Ricursive Intelligence Bets on AI-Designed Chips to Ease Compute Demand

Summary

Ricursive Intelligence founders Anna Goldie and Azalia Mirhoseini say more efficient chip design could help address the expected growth in demand for computing. At TechCrunch Disrupt, they described an AI system that designs chips, learns from each iteration, and improves later designs, with the aim of automating and accelerating a process that can take months or years. Goldie said demand for compute could be at least 10,000 times higher within five years, arguing that efficiency improvements may be preferable to building 10,000 times more data centers. The founders said AI can find layouts that challenge conventional engineering assumptions. Their earlier AlphaChip system produced a circular, donut-shaped arrangement of components that initially surprised engineers but worked successfully. Ricursive is also optimizing entire chips rather than dividing them into many smaller blocks, allowing it to explore complex patterns such as wave-like designs. The founders believe faster design could make custom chips financially practical for applications including space, augmented and virtual reality, and biological computing, while improving energy and manufacturing efficiency. They present AI as handling more of the execution so engineers can focus on applications and new computing ideas.