## The AlphaChip Era Anna Goldie and Azalia Mirhoseini spent five years at Google DeepMind solving a problem most people didn't know existed: chip design is slow. Traditional semiconductor architecture takes 2-3 years from concept to production. Human engineers manually optimize layouts, run simulations, iterate. It's painstaking, expensive, and bottlenecked by talent. AlphaChip changed that. Their AI could design chip layouts in hours that matched or exceeded human performance. Google deployed it internally. The research paper made waves. Then they left. ## The Departure The exit wasn't about compensation. It was about velocity. Google moves at Google pace — committees, reviews, alignment meetings. The founders had a question they couldn't answer inside a large organization: what happens when the AI that designs chips also designs better versions of itself? Recursive self-improvement requires speed. Each iteration teaches the next. Waiting for approval breaks the loop. Two months after leaving, Ricursive Intelligence hit $4 billion valuation. $300 million Series A led by Lightspeed. The bet isn't on a chip. It's on a feedback loop. ## The Vision Their first product is the Temporal Processing Unit — a chip architecture optimized for AI inference, designed entirely by AI. But the product isn't really the chip. It's the process. Each generation of TPU gets designed by the previous generation's AI. The architecture improves itself. Goldie and Mirhoseini become reviewers, not architects. If they're right, the semiconductor industry reorganizes around whoever has the best recursive design loop. If they're wrong, they've raised $300 million to find out.