Ricursive Intelligence Bets on AI-Designed Chips

The AI industry's appetite for compute keeps growing, and most proposed fixes involve building more data centers. Ricursive Intelligence wants to approach the problem from a different direction: make the chips themselves better, and let AI do the design work.

Co-founders Anna Goldie and Azalia Mirhoseini presented the idea on stage at TechCrunch Disrupt on Wednesday. Their startup is building AI that designs chips, learns from each attempt, and uses those lessons to improve the next design. The goal is to automate chip design and make it much faster.

The problem: demand outrunning supply

Goldie described the scale of the challenge in stark terms. "There is this 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 said.

Her conclusion is that building infrastructure alone won't be enough. "So we don't want to build 10,000 times more data centers," she added. Instead, she argued for chips that are more efficient and make better use of data center bandwidth, wafers and fabs (the fabrication plants where chips are manufactured). She also dismissed more exotic options, joking that even data centers in space would come at a cost: "you won't see the sun anymore."

Unusual layouts that work

According to the founders, AI is already questioning long-held assumptions in chip design by finding layouts human engineers would not normally try. Their earlier system, AlphaChip, produced a circular, "donut-shaped" arrangement of components. Engineers were surprised at first, but the design turned out to work.

Ricursive's current approach goes further. Traditional chip design splits a chip into dozens of smaller blocks that are handled separately. Ricursive instead optimizes the entire chip at once. The founders say this lets their AI create complex patterns, including wave-like designs, and they plan to share more details later.

Humans set direction, AI does the work

Some people find the idea of AI designing its own hardware unsettling. Goldie takes a different view. She sees AI taking over execution so people can focus on ideas, applications and new uses for computing.

"I think the way everything is shifting is that humans bring their great ideas and creative thoughts, and execution is handled by AI," she said. She added that this shift is happening in chip design, software development and other fields.

Faster design, more specialized chips

The strongest argument the founders made is about economics. When a new chip takes six months to two years to design, companies have to be very selective about which chips they build.

"But if you can design a chip in like a week," Goldie said, it becomes financially sensible to make custom chips for many applications, including space, AR and VR, and biological computing. She added that this would also improve energy efficiency.

Our Take

Ricursive's pitch is a reminder that the compute shortage isn't only about money for more buildings and GPUs. Much of the current conversation focuses on capacity, from neocloud providers like Lambda raising capital ahead of an IPO to ever larger infrastructure pledges. Efficiency gets less attention, and it is worth taking seriously.

The idea also fits a broader trend of AI moving into engineering work itself, similar to startups such as Vinci4D building AI for engineering simulation. Still, many claims remain untested. The 10,000x demand figure is a forecast, and a one-week design cycle is a goal, not a demonstrated result. Two things to watch: whether the promised details on wave-like layouts show measurable gains, and whether chipmakers actually take whole-chip AI optimization into production.