Ricursive Intelligence’s founders outlined an AI chip-design system that learns from successive designs, aiming to speed development and improve hardware efficiency.
At TechCrunch Disrupt on October 14, 2026, Ricursive Intelligence co-founders Anna Goldie and Azalia Mirhoseini described their effort to use AI to automate and accelerate chip design. The startup’s system is intended to create designs, learn from each iteration, and apply those lessons to subsequent chips.
The founders said their approach focuses on optimizing entire chips rather than splitting the work into dozens of smaller blocks, a conventional design method. They said this lets the AI explore arrangements that may look unfamiliar to engineers, including intricate, wave-like patterns. As an earlier example, they pointed to AlphaChip, an AI system they previously developed that generated a circular arrangement of chip components. The layout initially surprised engineers but, according to the founders, proved successful.
Goldie argued that more efficient designs could help address rising compute needs without relying only on building more data centers. She cited an estimate that demand for computing could grow by at least 10,000 times over the next five years, while describing that projection as potentially conservative. Her point was that efficiency across chips, data-center bandwidth, wafers, and fabrication capacity could help meet demand.
The company’s broader proposition is that faster design cycles could make it practical to create custom chips for a wider range of uses. Goldie contrasted current chip-design timelines—which she said can run from six months to two years—with a possible process measured in about a week. She named space, augmented and virtual reality, and biological computing as areas where specialized chips might be useful, provided the economics work.
Those timelines and potential applications are the founders’ projections, not confirmed product results or availability commitments. Ricursive says it plans to share more about some of its design patterns in the future; the Disrupt discussion did not specify a release schedule.