.jpg)
The Idea: In early 2023, Alex Atallah had just walked away from OpenSea - the NFT marketplace he co-founded and took to a $13.3B valuation as CTO - to "build something from zero to one." He spent the gap asking one question: would AI inference end up winner-take-all? Two things convinced him it wouldn't. Meta released Llama 1, and Stanford's Alpaca experiment showed you could distil a genuinely capable model for under $600. If anyone could make a decent model, there wouldn't be one model - there'd be hundreds, each with different pricing, latency, and moderation rules, and no sane way for a developer to choose between them. He started with Window AI, an open-source Chrome extension that let you pick which LLM a website talked to. Then he and co-founder Louis Vichy turned it into something bigger: one API, every model. This was OpenRouter.
The Execution:
The Lesson: Atallah has now built the same company twice. OpenSea sat between people making digital assets and people buying them. OpenRouter sits between people making models and people burning tokens. Neither was the exciting part of its market - the exciting part was CryptoKitties and GPT-4, and he built the plumbing underneath both. A neutral tollbooth between every app and every model vendor is precisely the asset a giant buys defensively, and that's the real tell: infrastructure gets acquired at a premium exactly because it's boring, load-bearing, and impossible to rip out. The best founders don't chase the trend. They own the road it drives on.