Ricursive Intelligence founders Anna Goldie and Azalia Mirhoseini say AI that designs chips could ease the coming compute crisis. Speaking at TechCrunch Disrupt on Wednesday, the pair described a startup building AI that designs chips, learns from each iteration, and improves later designs, with the aim of automating and speeding up chip design.

Goldie said demand for compute could grow at least 10,000x over five years, a figure she called potentially conservative. Rather than build 10,000 times more data centers, she argued, the industry should make chips more efficient, using data center bandwidth, wafers, and fabs more effectively.

Goldie said AI taking over execution lets humans focus on ideas and applications, a shift she said is happening in chip design and software development. The founders pointed to AlphaChip, their earlier AI system, which produced a donut-shaped arrangement of chip components that surprised engineers but worked.

They said they optimize whole chips instead of partitioning them into dozens of smaller blocks, letting AI produce intricate patterns such as wave-like designs, with more details promised later. Goldie said that if design time drops from months or years to about a week, custom chips become financially viable for applications including space, AR, VR, and biological computing.

Source: TechCrunch