According to Forbes, AI language tutoring app Speak has reached over $100 million in annualized revenue and boasts 15 million downloads since its 2016 founding. The $1 billion-valued startup, founded by former Thiel fellows Connor Zwick and Andrew Hsu, initially focused on South Korea’s massive English learning market before expanding to Japan, Taiwan, and now the U.S. this June. Speak’s voice-based AI coach uses OpenAI models to create role-playing scenarios for practicing real-world conversations in six languages including English, Korean, and Spanish. The company has raised $160 million from investors including Khosla Ventures, Accel, and OpenAI Startup Fund, and recently expanded into enterprise with 500 corporate customers like KPMG and HD Hyundai. Users start free then pay $80 to $200 for premium content, driving the company’s consumer-focused revenue growth.
The Duolingo elephant in the room
Now here’s where it gets interesting. Speak’s $100 million revenue sounds impressive until you realize Duolingo did $724 million last year and expects to hit $1 billion this year. That’s a massive gap. But Zwick makes a compelling case for differentiation – he basically says Duolingo is gaming-adjacent language practice while Speak is serious conversation training. “Mobile games that also teach you language are ways to feel less bad about using your phone when you’re bored,” he told Forbes. Ouch.
And you know what? He might have a point. Duolingo’s strength has always been that gamified experience that keeps you coming back, but how many people actually become fluent speakers versus just getting better at completing lessons? Speak’s entire premise is getting people talking – Zwick claims users speak five to ten times more on his app than competitors. That’s a meaningful difference if conversational fluency is your actual goal.
The South Korea sandbox
What’s really smart about Speak’s journey is how intentionally they approached market selection. Back in 2018, Zwick saw those English classroom-filled skyscrapers in Seoul and realized Silicon Valley wasn’t the target – South Korea’s “obsession” with English was. They used it as what investor Ben Quazzo called a “sandbox” to test in one of the world’s most competitive language markets. Think about that – instead of trying to conquer America first, they dominated a market where demand was insane but existing solutions were outdated and ineffective.
The timing was perfect too. When they started in 2016, nobody cared about AI or language learning apps. Some investors even asked if they were secretly data collectors. Now? AI tutoring is the hottest space in edtech. They were early in a way that’s paying off massively now that the technology has caught up with their vision.
Where the AI tutor wars go next
Here’s the thing – both companies are betting big on AI, but they’re approaching it from completely different angles. Duolingo rolled out AI characters for casual video calls while Speak is building what feels more like a true digital tutor focused on pronunciation and real-world scenarios. The enterprise push is smart too – when consumers start asking employers to pay, you know you’ve got something valuable.
But the real question is whether there’s room for both approaches. I think there might be. Some people want the low-pressure, game-like experience of Duolingo, while others need the serious speaking practice Speak offers. The market’s certainly big enough – language learning is a massive global industry, and AI is making quality instruction accessible in ways we’ve never seen before.
What’s fascinating is how fundamental OpenAI has become to Speak’s growth. The models keep improving, and Speak’s app automatically updates lesson plans and adds features like accent correction that weren’t possible back in 2016. Early investor Sven Strohband put it perfectly: “They were AI-first before it was actually cool to be AI-first.” That early bet is paying off now that everyone’s scrambling to add AI to everything.
