Ricursive Intelligence co-founders Anna Goldie and Azalia Mirhoseini spoke at TechCrunch Disrupt on Wednesday about their startup’s work: they are building AI that can design semiconductor chips, learn from each iteration, and improve subsequent designs with the aim of automating and accelerating the chip design process.
On stage Goldie said there is an understanding that “in the next five years we are going to have at least 10,000x more demand for compute, and that’s somewhat potentially conservative.” She argued it is impractical to build 10,000 times more data centers, so a complementary approach is to make chips that are more efficient and that use data center bandwidth, wafers and fabs more effectively.
Contrary to fears that AI designing chips might replace humans, Goldie framed the technology as enabling people to focus on developing applications and exploring new possibilities while AI handles execution. “I think the way everything is shifting is that humans come bring their great ideas and creative thoughts, and execution is handled by AI,” she said.
How the AI approach differs from traditional methods
The founders said the AI challenges conventional chip design by discovering unconventional layouts that can outperform standard designs. They cited a prior system, AlphaChip, which produced a circular, “donut-shaped” arrangement of components that initially surprised engineers but ultimately proved successful.
Ricursive Intelligence’s method optimizes entire chips rather than partitioning them into dozens of smaller blocks as is commonly done. This whole-chip optimization allows the AI to create intricate patterns, including wave-like layouts, which the company says it will reveal more about in the future.
Faster design cycles enable more and different chips
The founders also argued that AI-designed chips make it feasible to create many more types of chips. Goldie noted that when designing a new chip takes six months to two years, teams must be extremely selective about which chips to build. But if a chip can be designed in about a week, it becomes viable to manufacture custom chips for diverse applications — in space, AR, VR, biological computing and more — while also improving energy efficiency.
Ricursive Intelligence says AI not only speeds up design but can enable chip topologies that may not have been practical with traditional workflows.
Why this matters
The company’s approach directly addresses projected exponential growth in compute demand without relying solely on scaling data-center capacity. AI-based whole-chip optimization approaches could yield better energy and manufacturing efficiency and faster iteration cycles, which are important for supporting a broader range of future applications. Ricursive Intelligence has indicated it will share additional technical details at a later time.
