Ricursive Intelligence founders Anna Goldie and Azalia Mirhoseini took the stage at TechCrunch Disrupt on Wednesday to argue that AI which designs its own chips, learns from each iteration and improves the next design could help solve the compute crisis, rather than simply building 10,000 times more data centers.
The startup is developing AI that automates and accelerates the chip design process. Goldie told the audience that the industry faces at least a 10,000-fold increase in compute demand over the next five years, a projection she called "somewhat potentially conservative."
"So we don't want to build 10,000 times more data centers," Goldie said. "We think that we need to help by making chips that are more efficient, that use the bandwidth in those data centers more efficiently, that use the wafers and fabs more efficiently, because even if we make [data centers] in space or something, I don't know, you won't see the sun anymore."
Goldie acknowledged that some people are wary of the idea of AI designing its own chips, but argued the shift will let humans focus on ideas and creative work while AI handles execution. "And this is happening in chip design, software development, and other places," she said. "And that's something that recursive does for chip design."
The founders said AI is challenging traditional chip design methods, discovering unconventional layouts that can outperform traditional ones. The company previously raised US$335 million at a US$4 billion valuation in four months, according to TechCrunch's earlier reporting.
Analysis
Why This Matters
- The founders project compute demand will grow at least 10,000-fold over five years, a figure Goldie called "somewhat potentially conservative"
- Ricursive's counter-proposal is efficiency rather than expansion: chips and data centers that extract more work from the same wafers, bandwidth and power
- If AI-designed silicon delivers, it could ease a major bottleneck in the AI industry and change who gets to design advanced chips
Background
Chip design is among the most demanding engineering disciplines in existence. Producing a modern processor requires teams of specialists working for years, with the physical layout of billions of components determining how fast and how efficiently a chip runs. As AI models have grown, their appetite for compute has collided with the limits of fabrication capacity, energy supply and data center construction, pushing chip efficiency to the centre of industry strategy. Using machine learning to assist or automate parts of chip design has been an active area of exploration, but automating the full design loop remains largely unproven in production.
Key Perspectives
Anna Goldie and Azalia Mirhoseini: They argue the answer to surging compute demand is not 10,000 times more data centers but more efficient chips that use bandwidth, wafers and fabs better. Goldie says AI will take over execution in domains such as chip design while humans supply ideas and creative direction. She acknowledged some people are unsettled by the prospect of AI designing its own chips.
Practitioners of AI-assisted design: The founders point to examples in which AI has found unconventional chip layouts that beat traditional designs, evidence that the approach can challenge established methods.
Sceptics: The open question is whether AI-generated layouts can be trusted at scale. Chips must be manufacturable, reliable and verifiable, and design errors are enormously expensive. The 10,000x demand projection is also an estimate, not a certainty.
What to Watch
- Whether Ricursive moves from demonstrations to designs that reach production silicon
- How its AI compares with established electronic design automation tools and human experts on real chip tape-outs
- Further public milestones from the startup after its Disrupt appearance, including any product or partnership announcements