Duolingo is the company with perhaps the largest disconnect between its current market cap and the value many investors see online. Duolingo trades at a market cap of just $5 billion, while some project it will become a “trillion-dollar company.”

Two things upfront. People who tell you Duolingo will become a trillion-dollar company are delusional. Nothing else. But… I was surprised to see it at such reasonable valuations. Less than 20x free cash flow, with revenue growing by more than 30%, is rare. And you still have the founder onboard, which is something we love to see as well.

With Duolingo, you have to jump into the role of a business analyst, not a financial analyst. We will have a valuation at the end of the newsletter, as always, but the main question will be whether Duolingo’s business model is sustainable in a world of AI.

Given the size of the industry, if Duolingo comes even close to becoming the world’s largest learning platform, it will be a multibagger. Today, we want to find out whether that’s realistic.

Let’s dive in!

— Daniel

Duolingo – Attractive Setup or the Beginning of the End?

If you’ve spent time on Social Media in the last few years, there’s a good chance you’ve come across a chaotic, passive-aggressive green owl. Duolingo’s social media marketing is certainly unique in corporate America. And as we will see later, this marketing strategy has been a massive part of Duolingo’s success.

But Duolingo’s success goes beyond a great marketing engine, of course. Some years ago, I used Duolingo for quite a long time. I had a 300-day streak of learning Spanish. Spoiler: I can’t speak any Spanish anymore, and I never really could.

But upon reflection, there are some interesting aspects to this. First of all, how often have you heard anyone talk about their Duolingo streak? I’ve heard that quite often. Or at least I’ve seen it on Social Media. People are proud of their streaks because it shows that they are “committed learners.” The other side of that story is that most people can’t actually speak the language, just like me, but still, a streak on Duolingo is something you are proud of.

That’s why the friction to pay for Duolingo is low compared to other mobile apps. Duolingo gives you a good feeling. Other games or apps don’t. I have never heard anyone proudly state that they have been a paying member of Tinder or Hinge.

It’s a similar dynamic to gym memberships. Gyms don’t really make money from the people who actually go, but from the people who pay for their membership but never show up. I wouldn’t say that’s the case for Duolingo to the same extent, but they benefit from similar dynamics on the margin.

Another possible advantage is reactivation rates. Duolingo doesn’t publish reactivation numbers, but from anecdotal evidence, many people pay at some point, leave, and eventually return as paid members. In part, that’s because Duolingo is the strongest brand in the industry, and, judged by what they optimize for, and we will get into that, there is no viable substitute.

What’s notable, too, is that Duolingo has evolved significantly in the last few years. It’s no longer just about languages. The company has rolled out courses in music, math, and chess. Chess, in particular, is growing at a clip that outpaces even the early days of the language product.

Founder and Management

Duolingo was founded in 2011 by Luis von Ahn and his co-founder, Severin Hacker. Luis was a math professor at Carnegie Mellon, and Severin was his PhD student. Luis also founded reCAPTCHA — you know, that box where you have to tick “I’m not a robot” — and later sold it to Google.

And although reCAPTCHA was sold rather quickly, and it’s not the same as scaling a business like Duolingo, it’s still reassuring to see that he’s a “serial entrepreneur.” And it’s even more reassuring that he’s a math professor, so clearly a very intelligent guy. I might fall prey to the authority bias here, but I take that. Especially since my first impression of Luis was a bit, well, weird.

As I told Shawn in our podcast, my first impression of Luis was on the Q3 earnings call. For context, after those earnings, the stock plummeted 30% (so even more than after Q4 earnings…). On the call, Luis was explaining to analysts how big the opportunity ahead is. While doing so, he pointed at the camera and pressed a button to shoot cash emojis up the screen.

The fact that I keep bringing this up might give you an idea of how irritated I was. Anyway, after doing some more work on the company, I’m quite confident he’s a good CEO. Highly motivated, smart, and long-term focused.

He’s also highly aligned with shareholders, considering his 7% stake in the company.

Gamification – Duolingo’s Secret Weapon

Duolingo appears to be a young company. And it is, compared to most other firms in the market. However, compared to other apps, especially language-learning apps, Duolingo has actually been one of the first movers.

But its success is really not about being the first mover. Duolingo has been one of the few companies that successfully integrated gamification into its app.

Language learning is one of those things that is a perfect fit for a gamified experience. Lessons can be broken down into countless tiny, repeatable actions with fast feedback loops. Whether it’s practicing vocabulary, grammar, or speaking, all of this can be taught in quick 60-second lessons with immediate feedback.

Anyone who has played video games in the past knows that they’re incredibly good at taking you on a journey and making you get back at it quickly. That’s what Duolingo does, too. Language skill comes down to spaced repetition and retrieval practice. All you need to do is show up often and stick with the journey. Again, as in a video game.

You’d think that’s applicable to most other things as well. In fact, that’s why there was such a big momentum for gamifying just about anything 10 years ago. However, as it turns out, some things are easier to gamify than others.

Gamification fails when feedback is delayed or progress isn’t linear. Then, gamification actually works against you. Users would start optimizing for the metric (for example, a streak) rather than the outcome (for example, finishing a lesson) to improve their learning.

That exact thing happens to Duolingo users, too, by the way. Both Shawn and I have experienced firsthand how, after some time, you find yourself working on the shortest lesson possible to keep your streak, but spend as little time on the app as possible.

So even when gamification works well in general, you will have these experiences. Charlie Munger would say that you get what you incentivize. If you incentivize building a streak, most people will eventually focus more on keeping that streak than on making progress in learning a language.

This is a general problem that I have with Duolingo. Is this a serious learning platform that will expand into many verticals and become the dominant learning platform of the 21st century, or is it ultimately just a mobile game that is rather ineffective at teaching languages?

Most likely, the truth is somewhere in between. Duolingo is a game with learning capabilities. The important question we will have to answer regarding that is: what does this definition mean for monetizing its user base?

But let us quickly look again at how Duolingo works and the balance it has to strike between converting users to the paid tier and growing active users overall.

The company runs about 750 A/B tests a quarter to optimize everything, from the color of icons to more significant structural changes like replacing the old “tree”-style home screen with a linear path to better guide learners.

User Metrics and the TAM Opportunity

Duolingo has about 50 million daily active users and more than 135 million monthly active users. And those numbers have been growing rapidly in recent quarters and years. However, we currently see a slowdown. And it’s not totally clear why. Luis von Ahn discussed the impact of shifting the social media strategy from unhinged content to content about the mission.

However, it is also important to note that the decline in users has occurred on a quarter-on-quarter basis, and only for Monthly Active Users (MAUs).

The focus is on growing Daily Actives (DAUs) rather than monthly actives. For an app that uses streaks to keep users engaged, daily active users are much more important. They are more likely to stay longer and upgrade to the paid tiers.

The Super and Max tiers are the paid alternatives to the free version. The Super membership gives you unlimited energy, meaning you can learn as long as you want, but it doesn’t include new AI features. That’s what you get in the Max or Super Family tier. These new AI features include things like calls with Lilly or other Duolingo characters that let you have an actual conversation in the language you’re learning. Since the feature launched, bookings for the Max tier have doubled.

Currently, about 9% of monthly active users are paid subscribers — roughly 12 million people. And while growth is naturally slowing, paid subscribers are still growing at close to 30% a year.

However, there’s something interesting in the data. Daily active users have been growing faster than paid subscribers. So if we use daily actives instead of monthly actives as the denominator, the paid conversion trend is actually declining.

That isn’t necessarily alarming because there’s a time lag between someone becoming a daily active user and deciding to upgrade. Still, it’s an important metric to monitor. In theory, as DAU growth normalizes and cohorts mature, paid subscriber growth should re-accelerate relative to DAUs. If not, that could mean churn rates are rising, or conversion is worse for newer cohorts.

Based on active users, Duolingo currently captures about 7% of all language learners. But it captures only about 1% of the money spent on language education. As more of that moves online and Duolingo improves its monetization, you could have a market leader with double-digit market share in a trillion-dollar industry. That’s the bull case in a nutshell.

The AI Question — Risk or Opportunity?

Let’s switch gears and look at the other side of the story. AI might be the primary driver for Duolingo’s massive 80% drawdown (besides valuation).

The stock went from an all-time high with a price-to-sales ratio of 30x to below 6x. On an EV/FCF level, the company came down from ridiculous levels over 80x to a moderate 12x times — and adjusting for stock-based compensation, you get a more realistic EV/adjusted FCF of around 20x.

So what is the market afraid of with Duolingo? There are multiple things. But the most obvious one is that LLMs or AI-native apps will be able to recreate Duolingo more cheaply and even more effectively, increasing competition.

And we will dig deeper into whether that’s realistic, but first, I want to show you the numbers. I often see the comparison to Chegg, an adtech company that provides help with homework and school tasks. Since ChatGPT launched, Chegg's stock has been completely crushed, down over 99%.

Now compare that to Duolingo, which has grown its daily active users from 15 million to 50 million, revenue from $250 million to almost $1 billion, and operating margins from negative 17.5% to positive 11%. Up until now, Duolingo has nothing but benefited from AI.

And that doesn’t come as a surprise to anyone who has ever tried learning a language with the help of any LLM. It’s a complete nightmare. There’s no way anyone would settle for that when an app like Duolingo exists.

The hard part is getting users to log in every day for a long time and matching the lessons to their current level. The fact that Duolingo runs about 750 A/B tests a quarter shows how much work goes into optimizing the experience.

One insight Shawn brought up that I think is underappreciated: LLMs are only trained on outputs, not inputs. They can see how Duolingo works from a user perspective, but the actual details behind the algorithms are closely guarded internally. Duolingo’s 50 million daily active users generate a ton of data — every exercise produces labels like correct or incorrect, time taken, hints used, and type of error. All of that is why Duolingo can train its AI models so efficiently.

That said, Duolingo is no Meta. Meta’s ecosystem generates a magnitude of valuable data more than Duolingo ever could. If I had to bet, I would say Duolingo’s data advantage alone wouldn’t be enough to prevent a competitor from building a similar app with AI long-term.

There’s another argument and possible AI threat that Shawn brought up in the podcast. What if real-time translation keeps improving? Will people even bother to learn new languages? He told me how he has no problem communicating in English in most places anyway, and beyond that, he never had any trouble even using Google Translate. And real-time translation could magnify that.

I could certainly see a future where Zoom and Google Meet offer seamless, live translation as a built-in feature. That might reduce the need to learn a language in the professional environment, an important market for language-learning apps. However, I believe that argument underestimates the complexity of communication. Tone, facial expressions, timing, the way someone emphasizes a syllable — all of that gets lost or delayed when you're routing a conversation through a translator.

I'd actually argue that someone speaking imperfect English in a meeting is a far more effective communicator than someone speaking fluently in their own language while waiting for an AI to relay the message. And there's something more personal at play, too, something a native English speaker might not appreciate to the same extent. When someone addresses you in your native language, knowing it's not their mother tongue, it signals effort, respect, and commitment in a way that no translation tool can replicate.

Maybe that all sounds a bit woo-woo, but why did Google Translate not disrupt language learning at all? Why did the fact that I can now watch all international shows on Netflix in my native language not disrupt language learning?

If anything, all of these innovations increased people's motivation to study new languages.

All in all, I don’t see an immediate threat to Duolingo’s business from AI. However, that doesn’t mean that I believe Duolingo has an impenetrable moat.

Does Duolingo have a Moat?

The more companies I study, the more I feel the term “moat” is thrown around too often. A moat is more than a short-term competitive advantage. Duolingo’s first-mover advantage, for example, was a competitive advantage, but it’s not a moat.

I believe there are three competitive advantages that could evolve into a moat over time. In my opinion, and I’ll elaborate on this, none of them is a moat yet.

The first is the brand. At this point, Duolingo is the first thing that pops into your head when you think about language learning. And that’s the case in most countries on earth.

The second is data. 50 million daily active users produce a treasure trove of learning data. Duolingo can not only personalize lessons for you, but it also knows that, say, German speakers struggle with certain English words or structures and can include that in lessons for all German users. In that aspect, Duolingo is similar to what Spotify can do with its data. However, Spotify has a flywheel where users and artists reinforce each other. You don’t have a flywheel like that at Duolingo.

Duolingo also has billions of data points that help find the right balance between increasing paid conversions and increasing DAUs. It’s not impossible for AI to speed run such tests itself. However, you would first need the users. And that will be difficult.

The third is Duolingo’s ability to change user behavior. Uber and Airbnb changed people’s behaviors, and that unlocked hundreds of billions of dollars in value. Duolingo is pretty good at that, too. It gets people to practice languages daily, which is genuinely hard. If it can do the same for other verticals, they can create a huge market for themselves.

Perhaps Duolingo creates a category in between mobile games and learning platforms, as Airbnb has created a category between hotels and, let’s say, camping, or any other way to travel.

A big weakness, though, is that Duolingo has no obvious network effect. You could start a learning challenge with a friend, but it’s way less direct than platforms like TikTok or Instagram, where you directly benefit from other people being on the platform. One possible exception is chess, where player-versus-player creates a real network effect. If that vertical succeeds, it changes the equation.

Another difficulty is that Duolingo has no physical footprint. Historically, apps without physical infrastructure have been easier to disrupt. Think about Clubhouse, which had a $4 billion valuation at one point and has now vanished into irrelevance.

Most successful app businesses have a mix of digital and physical infrastructure — Uber has drivers, DoorDash has restaurants. Duolingo doesn’t have any of that, and that usually attracts competition due to the lack of barriers to entry.

Sometimes, companies can survive strong competitors. I always think about Netflix in that regard. I used to believe that the “Streaming Wars” would be a problem for Netflix. Disney and Amazon have deep pockets, and Disney’s IP put them in a strong position.

And yet, Netflix was never really hurt. They were simply better. Sometimes it seems to be that easy. And obviously, Netflix had a whole lot of capex, too, but my point is that, while it shielded Netflix from further competition, there was already sizeable competition. So perhaps Duolingo will succeed simply by working slightly better than anyone else. Again, LLMs can only be trained on output! It’s a difference on the margin, but it could be enough.

To summarize, I believe Duolingo’s best chance at a moat will be the data advantage. Brand recognition is nice, but given the sharp slowdown in growth after changing the social media strategy, it seems Duolingo’s brand isn’t yet enough to attract users on its own. The dependence on social media leaves me with some questions, generally.

If the business cannot sustain its growth without heavy, viral marketing campaigns, that says something about the staying power and quality of users.

The Monetization Challenge Outside the U.S.

One last, and in my opinion, not enough-talked-about point is the inherent monetization problem of Duolingo. On one of the latest earnings calls, Luis discussed how difficult it can be to monetize users outside the U.S. or Europe. He said something that reminded me a lot of the “paying for convenience” dynamic that we discussed when we covered Mercado Libre.

In many South American countries, people would rather wait a week for delivery than pay extra to have it arrive early. And it’s the same for Duolingo. As long as you can use the app for free, people do that. Even the power users. They prefer watching ads and being limited in how much they can study to paying for Super or Max, even though it’s significantly cheaper than in the U.S.

Those users, though, are the ones targeted when Luis speaks about the mission of teaching languages like English to open career opportunities for them. U.S. or European users tend to see language learning as a hobby.

About 50% of revenue comes from the U.S., but only 20% of daily active users are located there. That means 20% of users are responsible for 50% of revenue. And, again, you might speculate that these customers are also less sticky.

Valuation and Investment Decision

Valuing companies like Duolingo is difficult because it's hard to judge how quickly they will continue to grow. There's a huge difference between a company growing at 30%, 20%, or 10%, obviously. And yet, sometimes, high-growth companies go from 30% growth rates to 15% within a quarter or two.

That just happened to Duolingo. After growing revenue by almost 40% in FY2025, the guidance for next year projects top-line growth of only 15%. That's well below market expectations, and the stock reacted with an immediate 20% drop.

The reason for the slowdown is that management wants to focus more on growing DAUs than on conversion to the paid tiers. Again, this is a balance Luis von Ahn has to continuously strike. Still, up until today, both metrics, DAUs as well as top-line growth, were growing simultaneously, so it's not a good sign that this no longer seems to be the case. Or at least temporarily it's not.

Based on that, I also adjusted my valuation model from when we recorded the podcast episode.

One accounting note before we get to the numbers: Duolingo reports bookings alongside revenue. When a subscriber pays $120 for an annual plan, the company receives the cash upfront, but recognizes revenue at, for example, $10 per month over the following year. Bookings, therefore, give you a more real-time picture of demand, while revenue lags behind.

In my updated base case, I assume bookings growth of 17% through 2028, decelerating to 12% through 2030. That's meaningfully slower than what I modeled before the guidance revision. Adjusted free cash flow margins start at 20% and gradually increase to 23%, with share dilution of about 2% per year. Using a weighted-average exit multiple of roughly 21x, I arrive at a fair value of about $150 before accounting for any margin of safety. Apply a 30% discount for the range of outcomes, and the intrinsic value target drops to around $105 — roughly where the stock sits today.

The bull case still paints an attractive picture. If bookings growth reaccelerates back to 25%, free cash flow margins expand to 30% by decade's end, and the market assigns a 25x exit multiple, fair value rises to approximately $345. Even after a 30% margin of safety, that implies an intrinsic value north of $240 — more than double the current price. From today's entry at $100, the implied return in that scenario would be close to 30% annually. This might sound unrealistic after the recent slowdown, but one shouldn’t forget that the business's earnings power hasn’t changed as dramatically as the guidance suggests.

It reflects a focus on increasing DAUs, and theoretically, this should boost growth in the future, meaning the growth Duolingo is currently missing out on should appear then.

The bear case, however, has gotten uglier. If bookings growth slows to just 10%, margins stall at 15%, and dilution accelerates to 4% per year — all reasonable assumptions if the business struggles — free cash flow per share barely grows over the next five years. With a 17x exit multiple, fair value falls to roughly $58, and the margin-of-safety target drops to about $41. That implies a negative return from today's price. This is what keeps me up at night with this stock.

The spread between these scenarios is enormous. Duolingo is one of those cases where you primarily need to be a business analyst rather than a financial analyst.

And from the standpoint of a business analyst, I must say that I see too many yellow flags to be confident in Duolingo's success. My main problems are the stickiness of the product with customers, or better said, the lack of, the ARPU problem of users outside of the U.S. and Europe, and the lack of a strong moat. Although I’m aware that the data could turn into such a moat over time.

As mentioned in the podcasts, I bought a very, very small position for my personal portfolio to keep an eye on Duolingo, but I won’t suggest we add it to the Intrinsic Value Portfolio.

For more on Duolingo, you can listen to our podcast here.

More updates on our Intrinsic Value Portfolio below 👇

Weekly Update: The Intrinsic Value Portfolio

Notes

• Netflix stepped away from its ambition to take over Warner Bros for a price target beyond $80 billion. The market’s reaction showed very clearly what it was thinking about the potential deal. Netflix is up 30% since the announcement, while Paramount is down by double digits.

  • While we loved the idea of Netflix owning IP like Superman, Harry Potter, and Game of Thrones, there is always a price too high. Therefore, we are relieved that, when Paramount, once again, increased its bid, Netflix was willing to walk away.

  • We were just being lazy when not accounting for the acquisition of Warner Bros. in our Netflix model, but now that seems prescient. The thesis that Netflix is undervalued is certainly a lot cleaner without the WBD acquisition, and for Netflix's troubles, they will walk away from the deal $2.8 billion richer, thanks to a termination fee payout from WBD after WBD chose to renege on the previously agreed-upon deal in favor of Paramount's updated offer.

Quote of the Day

"Even in those earlier times, finding the really outstanding companies and staying with them through all the fluctuations of a gyrating market proved far more profitable to far more people than did the more colorful practice of trying to buy them cheap and sell them dear.”

— Philip A. Fisher

What Else We’re Into

📺 WATCH: Norges Bank Interview with the co-founder of Zalando

🎧 LISTEN: A new round of stock pitches by Stig Brodersen, Tobias Carlisle, and Hari Ramachandra

📖 READ: A WSJ article on whether insider buys are actually a good indicator of future returns

You can also read our archive of past Intrinsic Value breakdowns, in case you’ve missed any, here — we’ve covered companies ranging from Alphabet to Airbnb, AutoZone, Nintendo, John Deere, Coupang, and more!

See you next time!

Enjoy reading this newsletter? Forward it to a friend.

Was this newsletter forwarded to you? Sign up here.

Join the waitlist for our Intrinsic Value Community of investors

Shawn & Daniel use Fiscal.ai for every company they research — use their referral link to get started with a 15% discount!

Use the promo code STOCKS15 at checkout for 15% off our popular course “How To Get Started With Stocks.”

Follow us on Twitter.

Read our full archive of Intrinsic Value Breakdowns here

Keep an eye on your inbox for our newsletters on Sundays. If you have any feedback for us, simply respond to this email or message [email protected].