Andrew Hsu Built a $1.5M Startup at 19 — It Collapsed. His $1B Comeback Broke Every EdTech Rule
A child prodigy who entered university at 12, got a Thiel Fellowship at 19, and watched his first startup implode. Andrew Hsu's second act — the AI language app Speak — hit $1B valuation by doing the opposite of what made Duolingo famous.
TL;DR
- ✓ Andrew Hsu entered university at 12, earned three bachelor's degrees by 16, and dropped out of a Stanford PhD to start a company at 19 — which collapsed within a year (Wikipedia)
- ✓ His first startup Airy Labs raised $1.5M from Google Ventures, then imploded amid accusations of family-run micromanagement (TechCrunch)
- ✓ His second startup Speak hit $1B valuation and $100M+ annualized revenue by refusing to copy Duolingo's gamification playbook (Forbes)
- ⚡ The real unlock wasn't genius — it was finding a co-founder (Connor Zwick) and picking a hyper-dense market (South Korea) instead of trying to boil the ocean
- 💬 "Child prodigy" is a branding curse: it attracts capital but repels the organizational maturity you actually need
The Prodigy Trap: When Your Resume Is Better Than Your Company
Here's a resume that should guarantee startup success: entered college at 12, three degrees by 16, Stanford PhD candidate at 16, published author, featured on CBS and Time Magazine.
✓ Andrew Hsu had all of that before he turned 20 (Wikipedia). And his first company still fell apart in under a year.
✓ In 2011, Hsu accepted a Thiel Fellowship — Peter Thiel's $100,000 bet on brilliant kids who drop out to build companies (Forbes). He left Stanford to build Airy Labs, an educational gaming startup. ✓ He became the first Thiel Fellow to raise venture capital — $1.5M from Google Ventures and others (Forbes).
Then it all unraveled.
✓ By February 2012, Airy Labs had laid off most of its team. Former employees told TechCrunch the company operated like a family business — Hsu's parents were deeply involved, workers were expected to put in 10+ hour days seven days a week, and management style was described as micromanagerial (TechCrunch). ✓ The company couldn't turn a profit and eventually stalled out (Hack Education).
💬 The lesson isn't "prodigies fail." It's that intelligence and organizational competence are completely different skills. You can solve differential equations at 12 and still not know how to delegate to a 30-year-old engineer.
Decision #1: Find a Co-Founder Who Isn't You
✓ Four years after Airy Labs collapsed, Hsu co-founded Speak in 2016 with Connor Zwick — another Thiel Fellow who'd dropped out of Harvard (Forbes). Zwick had already built and sold a company (Flashcards+, acquired by Chegg).
⚡ This pairing solved Airy Labs' fatal flaw: Hsu was the technical brain, but Zwick was the operator and commercial thinker. Neither could have built Speak alone.
✓ Together, they snuck into AI classes at UC Berkeley and Stanford — even attending lectures by OpenAI co-founder John Schulman without ever being caught (Forbes).
💬 Most "second-time founder" stories focus on what the person learned. The more honest version: Hsu learned what he wasn't. And he found someone who was.
Decision #2: Pick Korea, Not Silicon Valley
✓ In 2018, Zwick visited Seoul and noticed something strange: entire skyscrapers were filled with English language classrooms, and taxi tops were plastered with ads for English tutoring (Forbes).
✓ Speak chose South Korea as its first market. It became the #1 education brand in the country (ASU GSV Summit).
⚡ Korea wasn't random. It was a calculated bet: extremely high English-learning demand, willingness to pay, and a concentrated market that allowed fast iteration. Accel's Ben Quazzo later called it a deliberate "sandbox" to test the product in one of the world's most competitive language learning markets (Forbes).
💬 Most startups die from "market too big" — they try to win everywhere and win nowhere. Hsu and Zwick picked a single, dense market and went deep. Only after dominating Korea did they expand to Japan, Taiwan, and eventually the U.S.
Decision #3: Bet on Efficacy, Not Gamification
✓ Speak explicitly chose not to gamify. CEO Zwick stated: "When there's a tug of war between gamification and engagement and efficacy, we will pick efficacy 100% of the time" (TechCrunch).
✓ Zwick's assessment of Duolingo: "Mobile games that also teach you language are, at the end of the day, ways to feel less bad about using your phone when you're bored. It's a healthier alternative to Candy Crush" (Forbes).
⚡ This is a product philosophy bet with real consequences: Duolingo's 2024 revenue was $748M. Speak's is $100M+. That's a 7.5x gap. Speak is betting that "actually learning" will eventually beat "feeling like you're learning" — but the jury is still out.
💬 The counter-argument is legitimate: gamification keeps people showing up. If people don't open the app, efficacy is zero. Speak's bet only pays off if the AI tutor is compelling enough to drive retention on its own.
Decision #4: Stack Capital on Hard Metrics
✓ Speak's funding trajectory: Series B $27M (led by OpenAI Startup Fund) → B-2 $16M → B-3 $20M ($500M valuation) → Series C $78M ($1B valuation, led by Accel). Total raised: $162M (TechCrunch, Bloomberg).
✓ 15M+ downloads, 10-20 min average daily usage, 500+ enterprise clients including KPMG and HD Hyundai (Forbes, TechCrunch).
💬 After Airy Labs, Hsu knew what "capital without traction" looks like. This time, every fundraise had hard numbers behind it — not just a prodigy story.
FORKED Scorecard: The Second-Chance Founder Playbook
Use this to evaluate whether a founder's second startup has actually learned from the first.
| Dimension | Score Range | Andrew Hsu's Score |
|---|---|---|
| Co-founder Fit (complementary skills, not overlap) | 1-10 | 9 — CTO + operator/CEO, Thiel Fellows who'd known each other 4 years |
| Market Density (concentrated demand, fast feedback) | 1-10 | 9 — Korea: highest willingness-to-pay for English learning |
| Product Conviction (clear stance, not "we do everything") | 1-10 | 8 — "Efficacy > gamification" is a real position, not marketing |
| Capital Discipline (metrics before money) | 1-10 | 8 — Each round backed by revenue/usage milestones |
| Failure Integration (specific lessons applied, not just "I learned a lot") | 1-10 | 7 — Solved the management problem, but personal narrative still leans on prodigy story |
| Total | /50 | 41/50 |
💬 The biggest gap: Speak's success is deeply tied to AI model quality (especially OpenAI). If OpenAI's models degrade or pricing spikes, Speak's core product degrades with it. That's a dependency, not a moat.
The Contrarian Insight
Everyone talks about Hsu as a "child prodigy." That framing hides the actual lesson.
The real pattern is: the "genius" label attracts capital and press, but repels the organizational humility you need to build a company. Airy Labs got funded because of the prodigy story. It failed because the prodigy story gave everyone — founder, family, investors — permission to skip the boring parts: hiring experienced operators, setting boundaries with family, building feedback loops.
Speak succeeded not because Hsu got smarter. It succeeded because he found a partner, picked a specific market, and chose a product philosophy that forced discipline.
The Hidden Cost
✓ Speak is built primarily on OpenAI's models (TechCrunch). OpenAI is both an investor and a technology supplier.
⚡ This is a dual dependency: if OpenAI raises API prices, Speak's unit economics change overnight. If a competitor gets better model access (say, Duolingo builds in-house), Speak's differentiation shrinks.
💬 And there's the Duolingo elephant: $748M revenue vs. $100M. Speak's "we're more serious" positioning is intellectually appealing, but 7.5x revenue gaps don't close on philosophy alone. Speak needs to prove that "efficacy" converts to retention and word-of-mouth at Duolingo scale — and they haven't yet.
Also unaddressed: ✓ Hsu regrets not learning Chinese growing up — the language of his parents (CommonWealth Magazine). It's a genuine personal motivation. But personal pain doesn't automatically translate to market insight. Speak still hasn't launched Mandarin as a target language.
What Would You Do?
Your first startup failed publicly. You're 24 with a "prodigy" label that's now more burden than asset. What's your next move?
FAQ
Q: What is Speak?
A: An AI-powered language learning app focused on conversational fluency — getting users to speak out loud, not just memorize vocabulary. It uses OpenAI's models for speech recognition and AI tutoring.
Q: How is Speak different from Duolingo?
A: Duolingo emphasizes gamification (streaks, points, leaderboards). Speak emphasizes spoken practice and pronunciation correction with AI. Speak's founders explicitly describe Duolingo as "a healthier Candy Crush."
Q: How much has Speak raised?
A: $162M total. Series C of $78M at a $1B valuation in December 2024, led by Accel with OpenAI Startup Fund participation.
Q: What happened to Airy Labs?
A: Hsu's first startup (2011), an educational gaming company funded by Google Ventures. It laid off most staff within a year amid management controversies and eventually shut down.
Q: Why did Speak launch in Korea first?
A: Korea has extremely high demand for English learning, willingness to pay, and concentrated market density. It served as a "sandbox" to achieve product-market fit before expanding.
Q: Is Speak profitable?
A: Speak has crossed $100M in annualized revenue but hasn't publicly disclosed profitability.
Q: What languages does Speak support?
A: Six: English, Korean, Spanish, Japanese, French, and Italian. Notably, Mandarin Chinese is not yet available.
Q: What is the Thiel Fellowship?
A: A program created by Peter Thiel that gives $100,000 to people under 20 to drop out of school and build companies. Both Speak co-founders (Hsu and Zwick) were Thiel Fellows.
Related Reads
- From Snowboards to $292B: Shopify's 5 Pivot Decisions — Another founder who pivoted from a failed first idea to a platform giant
- Canva's Melanie Perkins Heard "No" 100 Times — The persistence playbook for founders who keep getting rejected
- Brian Chesky's Founder Mode Isn't Micromanagement — When a founder goes back to the details after scaling too fast
Sources
- Wikipedia — 徐安廬
- TechCrunch — Big Cuts at Airy Labs (2012)
- Forbes — Andrew Hsu Raises $1.5M (2011)
- Forbes — What Happened to Thiel Fellows (2017)
- Hack Education — Airy Labs Failure (2012)
- Forbes — How Speak Takes On Duolingo (2025)
- TechCrunch — Speak Raises $78M at $1B (2024)
- Bloomberg — Speak Doubles to $1B (2024)
- ASU GSV Summit — Andrew Hsu Speaker Profile
- CommonWealth Magazine — Andrew Hsu Regrets Not Learning Chinese (2024)
- Business Insider — Airy Labs (2011)
- Y Combinator — Speak Company Profile
🔀 What Would You Do?
Your first startup just failed publicly. What's your next move?
🔀 One CEO decision, dissected weekly
No fluff. No hero worship. Just frameworks, data, and a decision model you can steal.
Unsubscribe anytime. We don't sell your data.
Disclaimer
This article was researched and written with AI assistance by the FORKED editorial team, with human review. Markers: ✓ = verified fact, ⚡ = reasoned inference, 💬 = editorial opinion. While we strive for accuracy, information may contain gaps or errors. This is not investment, legal, or business advice.
All analysis is based on publicly available information. If you spot a factual error or have copyright concerns, please use the report button below or contact us.