Duolingo

The business exceeded every promise in the prospectus -- and the stock still made a round trip from +419% to a 2026 low near the IPO price, punished first by the 2022 growth crash and then by the very AI-obsolescence fear its own S-1 had buried in a competition paragraph.

Duolingo priced at $102 on July 28, 2021, closed its first day at $134.26 (+31.6%), and then rode the 2022 growth-stock bear market down to $67.70 that November, 34% below the IPO price -- all before anyone doubted the fundamentals. The fundamentals, in fact, never wobbled. Revenue compounded from $250.8 million in FY2021 to $369.5 million, $531.1 million, $748.0 million, and $1.04 billion in FY2025; net income turned positive in FY2023 ($16.1 million) and reached $414 million by FY2025. Paid subscribers grew from 1.8 million at the S-1 to 10.9 million by mid-2025, and MAUs from roughly 40 million to about 130 million by Q1 2025. The streak-and-leagues gamification engine that the S-1 described as the core asset proved to be exactly that, and the marketing chaos of the green owl -- culminating in the February 2025 'Duo is dead' Cybertruck stunt -- kept acquisition essentially free.

AI arrived as both the bull and the bear case. In March 2023 Duolingo launched Max, a premium tier built on GPT-4 with AI conversation and explanation features, and in April 2025 it shipped 148 new AI-generated language courses. The market loved the first act: the stock peaked at $529.05 on May 11, 2025, up 419% from the IPO. Then the second act began. In January 2024 the company had already cut about 10% of its contractor workforce in favor of generative AI; in April 2025 CEO Luis von Ahn's 'AI-first' memo -- announcing AI would replace contract workers -- triggered a ferocious backlash in May 2025, with users ending streaks in protest, flooding the company's social accounts, and forcing Duolingo to scrub its TikTok and Instagram videos and issue a CEO clarification. It was almost precisely the mission-betrayal scenario the S-1 had warned about.

The stock's collapse, though, was mostly about the other AI story. As DAU growth decelerated through late 2025 and into 2026 -- down to 23% year over year by mid-2026, with revenue growth guided to around 11% -- investors reframed Duolingo from AI winner to AI casualty: if LLMs can tutor and translate for free, why maintain a streak? The shares fell from the $529 peak to $96.54 by late March 2026, an 82% drawdown that briefly took the stock below its 2021 IPO price, before recovering to $146.98 by August 2026 -- up 44.1% on the IPO over five years, roughly a market-lagging outcome despite flawless operational execution. The S-1's quietest sentence -- that better free translation could make 'people... generally become less interested in language learning' -- turned out to be the one that priced the stock.

Duolingo pitches itself as the world's most popular way to learn a language: a freemium mobile app with roughly 40 million monthly active users acquired almost entirely by word of mouth, a gamification engine that keeps them coming back daily, and a conversion funnel that turned about 5% of them into paying subscribers -- producing $161.7 million of 2020 revenue, up 129%, against a net loss of just $15.8 million.

If the bull case plays out, the flywheel keeps spinning: a beautifully designed free product attracts learners at near-zero acquisition cost (sales and marketing was only $35.0 million in 2020 against 129% revenue growth), those learners stay for months or years thanks to streaks, leagues and hearts, and an ever-larger fraction of them converts to Duolingo Plus -- penetration already rose from under 1% in 2017 to about 5% by early 2021. Revenue has grown every single quarter since the start of 2018, subscription bookings doubled in 2020, and free cash flow already turned positive, so profitability is a question of when, not whether. Beyond the core app, the Duolingo English Test rides the shift to online assessment, nearly 40% of US K-12 foreign-language teachers already use Duolingo in class, and the platform can extend into new subjects entirely. The learning and investment flywheels compound: more users generate more data, which makes the courses better and the product stickier, which brings more users.

With hindsight: Every element of the pitch over-delivered: revenue grew from $161.7 million (2020) to $1.04 billion (FY2025), the company reached GAAP profitability in FY2023 and earned $414 million in FY2025, paid subscribers rose from 1.8 million to 10.9 million, MAUs more than tripled to about 130 million, and the freemium flywheel kept acquisition costs negligible. The verdict applies to the business, not the shares: at $146.98 in August 2026 the stock is up only 44.1% from the $102 IPO after a +419% peak and an 82% AI-driven drawdown.

• Revenue has grown sequentially every quarter for over three years.

  “Although our revenue has increased each quarter since the first quarter of 2018, there can be no assurances that it will continue to do so.” source ↗

• Growth is organic and viral, not bought -- the brand itself is the acquisition channel.

  “We believe that our brand has significantly contributed to our word of mouth virality, which has in turn contributed to the success of our business.” source ↗

• Duolingo is already embedded in American classrooms, giving it a distribution moat no competitor has.

  “Based on a 2020 survey, we believe that almost 40% of foreign language teachers in US K-12 schools use Duolingo in their classrooms in some form.” source ↗

• Even after more than doubling, marketing spend remained small relative to the growth it accompanied.

  “For the years ended December 31, 2019 and 2020, our sales and marketing expenses were $15.0 million and $35.0 million, respectively.” source ↗

1. The entire model depends on keeping free users engaged and converting them to subscribers

   partly came true

   growth · structural

   Duolingo monetizes only a sliver of its users, so any erosion in engagement, stickiness or paid conversion flows straight through to revenue. The filing warns that if user growth slows, the company becomes increasingly dependent on squeezing more engagement and monetization from the base it has.

   “Our financial performance has been and will continue to be significantly determined by our success in adding, keeping and engaging users of our products and converting them into paying subscribers.” source ↗

   What happened: Engagement and conversion massively outperformed for four years -- 10.9 million paid subscribers by mid-2025 versus 1.8 million at filing, MAUs around 130 million. But the S-1's warning that a slowing growth rate would make everything else fragile proved exact: when DAU growth decelerated to 23% by mid-2026 and revenue guidance fell to roughly 11% growth, the stock lost over 80% from its peak even as absolute users and profits kept rising.

2. Improving machine translation could make people stop wanting to learn languages at all

   partly came true

   tech & security · structural

   Buried in the competition section is the existential version of the risk: not that a rival app wins, but that automatic translation gets good enough that the underlying demand for language learning shrinks.

   “As free online translation services improve and become more widely available and used, people may generally become less interested in language learning.” source ↗

   What happened: The most prophetic sentence in the filing. Generative AI made free, fluent translation and tutoring ubiquitous by 2025, and while there is no clean evidence that demand for language learning collapsed -- users kept growing -- the obsolescence narrative drove the 2025-26 selloff from $529.05 to $96.54. The risk materialized in the valuation before (and possibly without ever) materializing in the user base.

3. Apple and Google control distribution and take roughly 30% of most revenue

   didn't happen

   platform dependence · structural

   Around 70% of revenue flows through two app stores whose terms Duolingo does not control, whose agreements are terminable on 30 days' notice, and which levy an approximately 30% fee on in-app payments.

   “In 2020, we derived 51% of our revenue and 53% of our total bookings from the Apple App Store, and 19% of our revenue and 20% of our total bookings from the Google Play Store.” source ↗

   What happened: The dependence persisted but never bit: no delisting, no adverse terms change materially harmed the business, and gross margins supported $414 million of FY2025 net income. App-store fee pressure eased industry-wide as antitrust rulings and regulation forced Apple and Google to loosen payment rules.

4. Operating losses every year since inception with no assured path to profitability

   didn't happen

   profitability · serious

   Duolingo has never had a profitable year and expects expenses to keep rising as it invests in product, content and headcount; it cannot guarantee revenue growth will outrun costs.

   “We have incurred operating losses each year since our inception and we may not achieve or maintain profitability in the future.” source ↗

   What happened: Losses widened at first (a $60.1 million net loss in FY2021, $59.6 million in FY2022, inflated by stock compensation), then the model flipped: net income of $16.1 million in FY2023, $88.6 million in FY2024, and $414.1 million in FY2025, with operating income positive from FY2024. Profitability arrived on the exact freemium math the S-1 described.

5. Thousands of free language-learning apps and deep-pocketed potential entrants

   didn't happen

   competition · serious

   Switching costs are near zero, barriers to entry are low, and competitors range from venture-funded startups with no pressure for profitability to Apple, which shipped its own Translate app, and social platforms with enormous distribution.

   “We estimate that there are thousands of free mobile applications for language learning; free products are provided in at least 50 languages by private companies, universities and government agencies.” source ↗

   What happened: No rival app ever threatened Duolingo's position; it remained the top-grossing education app while legacy competitors faded. The competitive threat that eventually mattered was not another language app but general-purpose AI assistants -- a different risk factor.

6. The free-forever mission constrains monetization and invites backlash if betrayed

   came true

   other · serious

   Duolingo says it will be judged more harshly than other companies if it is perceived to abandon its mission of free education -- explicitly flagging paywalled features and degradation of the free product as reputational tripwires, and reserving the right to sacrifice near-term results for the mission.

   “our brand and reputation may be negatively affected by actions we take that are viewed as contrary to that mission, such as features that are only available to Plus subscribers or changes to the free offering that are viewed as undermining how fun or effective the free offering is” source ↗

   What happened: The April 2025 'AI-first' memo -- replacing contract workers with AI -- was received by users as exactly the mission betrayal the S-1 described: streak-ending protests, comment floods that forced Duolingo to pull its social-media videos, subscription cancellations, and a CEO clarification video. The January 2024 cut of about 10% of contractors had already drawn the same criticism. The disproportionate-damage mechanism the filing predicted worked precisely as written.

7. Dependence on founders Luis von Ahn and Severin Hacker

   didn't happen

   key person · serious

   The company names its co-founders as individually critical, notes active poaching of its talent, and ties the brand's credibility to key personnel staying aligned with the mission.

   “We currently depend on the continued services and performance of our key personnel, including Luis von Ahn and Severin Hacker.” source ↗

   What happened: Both founders remained in place through 2026 and execution never suffered from key-person loss. The nuance: the founder-led structure meant von Ahn's own AI-first memo became the company's worst reputational crisis, with no internal check on the messaging.

8. The Duolingo English Test depends on universities continuing to accept online assessment

   didn't happen

   market · serious

   The DET's revenue rests on institutional acceptance that could reverse: some universities were moving away from standardized testing altogether after COVID, and some educators questioned the validity of online-proctored language exams.

   “As a result of the COVID-19 pandemic, in 2020, a number of universities waived standardized test requirements for admissions requirements and some universities plan to phase out requirements for standardized testing altogether.” source ↗

   What happened: Acceptance broadened rather than reversed -- the DET grew in popularity and institutional acceptance after COVID, adding universities such as LSE and Imperial College and formal recognition in Ireland's student-visa program, though holdouts like Oxford and Cambridge remained. The feared retreat from online assessment never came.

9. China operations exposed to regulatory and geopolitical risk

   came true

   regulation · serious

   Duolingo was investing to expand in China -- an intensely competitive market where it acknowledged it might never be profitable -- while flagging that regulatory uncertainty could shut it out entirely, and that its Beijing office made it easier for authorities to act against it.

   “Moreover, geopolitical tensions or regulatory uncertainty in countries in which we operate, such as China, may prevent us from operating in certain countries or increase our costs of operating in those countries.” source ↗

   What happened: Within weeks of the IPO, the Duolingo app was removed from some Chinese app stores, staying unavailable from August 2021 to June 2022 amid Beijing's crackdown on private education. The impact on consolidated results was modest, but the specific scenario disclosed -- Chinese authorities abruptly cutting off access -- happened almost immediately.

10. Advertising revenue is hostage to Apple's privacy changes and ad-blocking features

    didn't happen

    platform dependence · serious

    About 17% of 2020 revenue came from ads served to free users, a stream directly threatened by iOS 14's IDFA opt-in requirement and the broader collapse of targeted-advertising identifiers.

    “internet-connected devices and operating systems controlled by third parties increasingly contain features that allow device users to disable functionality that allows for the delivery of advertising on their devices or reduce the ability to provide personalized or targeted advertising, which results in less valuable ads” source ↗

    What happened: IDFA deprecation never mattered much because subscriptions, not ads, drove the model; advertising shrank as a share of revenue while subscription revenue carried the company to $1.04 billion in FY2025. The risk was real for ad-dependent peers, not for Duolingo.

11. Dual-class structure concentrates 91.5% of voting power with insiders

    didn't happen

    governance · serious

    Class B shares carry 20 votes each -- double the usual super-voting ratio -- leaving public Class A holders with essentially no say over directors, mergers or charter amendments, and making the company ineligible for major indices.

    “As of March 31, 2021, our directors, executive officers, and 5% stockholders and their affiliates held in the aggregate 91.5% of the voting power of our capital stock.” source ↗

    What happened: No governance crisis materialized and the concentrated control coincided with strong execution. The structure did mean the controversial AI-first strategy and its reversal-by-apology were decided entirely by the founders, with public shareholders as spectators -- but no measurable harm followed.

12. A disclosed material weakness in internal control over financial reporting

    didn't happen

    operations · serious

    In 2019 Duolingo lacked adequate accounting personnel, risk assessment and journal-entry review; remediation was still unproven at IPO, no Section 404(a) assessment had ever been performed, and the auditors had never audited internal controls.

    “In fiscal year 2019, we identified a material weakness with our internal control over financial reporting as we had not designed or maintained an effective control environment and associated control activities to meet our accounting and reporting requirements.” source ↗

    What happened: No restatement, control scandal, or reporting failure emerged after the IPO; the company scaled to a billion dollars of revenue with clean financial reporting, indicating the 2019 weakness was remediated as described.

13. Security breaches have happened before and are promised to happen again

    partly came true

    tech & security · serious

    With a huge trove of personal data on a high-profile consumer platform, Duolingo states flatly -- not hypothetically -- that breaches and disruptions will recur.

    “Such security breaches or disruptions have occurred on our systems in the past and will occur on our systems in the future.” source ↗

    What happened: As promised, an incident occurred: in January 2023, data covering over 2.6 million users -- usernames, names, and phone numbers scraped through an exposed API -- was offered for sale on a hacking forum. Embarrassing but not material: no financial-data compromise, no discernible impact on growth or results.

14. Near-total reliance on AWS and Google Cloud with no formal disaster recovery plan

    didn't happen

    operations · serious

    Nearly all product traffic and data run on two third-party clouds, systems are not fully redundant, and the company admits it has no formal disaster recovery plan.

    “We do not have a formal disaster recovery plan, and system backups do not account for all possible eventualities and our property and business interruption insurance coverage may not be adequate to compensate us fully for any losses that we may suffer.” source ↗

    What happened: No prolonged outage or cloud failure materially disrupted the service; the infrastructure scaled from roughly 40 million to about 130 million MAUs without a disclosed incident of consequence.

15. Historical service to users in US-sanctioned countries

    didn't happen

    legal · serious

    Duolingo has served users in Iran, Cuba and Syria, relying on exemptions from sanctions law; a determination that it failed to comply could bring fines, enforcement actions and disgorgement.

    “we have historically provided services to users in Iran, Cuba and Syria, countries that are presently the subject of comprehensive sanctions by the US government” source ↗

    What happened: No sanctions enforcement action against Duolingo was ever disclosed; the exposure remained theoretical.

16. Headline user metrics are self-measured and unaudited

    didn't happen

    other · boilerplate

    MAUs, DAUs and paid-subscriber counts -- the numbers the whole story rests on -- come from Duolingo's own analytics platform, have never been independently validated, and can be miscounted by background app activity.

    “Our user metrics are calculated using internal company data gathered on an analytics platform that we developed and operate, have not been validated by an independent third party” source ↗

    What happened: No metric restatement or measurement scandal occurred. Ironically, the self-reported DAU growth rate became the single number the market traded on -- its deceleration in 2025-26, not any doubt about its accuracy, is what crushed the stock.

How the S-1 reads

This is a 2021-vintage consumer-subscription S-1 with all the era's furniture -- dual-class shares at an aggressive 20:1 ratio, a Robinhood retail tranche, emerging-growth-company exemptions, COVID framing -- wrapped around an unusually healthy business: near-breakeven, cash-generative, and growing organically. Two drafting choices stand out. First, the mission is treated as a genuine operating constraint rather than marketing: the risk factors explicitly warn that paywalling features or degrading the free app would damage the brand more than it would damage an ordinary company, effectively pre-registering the terms on which users would later judge management. Second, the sharpest risk in the document is a single clause inside a generic competition paragraph -- that better machine translation could make people less interested in learning languages at all -- an existential technology risk filed under competitive pressure, years before generative AI gave it teeth.

• “There may be risks associated with the use of the Robinhood platform that we cannot foresee, including risks related to the technology and operation of the platform, and the publicity and the use of social media by users of the platform that we cannot control.” source ↗

  Peak-2021 zeitgeist: Duolingo reserved part of its IPO for retail investors via Robinhood, then disclosed it could not control what that crowd might do.

• “Our office of approximately 17 employees in Beijing makes it easier for the Chinese authorities to bring enforcement actions against us.” source ↗

  An unusually blunt admission that physical presence in China creates enforcement leverage over the company.

• “maintaining a free version of the app that is both effective and enjoyable is central to Duolingo’s mission” source ↗

  The company defines its own free product as sacred -- a constraint it warns investors it will honor even at the cost of short-term results.

• “Our Class B common stock has 20 votes per share, and our Class A common stock, which is the stock that is being sold in this offering, has one vote per share.” source ↗

  The 20-votes-per-share structure is stated plainly: public buyers get one vote, insiders get twenty.

What this one teaches

• Execution and shareholder outcome can fully decouple: Duolingo beat every operating promise in its S-1 -- 6x revenue, profitability, 6x subscribers -- and an IPO buyer was still up only 44% after five years, having endured a -34% trough, a +419% peak, and an 82% drawdown. The multiple, not the business, was the investment.
• The most dangerous sentence in an S-1 can be a subordinate clause: the warning that better free translation could make people 'generally become less interested in language learning' was filed under competition in 2021 and became the entire bear thesis in 2025-26, cutting the market cap by more than half in a year.
• Mission-based-company risk factors deserve literal reading. Duolingo predicted it would be punished disproportionately for acts 'viewed as contrary to that mission' -- and the 2025 AI-first memo produced exactly that: streak-ending boycotts and a scrubbed social-media presence, a self-inflicted crisis no competitor could have engineered.
• AI cut both ways for the same company within 24 months: GPT-4-powered Max and AI-generated courses drove the 2023-25 rerating to $529, then AI-obsolescence anxiety and decelerating DAUs drove the crash below $100. The technology was one variable; the market's story about it flipped sign.
• Disclosed operational risks with named mechanisms (China enforcement, recurring breaches) materialized quickly and cheaply -- the app was pulled in China within weeks of the IPO and user data was scraped in 2023 -- yet neither moved the long-term outcome. What moved it was the growth-rate second derivative the S-1 could only gesture at.

The paper trail

1. 2021-06-28 S-1 filing index ↗ document ↗
2. 2021-07-19 S-1/A filing index ↗ document ↗
3. 2021-07-26 S-1/A filing index ↗ document ↗
4. 2021-07-28 424B4 filing index ↗ document ↗

Filed as Duolingo, Inc.. All documents are public domain, served by SEC EDGAR.