Ricursive Intelligence wants AI to design the chips that AI runs on. TechCrunch AI reports that founders Anna Goldie and Azalia Mirhoseini explained their plan on stage at TechCrunch Disrupt on Wednesday. They’re building AI that designs chips, learns from each design and uses what it learned to make the next one better. Their goal is to automate chip design and make it much faster. They’re betting that more efficient chips, rather than just more data centers, will help ease the compute shortage.
Key points
- The pitch: AI designs entire chips, improves with every pass and cuts design time from months or years to about a week.
- The problem it targets: Goldie expects demand for compute to grow by at least 10,000x in five years, and she says that figure may be “somewhat potentially conservative.”
- The track record: The founders previously built AlphaChip, an AI system whose unusual layouts held up in real chips.
- What’s new: Instead of splitting a chip into dozens of smaller blocks the usual way, they optimize the whole chip at once.
- What comes next: If design gets cheap and fast, custom chips for niche markets like space, AR/VR and biological computing start to make financial sense.
Why this matters
Pretty much every conversation about AI capacity ends up back at the same answer: build more data centers, find more power, buy more GPUs. Goldie argues that approach can’t keep up. “So we don’t want to build 10,000 times more data centers,” she said. She wants chips that “use the bandwidth in those data centers more efficiently, that use the wafers and fabs more efficiently.”
She also dismissed the more exotic ideas: “even if we make [data centers] in space or something, I don’t know, you won’t see the sun anymore.”
What stands out to me is the economics. Chip design is slow and expensive today, so companies only build chips for huge markets. Goldie spelled it out: “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.” Cut that to a week and suddenly lots of specialized chips are worth building.
The AlphaChip connection
These founders have done this before. Their earlier system, AlphaChip, came up with a circular, “donut-shaped” layout of chip components that surprised engineers at first and then worked. That’s the core argument for AI in chip design: it isn’t stuck with the habits and rules of thumb that human designers have built up over decades.
Ricursive is taking that idea further. Engineers normally split a chip into many separate blocks and lay out each one on its own. Ricursive optimizes the whole chip in one go. According to the founders, that lets the AI find intricate patterns, including wave-like designs, and they say they’ll share more about those later.
The human role shifts
Some people will find the idea of AI designing its own hardware unsettling. Goldie sees it as a split of responsibilities.
“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. “And this is happening in chip design, software development, and other places.”
That matches what’s already happening in software, where AI coding agents handle more of the execution while engineers set direction and review the results. Chip design looks like the next field where that happens.
What to watch
Here’s what readers should keep an eye on:
- Proof at scale. AlphaChip showed the idea works. Ricursive now has to show that whole-chip optimization produces chips that beat human designs on performance, power and area.
- The wave-like designs. The founders promised more details. That’s when we’ll find out whether the approach is a real step forward or a small improvement.
- Design cycle times. The “chip in a week” claim is the one that matters most. If it holds up, the market for custom chips could change.
- Industry adoption. Big chip companies and the established design-software vendors will either partner with startups like this, buy them or build their own versions.
My take
The compute shortage won’t be solved by one fix. Power, fab capacity and chip efficiency all matter. Still, Ricursive is going after a real bottleneck that gets little attention. Everyone talks about how many chips we can make, and far fewer people ask how good those chips could be. If AI can squeeze more performance out of every wafer and every watt, that buys the industry time while it builds out more capacity.
The founders say more technical details are coming. The full interview is available from TechCrunch AI.