Ricursive Intelligence Bets AI-Designed Chips Can Solve Compute Crisis

By Elena Vasquez

Anna Goldie stood on stage at TechCrunch Disrupt on Wednesday and put a number on the problem: 10,000x. That is the minimum growth in demand for compute she expects over the next five years, and she thinks it might even be "somewhat potentially conservative." Her answer is not 10,000 times more data centers. It is letting AI design the chips themselves.

Goldie and Azalia Mirhoseini are the founders of Ricursive Intelligence, a startup building AI that designs chips, learns from each iteration, and improves its subsequent designs. The goal, they told the Disrupt audience, is to automate and accelerate the entire chip design process.

The stakes are straightforward. Compute is the bottleneck of the AI industry, and the industry's current answer — building more data centers — runs into hard physical and economic limits. Goldie made that point with a joke that landed as a warning: "even if we make [data centers] in space or something, I don't know, you won't see the sun anymore."

Who are the founders, and what have they already built?

The pair bring a track record. They previously developed AlphaChip, an AI system that produced a circular, "donut-shaped" arrangement of chip components. Engineers were surprised by it at first, the founders said, but the layout ultimately proved successful.

That episode is their core argument in miniature. AI, they say, is challenging traditional chip design methods by discovering unconventional layouts that outperform traditional ones. Human designers carry conventions. A reinforcement learning system does not, and the donut shape is what that freedom looks like in silicon.

What does Ricursive's approach change?

The founders say they are optimizing entire chips rather than following the traditional approach of partitioning them into dozens of smaller blocks. That whole-chip framing lets the AI create intricate patterns, including wave-like designs, which the pair said they plan to share more about in the future.

The distinction matters. Partitioning a chip into blocks is how the industry has managed complexity for decades, but it caps how coordinated a design can be. Optimizing the whole chip at once is a harder problem — and it is exactly the kind of problem the founders argue AI is suited to grind through, iteration after iteration, with each design feeding what the next one learns.

The company's name, as Goldie described it on stage, points at the same loop: "that's something that recursive does for chip design."

Why does speed of design matter as much as efficiency?

Efficiency is half of Ricursive's pitch. Speed is the other half, and Goldie framed it around design cycle times that the industry treats as fixed.

"If it takes six months or a year or two years to design a new chip, then you really have to be extremely selective about the chips that you design," Goldie said.

She then flipped the arithmetic: "But if you can design a chip in like a week, and then say for manufacturing, you can design custom chips for all sorts of different types of applications in like space and AR, VR, and you biological computing, and it makes sense financially to do that, and we can be more energy efficient."

In other words, a one-week design cycle does not just accelerate existing chip roadmaps. It changes what chips are worth building at all. Applications Goldie named — space, AR, VR, biological computing — are domains where custom silicon rarely pencils out today because the design cost dwarfs the market.

The claim pairs with the efficiency argument from her opening. Making chips that "use the bandwidth in those data centers more efficiently, that use the wafers and fabs more efficiently," as she put it on stage, attacks the compute shortage from the supply side. Cutting design time to a week attacks it from the demand side, by opening custom silicon to markets that cannot justify a two-year design budget.

Should people be worried about AI designing its own chips?

The image of AI designing the hardware that runs AI unsettles some people. Goldie's response is that this is not a story about displacement but about division of labor.

"I think the way everything is shifting is that humans come bring their great ideas and creative thoughts, and execution is handled by AI," Goldie said. "And this is happening in chip design, software development, and other places."

Her framing extends beyond semiconductors. Chip design is, in her telling, one instance of a broader shift in which creative direction stays human and execution moves to machines. Humans freed from layout work can focus on developing applications and exploring what computing can accomplish, she argued.

What is the bet, really?

Ricursive Intelligence is making a layered wager. The first layer is that demand for compute really will grow at least 10,000x in five years, making current infrastructure strategies untenable. The second is that AI can out-design human engineers at the whole-chip level, as AlphaChip's donut-shaped layout suggested at the block level. The third is that speed compounds: once design cycles shrink from years to roughly a week, entire categories of custom silicon become financially viable.

The founders did not share benchmark figures, customer names, or funding details on stage, and the wave-like whole-chip designs remain undisclosed for now. What they did stake out is a direction — and a pair of numbers to hold them to: 10,000x demand growth in five years, and a chip designed in a week. If Ricursive can hit the second number, the first one starts to look a lot less frightening.