Ricursive Intelligence is betting on AI-designed chips to help solve the compute crisis

Ricursive Intelligence’s founders, Anna Goldie and Azalia Mirhoseini, are betting that AI can help ease the compute crunch by redesigning the chips that power it. They said at TechCrunch Disrupt on Wednesday that their startup is building AI that can design chips, learn from each pass, and improve the next one. The pitch is simple: if demand for compute is about to explode, the answer may not be building a lot more data centers. It may be making each one work harder.

Goldie said the team expects at least 10,000x more demand for compute in the next five years, and called that estimate potentially conservative. She argued that the industry should not try to answer that by building 10,000 times more data centers. Instead, the focus should be on chips that use bandwidth, wafers, and fabs more efficiently. Her line about data centers in space getting in the way of the sun landed like a joke, but the point was serious: brute force has limits.

The company’s argument is that AI can find chip layouts people would not have designed on their own. The founders pointed to AlphaChip, the system they previously developed, which came up with a circular, donut-shaped layout that surprised engineers before proving successful. They said Ricursive is optimizing whole chips rather than splitting them into dozens of smaller blocks, which lets the system produce more unusual patterns, including wave-like designs.

And that speed matters. Goldie said chip design can take six months, a year, or even two years, which forces companies to be picky about what they attempt. If AI can cut that down to about a week, she argued, then custom chips become realistic for far more uses, including space, AR, VR, and biological computing. That is the real promise here: not just faster chip design, but more of it, for more things.