Duolingo Crashed on AI Fears. Its User Numbers Don't Confirm the Doom Yet

The S&P 500 sits within a few percent of its record high while much of the financial press warns that a crash is imminent and the index keeps climbing. Duolingo never waited for that crash. The language-learning app has spent the past year living through one of its own: the stock is down more than half over the last twelve months, pressed down by a specific, testable fear. That fear — that free AI tutoring makes a paid language app obsolete — is stamped into the price. The question worth answering is whether the company's own numbers agree with it.

The selloff had more than one author

Two forces drove Duolingo down. The first is the AI-casualty narrative. Investors reasoned that as ChatGPT-class models tutor for free, nobody would keep paying a monthly subscription for a cartoon owl. The story won so completely that short interest ran near 21% of the float, and 19 of 27 analysts sat on Hold.

The second force was self-inflicted and, on its face, worse. In February, chief executive Luis von Ahn announced that the company would deliberately slow how hard it pushes users to pay, sacrificing an estimated $50 million in near-term bookings by giving away more of its premium AI features in exchange for user growth. Management guided 2026 revenue growth down to the mid-teens from roughly 40% a year earlier. The market read that as a company in trouble, and the shares kept falling.

The users say something else

Then second-quarter results landed on August 5, and the number that matters most for the AI thesis was not on the income statement at all. Daily active users grew 23% year over year to 58.7 million, and paid subscribers rose 17% to 12.7 million. Revenue came in at $298.5 million, up 18.3%, ahead of estimates, with adjusted earnings that also beat.

That is the whole crux of the AI argument. If free artificial intelligence were truly killing Duolingo, the destruction would show up in usage first — before it ever reached revenue. It is not showing up. Users keep returning in rising numbers, and the habit mechanics that make the product sticky — the streaks people refuse to break, a decade of word-of-mouth, the behavioral data used to train the app's AI — are the moat the bear case wants to pretend doesn't exist. This is the difference between "competition exists" and "the moat is breached," and the user data currently supports only the former.

The honest catch: this is a deliberate bet, not a done deal

But the market is not crazy, and I would not pretend otherwise. The stock fell 12% after that beat, and there was a legitimate reason. Duolingo guided third-quarter revenue to about $302 million, a hair below expectations, and bookings — the money users commit to — rose only about 8%, against 23% growth in users. Management is, by its own design, converting near-term dollars into users, betting that a bigger audience of 100 million daily actives by 2028 will monetize later. That is a trade, and until bookings reaccelerate, every guidance call will keep moving the stock more than the fundamentals will.

The clever part is that the company is using AI to fund the trade. Video Call, its signature AI conversation feature, has seen inference costs fall by more than tenfold since launch, which is what lets Duolingo hand out more premium access for free. In mid-August, DA Davidson upgraded the shares to Buy with a $160 target, arguing the "AI kills Duolingo" thesis had broken on accelerating engagement.

What it costs, and what could break it

Where this lands on value is where I have to stop short of a triumphant "buy it all today." The quality is genuinely intact: roughly 72% gross margins, a free-cash-flow margin near 35%, return on invested capital around 30%, no debt, over a billion dollars of cash, and a $400 million buyback the company has authorized. But the valuation is not dirt cheap in absolute terms — the market capitalization is about $6.5 billion, and the stock trades in the neighborhood of 40 times forward earnings. What has happened is that a multiple that used to carry triple-digit expectations in a high-growth SaaS era has collapsed, and the price now bakes in a substantial amount of near-term doom.

That is the honest middle. The market has arguably priced a disaster scenario its own user metrics have not produced yet, which shifts the burden of proof toward the bears — but the re-acceleration that would make this a comfortable contrarian buy is unproven. So the two numbers to watch are daily active user growth and bookings growth. If DAU growth holds above the high teens and bookings begin to converge back toward it, the discount gets bigger by the quarter. If DAU growth slips toward the low teens or bookings stay pinned near single digits with no clear path to monetization — or, worse, a real AI lab ships a serious rival and usage finally cracks — the market's pessimism will have been right.

For an entry, the price has stabilized in the mid-$130s, just above its 50- and 200-day averages and well above the $87.89 low, after the February collapse and the August post-earnings dip. That is the shape of selling exhaustion rather than a falling knife, but it still argues for scaling in rather than chasing — and for letting the quarterly user-and-bookings data confirm the thesis before adding meaningfully.

You do not need to wait for the next market crash for this setup to matter. Duolingo has already repriced on a specific, testable fear, and that fear now has a scoreboard that updates every quarter. The market was arguably premature to pronounce the company an AI casualty based on price alone. Whether it is a gift is a bet on whether users eventually pay — a live decision, not a certainty.