THE CRUNCH
Ricursive Intelligence, a startup founded by Anna Goldie and Azalia Mirhoseini, is developing AI that designs computer chips, learns from each iteration and improves the next design. Speaking at TechCrunch Disrupt, the founders argued that far more efficient chip design is a better answer to exploding compute demand than simply building more data centres. Goldie cited projections of at least 10,000 times more demand for compute within five years, calling that figure potentially conservative, and said the goal is to use bandwidth, wafers and fabrication capacity more efficiently rather than expand infrastructure to match.
The approach differs from traditional chip design, which partitions a chip into dozens of smaller blocks. Instead, the AI optimises the entire chip at once, producing unconventional layouts. Their earlier system, AlphaChip, generated a circular, donut-shaped arrangement of components that initially surprised engineers but ultimately worked, and the pair said they plan to share more about wave-like designs in future.
Goldie also argued that speeding up design changes what chips are worth building. If a design takes six months to two years, teams must be highly selective, but a week-long design cycle could make custom chips viable for niche applications in areas like space, AR, VR and biological computing, with better energy efficiency as a side benefit. On fears of AI designing its own chips, she framed it as humans supplying the ideas while AI handles execution.