from EVs, but this is another reason why the commercial market is a comfortable place to be. The path to electrification for commercial vehicles is much slower on the development and adoption side. While prominent companies like Volvo and Tesla have developed electric semis, it doesn’t work well for long-haul trips and is a large capital expenditure many companies can’t or won’t spend. Furthermore, even if companies do add electric semis, that will be in addition to their existing gas fleet rather than to immediately replace them.

The IAM market is so important and useful because it saves companies money, especially in countries with less wealth where they’re trying to stretch the lifespan of the vehicle, immediately purchasing a fleet of electric vehicles is the opposite.

Besides, many of Relais’s companies have already grown to service alternative vehicles like electric and hydrogen, capturing more value in the market and preparing for the future.

Business Segments

Vehicle Repair and Maintenance

Vehicle repair and maintenance is driven by physical workshops, earning money from vehicle work and the spare parts installed. Workshops often have a few dozen employees and a myriad of specialty equipment to service every part of a commercial vehicle. Since Relais purchases existing workshops, the additional capex investments are minimal, though they occasionally engage in greenfields as mentioned earlier, but land and machinery are often financed through leases. Most costs are operational — leases, machinery, and labor — meaning gross margins are higher but operating margins often lower compared to their parts business.

As a result of Relais’ acquisitions across the Nordics, they operate 62 commercial vehicle workshops, making them the largest independent operator in the Nordics. Being the largest operator with only 62 workshops indicates how fragmented the market is, particularly in non-dealer owned workshops.

Most large-scale competition comes from dealer networks such as Scania, Volvo, and affiliate dealerships. But IAM shops are a better offering for most customers. Relais can provide many of the same services as OEM shops but a greater variety of parts, including private label brands. Since IAM shops service a variety of vehicle makes, models and types, utilization improves margins and prices for customers. Expertise also isn’t a concern as Relais’ workshops partner with OEMs like DAF, Mercedes, and Iveco so workers are trained in servicing numerous brands.

Covering more vehicle models and types strengthens the workshop’s brand awareness, which continues to grow as they add more workshops to the network. Independent shops also provide a wider range of services including specialized repair, maintenance, installation, and the ability to bundle trailer and chassis work.

See the link for an example of the speciality services offered by their workshops. The video shows STS (now Team Verkstad) installing solar panels atop a truck to enable usage of vehicle electronics without the engine running.

Low competitiveness is present due to high fragmentation of independent shops and difference in incentives compared to OEM shops. For OEMs, repair and maintenance primarily serves vehicles under warranty and those who want the simplicity of going to a OEM, but it isn’t the OEM’s core business. OEM’s core business revolves around the manufacturing and selling of commercial vehicles, while warranties and maintenance contracts tie into that they don’t seem to be a driver for intense growth and competition. For the reasons mentioned above, there’s also only so much they can compete on. The service offerings and customer set is different. Furthermore, many OEMs choose to partner with independent workshops as it alleviates some of the capex and further operational complexity needed. Operating an independent workshop is incredibly more capital efficient and profitable than six different OEM shops sitting next to each other in the middle of nowhere. It’s unrealistic for several OEMs to have a dense enough workshop network to cover customers whenever and wherever their vehicle goes down while maintaining high utilization and profitability levels.

The largest OEM workshop network in Relais’ market seems to be Scania, a premium focused brand who emphasizes workshops as part of their value offering, but not a large growth engine. Their service business also isn’t growing fast, with 3% growth in FY24, meaning they aren’t pursuing high growth.

In the workshop segment, Relais’ notable acquisitions are Team Verksted, Raskone, and recent small bolt-on of Wetteri Auto:

Team Verksted:

• Acquisition announced in April 2025, the company was previously held by VyBuss, subsidiary of the Norwegian Ministry of Transport

• Consisted of 21 owned and 3 franchised workshops in Norway and spare parts dealer Lastvagnsdelar

• Acquired for a EUR 34 million enterprise value, EUR 20 million equity value

• FY24 Sales of EUR 77 million and EBITA 6.3 million, 8.1% EBITA margin.

• Relais estimates a EUR 9 million EBITA contribution to annual EBITA, before synergies.

• Purchase Multiple: 0.44x Sales, 5.4x FY25 EBITA, 3.7x estimated EBITA contribution

  • Immediate ROI: 18.5%

  • Mgmt Est. EBITA ROI: 27%

Raskone:

• Acquired in January 2021

• Added 19 workshops in Finland and Raskone CEO Jan Popov to management team

• Helped initiate Relais’ entrance into repair and maintenance segment, becoming a leading workshop chain in Finland

• FY20 Sales of EUR 63 million and EBITDA EUR 4.7 million, 7.5% EBITDA margin

• Enterprise Value: EUR 30.7 million

• Purchase Multiple: 0.5x Sales, 6.5x EBITDA

  • Immediate ROI: 15.3%

Wetteri Auto Oy:

• Announced in July 2025

• Two workshops in Finland, providing maintenance to heavy commercial vehicles including Scania, Mercedes, Man, Sisu, and Mitsubishi

  • Illustrates the scope of OEMs serviced by IAM shops

  • Added to Raskone operating company

• EV purchase price of EUR 14.1 million

• FY24 Revenue EUR 10.3 million and Operating Profit EUR 2 million, 20% operating margin

• Purchase Multiple: 1.37x Revenue, 7x Operating Profit

  • Immediate ROI: 14.2%

Average acquisition multiples indicate an immediate return on investment of 15% or greater. These multiples are in-line with previous management communications:

We try to get average somewhere between 6-7x EBITA, for smaller companies it can be 4-5, and for rare situations where there’s an imminent synergy potential or huge growth potential it can be north of 7.

- Arni Ekholm on Q3 2024 earnings call

Technical & Wholesale Products

Technical & Wholesale Products is a more fragmented in its strategy and offering and where the origins of the company lie. The segment can be further divided into Spare Parts and Lighting and Equipment.

Lighting and Equipment is more specialized and diverse, providing products with a strong brand and position and contributing 36% of YTD sales. Lighting has been a main differentiator, providing a growth engine and prominent brands. Strands is a core pillar of the strategy, designing and selling quality lighting equipment for commercial vehicles in the U.S. and Europe. Equipment consists of workshop machinery such as Auto Materiell, and truck equipment like Awimex. Relais’ strategy has created a source of differentiation as house brands can be sold across company owned retailers and geographies. For instance, the recently acquired Matro Group sells and prominently displays Strands and Matro’s brand Nedking. Premium operating companies like Strands, Matro, and Lumise also have a strong e-commerce offering, with modern webpages making online ordering easy.

Spare parts consists of parts wholesalers supplying auto part dealers, retailers, and car importers, with SKU counts reaching over 90,000 for some of the operating companies. A concern with this business area is the seeming lack of differentiation. Presumably there is some relationship power, but the business runs on prices. Customers may purchase more from a competitor if they offer lower prices or better service, but you won’t see constant switching of suppliers as that is unsustainable in the commercial market. So far, the majority of disruption for IAM has been on the passenger car side, with commercial being a much steadier market. That may change with prominent supplier MEKO expanding into heavy commercial, though Relais hasn’t come across them yet.

Although MEKO is large, they haven’t been the most successful on the passenger side. MEKO hasn’t grown EBIT in the past 5 and 10 years and ROIC was 3.5% in the LTM. So, they don’t spark the most concern. An argument can also be made that more independent supplier competition will expand independent supplier market share versus OEMs, as has occurred with passenger aftermarket parts and repair.

This segment is also less exposed for Relais than at first glance, with only a quarter of sales coming from their four wholesalers, spread out over multiple geographies and hundreds of thousands of SKUs.

For Relais, acquiring more parts distributors and developing house brands has also improved Relais’ differentiation and price levels for this segment. Continuous acquisition of parts wholesalers, manufacturers, and original brands gives the company better purchasing power, higher margins, and a larger customer set to distribute private label parts under. All-in-all it gives them greater leverage and knowledge than individual parts distributors and those developing just one private label brand.

Relais management has commented on the future potential to capitalize on strong brands in both operating segments:

From a cross-selling perspective, I think the untapped potential, the biggest ones are cross-sales of commercial vehicle spare part products from Sweden to Norway because we have a wholesale operation in Sweden, which has a very large scale of products.

Also, the rollout of lighting products through the workshops and through Nedking Europe operating in Europe. Best practices we are sharing every day.

- Arni Ekholm Q3 2025 Earnings Call

Prominent acquisitions in Lighting and Equipment are Strands and Matro Group:

Strands:

• Completed January 2021

• FY20 Sales SEK 150 million EBITDA SEK 25 million, 16.7% margin

• 55% of Sales came from exports

• Purchase price SEK 182 million

• Multiple: 1.2x Sales, 7.3x EBITDA

  • Immediate ROI: 13.7%

• Strands has become a core pillar of the portfolio and growth strategy

Matro Group:

• Announced April 2025

• Acquired 70% of shares with a 5-year put and call option for remaining 30%

• FY24 Sales of EUR 18.7 million, EBIT of EUR 3.6 million, 19.3% margin

• EV: 27.9 million

• Purchase Multiple: 1.5x Sales, 7.8x EBIT

  • Immediate ROI: 12.8%

• Long-standing relationship as a customer of Strands, potential for greater collaboration and cost-savings

As we can see, the e-commerce acquisitions have stronger margins and garner slightly higher multiples. ROI of all acquisitions are substantially above ten percent before even considering margin expansion, cross-selling, and organic growth.

Competitive Advantages

Competitive advantages have already been weaved throughout the write-up, but I will lay them out clearly here.

Purchasing Power

Scale across SKUs, product types, and regions gives Relais stronger purchasing power and better payment terms. This means lower prices on products which they can sell at lower prices or gain greater margin on, it also improves their working capital as payment terms improve with scale. Better purchasing power also puts them in a strong position if a price war ever occurs.

Brand Power

Brands like Strands provide greater differentiation and pricing power. These aren’t commodity products, they’re specially developed and don’t have a direct replacement. The more companies Relais acquires, the greater all brand names can spread. Operating company STS Trailer Service was rebranded to Team Verkstad after the latter company’s acquisition. Team Verkstad was already a widely known name, being one of the largest Nordic repair chains, rebranding STS allowed them to further capitalize on this and expand STS beyond just trailer service.

Private Label Brands

Private label brand development has become increasingly important in the commercial IAM ecosystem. Aging vehicles and strained budgets in many European countries means companies want to keep vehicles on the road reliably at the best price possible. However, they won’t just go with some random upstart brand they don’t trust. Valuable brands that have been around for decades like Startax and Nedking provide a reliable source for lower cost products.

The advantage is furthered by Relais’ direct connection to repair workshops, being able to cross-market their products and sell them for even cheaper prices/better margins. It also widens the scope of where the brands are known. E.g. they can introduce parts from their Swedish companies to the Norway market.

Decentralization

Management places a heavy emphasis on their decentralized structure and an entrepreneurial culture, spending less time and money on acquisition integration and letting great leaders continue doing what they do best. Similar to other decentralized acquirers, they have operating group managers and key brands that head certain segments.

The level of decentralization depends on the region, scale, and services offered. If they acquire a prominent company/new market, that can serve as a platform for future growth and brand identity through bolt-ons. Team Verkstad has been acquiring individual workshops under the brand across Sweden and Raskone doing the same in Finland. On the lighting side, you may see small companies join the oversight of Strands so they can share best practices. Subsidiaries still remain individual entities such as Matro, Strands, and Lumise all maintaining their own brands and websites, but collaborating behind the scenes.

Successful serial acquirers have proven that decentralization improves the ability to scale while still sharing learnings about different markets and how to operate best.

Operational Efficiencies

We have a very detailed playbook on when we buy repair and maintenance workshops, kind of a hundred point checklist of what to fix.

- CEO Arni Ekholm on Q3 2025 Earnings Call

Efficiencies are a product of the other advantages created by being a growing player in the IAM space. Everything from purchasing power and payment terms to best practices shared by company leaders across geographies and business types. Relais has doubled the profitability of many of the workshop brands, stating the ability to push EBITA from 5% to 10%. Theoretically, this means a 15% ROI acquisition turns into a 30% ROI. Efficiency is driven through better working capital policies, improving capacity utilization, and dozens of other small initiatives.

For instance, the addition of Lumise in 2021 provided expertise on e-commerce and business analytics and gave them a direct connection to the end consumer.

Through Relais’ acquisitions they have vertically and horizontally integrated across the aftermarket parts ecosystem. Relais now develops private label parts, supplies spare parts as a wholesaler, sells them DTC through e-commerce, and are the ones actually installing the parts and equipment on commercial vehicles. Having intimate knowledge and data across the value chain provides insight into how each of these parties operate and how to best serve each party.

M&A Centered Growth

Growth for Relais is primarily driven by consolidation rather than trying to take market share, an important distinction for slow growing markets. High competition for organic growth can be very costly and result in a pricing war. This is especially true when management teams feel pressured to maintain double-digit growth. Instead, M&A centered growth relies on owning more of the ecosystem rather than taking it from others. Higher than average organic growth becomes a product of better knowledge and scope rather than competing solely on price.

M&A means they invest in already established brands and products, not risking heavy capital expenditures on unknown outcomes. They can find new areas of the market to add to their portfolio and expand geographically. IAM is the perfect market for this due to a lack of cyclicality and a heavily fragmented market meaning numerous opportunities and less sophisticated competition.

Compensation and Ownership

Depending on how you look at it, the ownership and compensation structure can be a strength or a risk. As of November 2025, the board and management held 32% of outstanding shares. The Chairman, Jesper Otterbeck has 6.5 years of tenure on the board and holds 16% of outstanding shares. Most managers hold at least EUR 1 million worth of shares, though a couple including the CFO don’t own much, which seems odd at first. Lower ownership is less strange when you consider that some executives are relatively new and salaries are generally low, meaning they don’t have excesses of money to purchase stock. Stock-based compensation is also more selectively awarded compared to American companies.

Those with significant shareholdings are generally from acquired businesses such as Juan Garcia from ABR and Jan Popov from Raskone. Managers without shareholdings in the image below do have stock options however, which is a less popular share compensation method in the US nowadays.

In 2024 CEO Arni Ekholm earned EUR 310k in salary, of which EUR 47k was performance based cash bonus. Short term incentives are based on annual financial and strategic goals, with the incentive reaching up to 50% of base salary.

There are also three outstanding option plans totaling 382,000 shares, which would have a market value of EUR 6.1 million if converted to shares. Below you can see the stock option plans for Finnish (A) and Swedish (B) employees, with 2025 being announced a couple days ago.

There was also a LTI incentive plan previously approved consisting of 258k of cash-based incentive units (of which 172k are remaining), this will likely be similar for the new CEO.

I am not a fan of the relatively unclear compensation structure for other managers and the use of stock options. The compensation policies could be better, but currently isn’t a significant enough of a detractor to turn me away from the company.

In terms of the board, they have strong experience and aligned incentives:

Jesper Otterbeck joined the board after ABR was acquired, a company that his investment firm owned while it scaled from 8 MEUR to over 400 MEUR.

Anders Borg was a KKR MD and Head of Nordics for 10 years. He holds 60,000 shares of Relais.

Katri Nygard has been on the board since 2015, serving as Chairman from 2015-2019. They worked at EY and KPMG and hold 106k shares and 383k options personally. They also oversee 2.5 million shares (13.42%) as the appointed board member for investment firm Ajanta Oy. Ari Salmivuori, the Chairman of Ajanta Oy, holds another 17.2% of shares, meaning roughly 44% of shares are controlled by management, board members, or affiliates.

Two previous insiders Ville Mikkonen and Jon Strand held another 557k of shares, 3% of shares outstanding.

The board holds a significant ownership stake and is quite strong in the areas needed to be a successful acquirer. Overall, this is a strong positive, but some may consider the dividend policy a detractor. The current policy is to pay out at least 30% of earnings as dividends, if the board members and their investment funds prioritize dividends over debt paydowns and acquisitions, it could be a drag on the company.

When looking at non U.S. companies it’s important to consider cultural norms. For Finnish public companies, the median payout ratio is ~54% compared to ~16% for the United States, so Relais’ policy is in no way surprising or excessive in the country’s context.4

Conclusion

Now that we understand the company and markets it operates in, part two will detail what the future looks like for Relais and what the market is pricing in.

Part Two:

1. Outlook

2. Financial Analysis

3. Valuation

Thank you for reading my work; every subscription, like, and comment helps me continue to produce these reports.

Join Investor's Odyssey

Share

Read our post on multi-bagger Sterling Infrastructure and latest update: STRL History, 3Q25 Update.

Understand my approach to portfolio construction: Types of Companies I Invest In

Read full disclaimer here

Refer to the Sterling Infrastructure write-up if you don’t know about the company.

Performance

Organic and acquisitive growth have both remained fantastic, mainly driven by data center, manufacturing, and e-commerce demand. Quarterly Revenue reached $689 million, 32% growth, with 58% E-Infrastructure growth (42% organically). The segment and overall company continue to perform well, adjusted operating margin grew for both E-Infrastructure and Transportation to 26.8% and 15.6%, respectively, and overall to 21.9%.

Expansion in gross and operating margin for E-Infrastructure is driven by projects becoming continuously larger. The more complex and lengthy the project, the more value Sterling squeezes out of it from overall bid price and efficiency gain potential. Management continues to see project sizes increase, thus potential for further margin expansion, as noted on the earnings call:

Our customers are talking to us about the next level of projects that are coming out, again, they’re getting bigger, which is fantastic for us.

Margin growth is also driven by acquisitions expanding their value capture on projects. Management noted the dry-utility tuck-in done in January has already seen 40% improvement in margins by integrating into site development jobs. It really goes to show, all the margin and efficiency to be captured by doing more phases of a project, proving the value in STRL’s M&A strategy.

On the transportation side, revenue grew 10% while operating income grew 40%, an outcome of their continuous bid-discipline and wind down of Texas Sterling’s low-bid heavy highway business. As projects in heavy highway are burned through (out of backlog by first half 2026), not only will Transportation margin improve from an already strong 15.6% for the quarter, return on capital will grow as they can redirect TSC’s resources to higher earning jobs.

Outlook

Prospects for the next few years are strong, but depending on how you look at it, you could see concentration and valuation risk.

Management gave several positive notes on the earnings call:

As we look out to 2026 and 2027, there remains a very big pool of mega projects on the horizon.

This would include planned semiconductor fabrication facilities. The e-commerce market has strengthened significantly in 2025.

Data centers and semiconductor fabs require years of planning, communicating with trusted contractors like Sterling well in advance, giving management a strong pulse on future demand. Future acquisitions, integration of CEC into future projects — which they’ve seen strong interest in — and geographical expansion into Texas provide a strong case for continued growth in revenue and margins.

Transportation solutions, a more “boring” business, holds a strong outlook as well. Management is continuing to engage in specialized higher margin projects like aviation and rail, with plenty of federal funding left. Two years of spending are still left on the current bill, with management anticipating having two years of backlog going into the end of the funding cycle. Furthermore, once the bill ends, it generally results in a “bridge” that maintains funding at similar levels adjusted for inflation. STRL has a team in D.C. who has a good pulse on future government spending, indicating infrastructure bill progress is strong, six months ahead versus previous cycles.

On the not so positive side:

E-Infrastructure is now $3 billion out of the $4 billion of backlog. Within the $3 billion, data centers are roughly 75-80%. Meaning roughly 60% ($2.4 billion) of the backlog is data centers alone. Such concentration leaves question on what the backlog will be replaced with when data center spending inevitably slows down. On the other hand, Sterling’s diversified model has shown a continued ability to find areas of high growth and margin.

Building solutions has hit what management thinks is rock bottom, but doesn’t believe an upturn will be seen until at least H2 2026.

Valuation continues to be high, trading at 34x TTM FCF and 30x adj. operating income. Such multiples were previously unthinkable for a construction business. But Sterling has also continued to defy conventional construction companies, with a ROIC of >30% and Operating Margins over 20%. Nonetheless, for a cyclical industry, it should spark concern.

Valuation

Adjusting operating cash flow for depreciation and SBC, TTM free cash flow was $350 million, a 34x EV/FCF multiple.

For a very bullish case, I estimated year one FCF growth of 35%, slowing by 5% each year to reach 15% in year five. This yields a 5-year FCF CAGR of 25%. (Compared to YTD operating income growth of 45% y/y1.)

Given the cyclicality of construction companies and unknown long-term future of high revenue and margin levels, I elected to use an exit multiple. Assuming the multiple contracts from 34x to 25x, the 5-year CAGR will be 17%; a contraction from 34x to 20x yields 12%:

However, if the FCF CAGR was 15% and multiple contracts to 20x — still strong growth and a high-multiple for construction — the return is only 3%.

If you had asked me a year or two ago, this would have been close to my base case. A more realistic assumption for a strong growth construction company.

The Verdict:

Current valuation is pricing in very strong performance, a stark contrast from when I first invested a few years ago. I believe STRL will continue to do well, management will continue to execute, and the end-market will grow. However, I would be hesitant to recommend at current prices. I have opined before on the risks of paying up in expectation of high continued growth, and would be foolish to recommend otherwise.

Due to the high valuation and their strong ties to the data center boom, I won’t be purchasing more at these levels. I will likely be selling part of my position, as the strong run-up has made it too large a part of my portfolio anyway.2

With that being said, I don’t wish to fully cut the grass and water the weeds, and management has continued to surprise on the upside. When I first invested in this company, the stock price was pricing in low single digit growth, and management blew that out of the water.

I believe the company is strongly positioned for the foreseeable future, so I will continue to hold some shares and wait for a better price.

Stay tuned — I’ll be posting a new pick with strong prospects and a low valuation this week, along with other earnings updates.

Read full disclaimer here

Introduction

Sterling Infrastructure (NASDAQ: STRL) is an infrastructure construction company based in the United States. Over the past 5 years:

• Revenue compounded 10% per year to >$2bn

• Diluted EPS grew 30% per year to $7/share1

• Share price increased an average 84% per year to $336/share as of October 2025

It’s the perfect example of a well-executed turnaround.

When current CEO Joe Cutillo joined in 2015, they were in a decade long slump. Share prices had declined 90% from a peak of $30 to $3 and the company was losing money. The previous CEO had resigned and the Chairman was running the company while they searched for a CEO.

Cutillo was appointed CEO in 2017; he moved to solidify current operations, phase-out low-margin business lines, and used M&A to expand operating capabilities. Sterling looks completely different from 10 years ago, and they’re not done yet.

As a product of this transformation, valuation now embeds lofty expectations. The key question is whether organic growth and disciplined M&A can sustain double-digit EPS growth.

Contents

This report aims to provide context into how Sterling has evolved, what the current business looks like, and risks they face going forward.

It will be detailed in the following sections:

• Company History

  • Beginnings (1955 - 2005)

  • Rise and Fall (2005 - 2015)

  • Cutillo Era (2016 - Present)

• Competitive Advantages

• Acquisitions

  • CEC Facilities

  • CCS Arizona

• Risk Analysis

  • Segment-Level

  • Business-Level

Beginnings (1955 - 2005)

The origins of Sterling are traced back to 1955 where brothers James and Richard Manning started the Oakhurst Company in Sterling Heights, Michigan. The family business grew, with their successor Patrick Manning aiding in relocation to Houston in 1978. They sought to capitalize on regional economic growth, aiding in heavy civil infrastructure such as sewers, water mains, and light-rail transit.

In 1991 the company merged with Steel City Products (SCPI), a consumer products distributor, to diversify revenue streams and capitalize on SCPI’s NOL carryforwards. They acquired another parts distributor, Dowling’s Fleet Service in 1994, and funded construction of a waste-to-energy facility in Illinois in 1998.

By the early 2000s Sterling was a public company. Patrick Manning was Chairman and CEO holding 4.7% of the common stock, the entire director/executive officer group held 42%. Sterling claimed to be one of the largest regional contractors in Houston, also operating in DFW and San Antonio. Nearly all of their work still came from publicly funded transportation projects, with only the occasional private contract. Management continued to pursue M&A as Sterling acquired Kinsel Heavy Highway in 2002 and $2 million worth of equipment from RDI in 2006.

The company’s attempt to enter the distribution and energy businesses illustrates management’s poor capital allocation and management skills. The energy facility project sustained significant losses and Sterling’s relationship with them concluded in 2001. Dowling’s Fleet Service was disposed in the year 2000 and SCPI was sold in 2006 after revenue contracted from $33 million in 1990 to $5 million in 2006.

Rise and Fall (2005 - 2015)

At this point, Sterling could be seen in a more positive light. Revenue had been growing, they were focusing on the core business and divesting non-core unsuccessful ventures. The successor of the founder was at the helm and it was a new century. The dot-com bubble had passed at this point and construction spend was going up.

By 2005 Sterling’s revenues were $219M and operating profit was $14.7M, a remarkable increase of 47% and 41% from 2-years prior, and a 6.7% operating margin.

Backlog increased 32% year-over year to $307M, state highway work grew 98% to $77M, and municipal revenue grew 52% to $142M. The growth was especially remarkable as 2004 revenues had declined 11% from 2003.

The company attributed the growth to higher backlog which allowed crew and fleet expansion, growing capabilities to handle large and complex work, an increase in their share of state highway contracts, and better weather (2004 was reportedly one of the wettest years on record for them).

At this point in time, the company operated in Houston, San Antonio, DFW, and Austin. The strategy continued to be on infrastructure growth in Texas markets bolstered by infrastructure spending bills that were passed. $14.5 billion was allocated to Texas for federal highways and there was a mandate to replace all well water systems in Houston with surface water systems. The company desired to integrate more services and products to capture more margin and also expand through the Southwest. Management stated a continued eye for M&A, citing their 2002 acquisition of Kinsel Heavy Highway. Around the time the 2005 annual report was released they were trading at roughly 20x earnings.

Five years later in 2010, revenue had continued to grow, reaching $459M and $35M operating profit, over 100% growth since 2005 across both metrics. But the share price in 2010 alone had declined from around $20/share to $13/share, and was lower than the $18/share from 2005.

So what caused the price decline and continued deterioration?

In addition to the GFC which caused a broad decline in share prices and spending, Sterling had substantially diluted shareholders through new share issuances.

In 2005 Diluted Shares Outstanding was 9.5 million, by 2010 this number reached 16.2 million. This was a shocking 70% share count growth in 5 years. It caused Diluted EPS to only grow 2% from 2005 to 2010 while revenue had grown 110% and net income 72%.

After this point, the financial profile began to deteriorate as well. Starting in 2011, Sterling’s operating cash flows dropped while capex grew, goodwill was impaired, and G&A ballooned. They didn’t record another year of net profit until 2017, after the turnaround. While they had some strong years of growth, the company couldn’t sustain it with current cash flows. Management got in a rut of seeking revenue growth, diluting shareholders to fund investments, and losing money in the process.

2

Referencing the MD&A section from this period, they attribute margin decline to downward revisions in expected revenues (and thus margins) on projects they had completed. As can be seen in 2013, this became so drastic that the gross margin was negative.

They listed out reasons including skilled worker shortages, onsite conditions different than described, lower than expected activity levels, and delays in identifying and addressing issues during the projects. In my opinion, a well run company would not continuously have all the issues stated here.

By January 2015, financial performance continued to lag; Peter MacKenna stepped down as CEO (he was appointed in mid-2013) and Chairman Paul Varello became interim CEO. He believed the Company could improve financial results and shareholder value, they started searching for someone to aid in these efforts.

“Our fourth quarter results are far from representative of what this business is capable of delivering. Since I became CEO, we have been engaged in an extensive review of all aspects of our Company including processes, procedures and personnel at all levels.”

- Varello’s commentary in the Q4 ‘15 Earnings Announcement

Along with the passage of a new federal transportation bill in 2015 and a greater focus on financial performance, the margins started to improve. Gross margin increased from 4.6% to 6.4% in 2016 and operating margin neared break-even.

Cutillo Era (2016 - Present)

After a decade of overexpansion, dilution, and margin erosion, Sterling entered 2016 searching for a new direction. This transition would mark the start of one of the most successful construction turnarounds in the last decade.

The next two sections briefly discuss how Cutillo’s strategy reshaped Sterling’s business mix, leading to the inflection in earnings and valuation.

When Cutillo joined, Sterling only had one operating segment, currently named Transportation Solutions. This is what the company had been doing for decades, government contracts for infrastructure like highways and bridges. The segment was incredibly low margin, highway work was undifferentiated and contractors won jobs primarily on price. With highway projects the engineering often isn’t fully fleshed out so it’s hard to accurately estimate material and labor costs. Sterling was completely at the whim of supply, if supply decreased, contractors would bid lower and lower as margins degraded away.

Noticing the lack of value and differentiation, the company expanded into Building Solutions. In 2017, Sterling acquired Tealstone, a Texas based company that built concrete foundations for commercial and residential homebuilding. They also became more selective about which transportation projects they bid for. At 14% operating margin versus 2.4% for transportation, it was immediately accretive, company-level operating margin reached 4% in 2018.

In 2019, the third operating segment was added, E-Infrastructure solutions. Plateau Excavation was acquired, a Georgia based company that specialized in site excavation, storm drainage, and infrastructure for data centers and distribution centers. Expanding into this segment has proved particularly beneficial due to its versatility. They could capitalize on e-commerce centers when in high demand, work on manufacturing onshoring as that has grown, and take part in the surge in data center construction.

Over the next six years, Sterling acquired six more companies to strengthen each of these segments and expand to new geographies. Now that these platforms have been established, management is focused on increasing the depth of their offerings. For instance, the 2023 acquisition of PPG allowed them to bundle underground plumbing work into their homebuilding slab construction, speeding up the process for customers and capturing more of the project value. Their largest and most recent acquisition, CEC Facilities will help them expand into electrical services.

Growing from just transportation projects in Texas, Sterling now completes projects across thirty states and plans to expand further.

Source: STRL 2024 10k

Competitive Advantages

There are a few reasons why they achieve uncharacteristically high returns for a construction company:

• Diversified operations = Lower cyclicality

  • Construction can be cyclical, Sterling’s exposure to different industries and funding sources helps level out financial performance. Maybe the housing market is soft one quarter, but data centers are booming. Sterling is project agnostic, so they don’t care where the job comes from, they can execute on it.

  • Typically project and weather fluctuations can impact profitability as you have idle workers. Sterling has been able to shift crews across projects and even segments when needed. This allows greater flexibility and higher utilization of equipment.

• Speed + quality over price

  • In the public sector, projects are often approved based on bid price. This causes companies to bid low and earn poor returns or even lose money. Sterling recognized this and has put more weight on private sector work and specialized public jobs such as aviation. This is one of the main reasons for margin expansion.

  • For example, a hyperscaler like Amazon values reliability and speed far more than marginal bid price. Each day a data center or distribution center runs over schedule ramps up the opportunity cost more than the higher bid price does. Private companies with an abundance of capital value contractors that can reliably execute on time or ahead of schedule. Sterling has built up a reputation for consistent execution across their subsidiaries.

  • With Sterling’s expansion of operating capabilities, this advantage has been magnified. Bundling more of the operations under them, Sterling can significantly speed up a project, there’s no waiting for a new company to come on site and get set up. This will boost Sterling’s value add and the margin they can obtain.

  • In the latest earnings call, Cutillo highlighted how their recent acquisition of CEC Facilities will improve their E-Infrastructure offering:

    [On data center projects] we actually have all the dry utilities in, dug, put in, concrete poured, finished before the electrician is even selected for the project. Think of how much time that can pick up, and think of how much productivity that can pick up.

    “By integrating these businesses together, we really believe we can take months out of the development of this project and months out of the total cycle time to build these”

    - Q2 ‘25 Earnings Call

• Decentralized but collaborative

  • Sterling is highly disciplined when they choose what companies to acquire. They want organizations with great leaders in place (often the founder), a strong culture, and a reputation for quality work. The companies are strong on their own and don’t need excessive effort by central management to perform well. Instead, time can be spent leveraging skillsets, cultivating relationships, and integrating new acquisitions into current operations.

  • Although they are decentralized, they have also demonstrated an ability to share learnings and collaborate on projects. As mentioned, crews and equipment can be shifted from one operating company to another, allowing them to accept a greater quantity of high-value jobs that come along. This improves utilization and the number of relationships they are able to build. Understandably, this also gives their employees more consistent work and could reduce turnover.

    Here’s an example of how they’ve leveraged multiple subsidiaries in the past:

    “Using our relationships and kind of skillsets around the project management of Plateau and teaming up with our RLW business in Utah -which is outstanding in executing projects - put that together we did our first Meta project.

    That project has been so successful that the teams they’re working with are now talking to them on two data centers within the Rocky Mountains.”

    - FY23 Earnings Call

• Earnings visibility

  • By nature of the industry and the direct relationships they maintain, there is visibility into future revenue and gross margins. Projects start to be planned years in advance, and prices are contractually agreed upon well before the start date. Clauses for building materials costs are often included in contracts, helping shield them from material cost inflation. Contractual revenue that hasn’t been completed is termed Backlog, and Combined Backlog includes unsigned awards, keeping track of these metrics can give some insight into future revenue growth.3 Sterling also reports gross margins for their backlog so you can ensure it is profitable pipeline growth.

  • In addition to contractual backlog, Sterling has become more involved in large multi-stage projects. Multi-stage projects aren’t approved all at once, a company has to execute on one stage then win the next one as well. Assuming they execute, which Sterling does, they generally win the following stages. This leads to even more earnings in the pipeline than their backlog metrics suggest.

  • Due to recent backlog growth and increase in multi-stage projects, Sterling is arguably in the best position they ever have been.

Source: STRL 2024 10k

Acquisitions

“We’re typically looking for businesses that are accretive day 1 … and have the potential to double within a 3 year time period.”

- STRL at the 2024 Southwest IDEAS Conference

Sterling’s acquisitions have ranged greatly in size and operations scope. The smallest was $7.6 million for Kimes & Stone in 2021 and largest is $505 million for CEC in late 2025. For this section I want to analyze a couple of the acquisition valuations and how they contribute to operations.

CEC Facilities

The CEC acquisition was announced in June 2025 (finalized in September) with $450m cash consideration and $55m in common stock, plus an unspecified earnout based on 2029 operating income targets. The Founder and Chairman Ray Waddell will lead Sterling’s electrical services growth and their CEO Daniel Williams will continue to lead the subsidiary’s operations.

CEC specializes in electrical services for “mission-critical” end markets like semiconductors, data centers, and manufacturing. Customers include Samsung, Intel, Texas Instruments, and Meta. They capture revenue not only from installation, but also ongoing maintenance, retrofitting, and upgrades. This adds a small recurring revenue component to Sterling and expands the scope of work they can do on a project.

Revenue Sources:

• >80% of sales are from “mission-critical” markets

• 50% of business is based in Texas

• 20% of backlog is related to existing facilities work

• 6% of 2024 revenue came from on-demand services, primarily in the semiconductor market

Sterling Provided CEC Financial Estimates:

• 2-year Revenue CAGR of 20%

• 2025 EBITDA margin of 13%

• FY2025 Revenues $390-415m and EBITDA $51-$54m

• $0.63 - $0.70 EPS accretion per diluted share

Valuation:

• $505m deal value (excluding earnout)

• 9.5x FY25 EBITDA, 1.25x Revenue

  • 8x 2026 EBITDA, assuming 20% growth

Sterling management noted the earn-out as “substantial” and requires CEC to nearly double by 2029. CEC stated they believe this can be achieved organically. Assuming they mean year-end 2029, CEC would compound EBITDA at 19% organically to roughly $105 million in 2029.

Management also stated plans to pursue tuck-in acquisitions over the next 12 months to build Sterling’s electrical and mechanical platform. This aligns with previous comments on the desire to build such a platform as their fourth segment.

Considering the projected revenue growth and how it will assist growth in Sterling’s operations, it’s a very good price. Additionally, the Founder and the CEO coming on-board and being incentivized with income targets out in 2029, shows they are aligned for long-term performance.

If executed well, this will be one of their most transformational acquisitions. The size of the company and scope of capabilities it adds to operations, will push Sterling forward. A company that has been able to grow quickly like CEC and establish a good market position will also have valuable learnings and expertise to contribute.

CCS Arizona

In December 2022, Sterling acquired CCS, a construction company focused on concrete foundations for homebuilding This was folded into their Tealstone business already based in Phoenix and served to expand their capabilities to service blue-chip homebuilders in the growing market.

Valuation:

• Purchase price of $21m, with $4.8m being stock

• .25x Revenue ($70m annual revenue)

• 2.5x-3.5x EBITDA, implying a roughly 10% EBITDA margin

While it’s hard to evaluate how much CCS has grown or contributed to building solutions, this seems to be a reasonable price especially for the size of business compared to their building segment revenue overall of $321 million in 2022.

The market has not performed well in recent years, but the purchase price makes it hard to think they got a bad deal.

Risk Analysis

Segment Level

It is no surprise that performance of their operating segments are based on the construction demand and capital outlays in those respective industries. For Transportation Solutions it relies primarily on state and federal funding, Building Solutions depends on the single and multi-family home market, and E-Infrastructure centers around onshoring of manufacturing, e-commerce growth, and data centers. Sterling is not the only company of its type to have seen large revenue and share price growth in recent years, but they are one of the highest performers. The main question lies in the confidence of whether these domestic investment trends will continue, and if Sterling’s operating segments are uncorrelated enough for the company to be somewhat protected from spending downturns.

There is also the difference between a spending slowdown and the pop of a bubble. In the segment specific discussions below I will reference US Census Bureau data on public transportation spend, private residential, and total manufacturing, covering the period surrounding the GFC. These will aid in giving historical comparisons of how long it took for spend to reach back to peak levels and how different industries were affected.

In terms of economic downturn risk such as a recession, if you are one who believes we are in a bubble (corporate investment, AI, whatever), then after the bubble pop, the economic downturn could last for a significant period.

In the GFC, aggregate private construction spend peaked in 2006 and didn’t reach the same levels until 2016, on a nominal basis. However, looking at just this figure is cloudy, and I find segment specific spending data to be clearer indicators.

The following sections will discuss risks for each operating segment, incorporating some Sterling level discussion and some macro-economic discussion.

Transportation Solutions

In recent years, Sterling’s Transportation Solutions has been supported by the substantial Infrastructure Investment and Jobs Act (IIJA) that runs from 2022-2026. It was authorized under the Biden administration, allocating $1.2 trillion for transportation and infrastructure, with $55bn being for “new” investments and programs. As of March 2025, $712bn was still available for disbursement, with over $300bn being for large transportation projects like roads and bridges. Although the spending bill is meant to range from 2022-2026, 35% of the available funding, $250bn does not have a fiscal year limitation, with $73bn being for the DOT.4

Sterling echoed a similar sentiment in their 2024 Annual Report, stating that transportation demand remains at very high levels as they move deeper into the funding cycle. This makes sense as the bill wasn’t signed until November 2021 and it takes time for funding to be allocated to federal departments and non-federal jurisdictions, then spent on specific projects.

While Sterling has been shifting resources to E-Infrastructure due to recent high-demand, the continued availability of funding for this segment shows that it is a reliable backup they can shift to if other segments slow down. It is important to note that would likely result in some margin decline for Sterling.

In terms of greater macro-economic trends. Transportation spend was last seen peaking in 2010, before reaching similar levels within four years. The growth in the mid-2000s began with higher tax revenues from homebuilding and a focus on transportation investment in metropolitan areas. The GFC resulted in the ARRA funding bill passing in 2009. It quickly injected tens of billions of dollars into infrastructure spend, with most being deployed by the end of 2010. After this, there is a noticeable drop off before a steady increase over the next few years. These historical fluctuations are incomparable to the current environment, where spend has been similarly high for several years. But, it is still illustrative of government spending patterns in response to government downturns and financial crises.

E-Infrastructure

After a slight decline in revenues in 2024, E-Infrastructure has seen a significant rebound, bolstered by data center projects underway and in the pipeline. This is reflected in Q2 ‘25 backlog numbers with data centers comprising 62% of segment backlog. E-commerce has also seen a rebound, backlog increasing almost 700% for the latest quarter.

On face value, you could argue that there is significant risk if/when data center manufacturing slows down. But, the E-Infrastructure segment doesn’t rely solely on data center demand, they have historically shown an ability to pivot as the market demands do too. They were able to pick up manufacturing onshoring jobs when Amazon slowed down distribution center construction, and as some of those slowed down, they have been able to take on data center jobs.

The main bearish case for this segment would be its economic cyclicality. Countrywide private (+government support of) investment in manufacturing, renewable energy, commerce, and data centers will not always be high. If we saw a bubble pop or an economic downturn, many companies would slow down construction as with previous downturns. I believe this is one of the most significant concerns, though out of our control, the inherent cyclicality of this line of business. With that being said, the country continues to grow and innovate in one way or another and there will continue to be high-growth high investment industries. Based on Sterling’s track record, they have demonstrated an ability to gauge future demand and find construction trends that will last years.

There wasn’t a US Census Bureau category that perfectly encapsulated the E-Infrastructure segment, but I found manufacturing and commercial to be close enough. While it’s hard to distill down country-wide and global spending trends, domestic manufacturing was going strong in the mid 2000s, substantially growing until the GFC froze credit markets and collapsed global demand for goods. This downturn led companies to postpone expansion projects as margins were likely contracting and their current capacity was enough to supply lower demand levels. The more recent spike starting in 2022 is in line with the manufacturing on-shoring trend, further bolstered by the AI boom and increasing demand for semi-conductors. It’s honestly shocking to see a near doubling of manufacturing construction spend from 2022 to 2024, and triple 2021 levels. Similar but less drastic growth can be seen with the commercial spend category. I am generally bullish on the recent long-term infrastructure investments in the US, but this is definitely an area to keep tabs on. We need to ensure this is a long-term trend rather than a government driven spending fad.

Building Solutions

I don’t think we need to go over the GFC talk track again with this segment as the causes and results are much less applicable to the current environment. It is interesting to note, however, that spending peaked in 2005 and didn’t reach similar levels until 2020 (though actually later if you adjust for inflation).

The Building Solutions segment has been the victim of a downturn recently. Likely a combination of high mortgage costs and a tight economic environment, people have been less willing to buy homes and the supply has grown. The housing inventory in the DFW area is back to the highest it has been in over ten years. Although there was a lot of discussion around long term housing shortages in recent years, the number of active listings has quickly rebounded since the trough in 2022:

At face value housing prices appear higher, but when adjusting for overall inflation, the median price as of August was lower than pre-covid. The higher listing count could be a combination of factors such as overproduction by homebuilders in the last few years and tightening wallets making the same price of home harder to afford. I don’t like to overly discuss the macro environment as it’s impossible to predict, but the supply trends are important to note for the upcoming future. This is one of the aspects that makes the housing market a harder one to invest in for the short-term and why it’s useful that Sterling has other segments to balance it out.

Business-Level Risks

Key Man Risk

By the amount of praise that has been put in this article on the turnaround executed by Cutillo, it is reasonable to assume there is some key man risk here. Comparing the previous state of the business to its current performance, there is concern if this can be continued when Cutillo eventually retires. Very few CEOs can achieve high ROICs, especially in construction, and even less are good at M&A as well.

The decentralization of the organization lowers the concern slightly, but their level of decentralization is nowhere near those of well-known companies like Berkshire Hathaway and Constellation Software. Cutillo has mentioned many times how they have coordinated shifting teams to different geographies and operating segments to satisfy demands. He also appears to be actively involved in large contract negotiations, or at least know of them in depth. At the same time, he has been successful in selecting great companies to acquire at fair prices and the operating managers seem to be competent.

The board has recognized this risk and implemented special compensation plans to ensure he is retained and well rewarded. Cutillo’s base salary sits at $1 million, which is just 18% of his total compensation. Short term cash incentives are 20% of total target compensation, based 75% on Adjusted EBITDA targets and 25% on safety performance. The bulk of his pay comes from PSUs and RSUs, 60% of target comp. Of the PSUs, 67% is based on EPS targets which ensures that earnings grows without diluting shareholders, a problem STRL previously faced. The other 33% is based on relative Total Shareholder Return of the public shares versus comparable companies. The total target value of compensation for Cutillo was $5.65 million in 2024, but has generally been higher due to beating performance targets.

In addition, the board approved a special grant plan in 2024 where based on certain stock price targets he would receive 160k PSUs at the end of 2026. If he agrees to extend his employment through 2027, he will be granted another 40k PSUs since he already achieved all three stock price metrics.

I don’t completely agree with the board’s compensation plan, the performance targets seem to be low as performance often reaches 200% of the target. Additionally, there is more of a focus on stock price than I would like. Even though the special grants incentivized Cutillo to remain until at least the end of 2026, it seems the stock price targets were just a veiled way to ensure he receives substantial grants as long as he stays at the company. The targets were low compared to the near term future performance of the company, with the first two being reached by March and May 2024 ($100/share and $120/share) and the last by September ($140/share). Using share price as the target also incentivized him to increase the stock price rather than the underlying value of the company.

Competition

This is a harder one to gauge (if desired I will make a part two analyzing public competitors). In earnings calls, management often says most of their competition comes from local companies. If you analyze public competitors there aren’t many that are similar to Sterling and have as strong of returns as them.

I don’t believe their success will lead to many large competitors due to the barriers to entry. For a new entrant, it is time and somewhat capital intensive to accumulate enough skilled workers and specialized equipment to compete. It takes specialized knowledge on construction operations, project bidding, and relationship building.

For an established construction company trying to follow Sterling’s path, it can be challenging as well. Although Cutillo mentions that it doesn’t matter to them whether they’re working on a data center or battery plant, this type of work still takes specialized workers, project management, and experienced budgeting. This is why they’re able to charge such a premium compared to easy public transportation work. Sterling didn’t even enter these markets themselves, they utilized acquisitions of established companies in those sectors to capitalize on their expertise and learn from it. There is also the consideration of brand power. If Meta wants to build a new data center quickly, would they go with the contractor that has consistently performed ahead of schedule and has cultivated a direct relationship with, or use a new entrant that’s offering a lower price? Clearly Meta would go with the established provider. Sterling has the combination of their name and brand behind it plus those that they acquired in each operating segment.

M&A

M&A has been a large value driver for Sterling, immediately adding per share value and growing the operational power of the company. This of course relies on strong companies being available to acquisition and low prices to buy them at. If more entrants were to pursue a heavy M&A strategy, this could impact a significant part of Sterling’s strategy.

An aspect that benefits Sterling is the stigma against the construction industry in general. Great capital allocators looking for their next venture are unlikely to choose infrastructure construction even if a company like Sterling has been successful. Construction is known for being a lower margin often undifferentiated business and investors would be concerned if there’s enough room for growth. It takes a lot of experience to understand the dynamics of project timing, submitting bids, what a good construction company looks like, and how to form relationships with clients.

For current construction companies that could expand their M&A efforts, I doubt many will be able to successfully compete. The skills to run a construction company are far different from those needed for M&A. Many construction CEOs are technical by training such as an engineer or someone who has moved up through the ranks. Though being technical doesn’t mean they can’t think like a capital allocator, the average engineering minded CEO is likely not suited for the process that goes into doing many M&A deals. This is supported by the fact that many construction companies don’t achieve even adequate returns on capital, so many managers have not been bred to consider return on investment and valuation.

Margins/Fixed Costs

Due to fixed costs inherent in the construction business there is absolutely risk if pricing were to contract. The greater amount of fixed costs means that any decrease in pricing or volume will magnify the impact on earnings. For instance, in Q2 Building Solutions saw revenue decrease -1% and adjusted operating income declined by an astonishing -28%.

While demand has been great and earnings have grown exponentially more than revenue, it can’t continue forever. Margins are astonishingly high for a construction company (23.3% gross), no doubt fueled by the immense amount of capital the hyperscalers are spending in the AI race. I don’t see as much risk in the next few years due to project pipeline visibility, but even that could take a hit if there were an economic or industry wide downturn.

The margin considerations is one to keep a strong focus on and question management on the sustainability of these levels if possible. They have mentioned their belief that margins can grow further, but they should be tested on how sustainable these are over the long term.

Conclusion

I hope you found the write-up on Sterling Infrastructure interesting and useful, it has been an enjoyable ride to follow Sterling over the past few years and I’ve been overdue to write about it. As can be gathered from this piece, there has been a lot of positive in their recent history, and management continues to communicate more good times to come. In contrast, we have seen similar growth and contraction before with this company, but it is undeniable they are stronger and better led today.

The future prospects of the company rely on their ability to continue to find strong acquisition opportunities, high construction demand for US segments they operate in, and Sterling having good management to steer the ship.

That is all that will be covered on this part of the deep dive. If there’s enough interest I’ll release a second part, going more in-depth on: Industry Background, Competition, Financial Analysis, and Valuation.

Please subscribe to the substack if you enjoyed! It has been great fun writing this, I hope to spend more time on it in the coming months and grow this into something sustainable in the future.

Subscribe now

Also, reach out if you have any questions, comments, advice, or support!