Anna Goldie and Azalia Mirhoseini took the stage at TechCrunch Disrupt on Wednesday to make the case that the fastest way out of the compute shortage is to let artificial intelligence design the chips themselves. The pair, who lead the startup Ricursive Intelligence, argued that a self-improving loop between AI and hardware is the only realistic answer to demand that is outpacing what today's factories and data centers can support.
Goldie told the audience she expects compute demand to grow at least 10,000 times over the next five years, a figure she described as possibly conservative. Building 10,000 times more data centers is not the answer, she said — the industry needs chips that use bandwidth, wafers, and fabrication capacity far more efficiently.
Unconventional layouts, whole-chip optimization
The founders say AI is already challenging traditional design habits by discovering layouts no human team would try. Goldie pointed to AlphaChip, the AI layout system the pair previously built, which once produced a circular, donut-shaped arrangement of chip components that surprised engineers but worked. At Ricursive, they are now optimizing entire chips at once rather than splitting them into dozens of smaller blocks, an approach that lets the AI generate intricate wave-like patterns they plan to describe in more detail later.
From years of design to a week
The bigger promise is speed. Goldie said designing a chip in about a week — instead of six months to two years — would change the economics of custom silicon. Chips for space, AR and VR, and biological computing could suddenly be worth building, she said, and the resulting hardware would be more energy efficient.
Why it matters
Goldie and Mirhoseini are not newcomers to the idea. Before founding Ricursive Intelligence, they co-led AlphaChip at Google, a system that generated chip layouts in hours instead of months and fed multiple generations of Google's Tensor Processing Units. The open question now is whether that approach works beyond one company's walls — for outside chipmakers, at commercial scale, and on fabricated silicon rather than faster layouts alone.
Sources:
- Ricursive Intelligence is betting on AI-designed chips to help solve the compute crisis | TechCrunch
- When Will AI Start Designing Its Own Chips? | AI Daily Post
Goldie told the audience she expects compute demand to grow at least 10,000 times over the next five years, a figure she described as possibly conservative. Building 10,000 times more data centers is not the answer, she said — the industry needs chips that use bandwidth, wafers, and fabrication capacity far more efficiently.
Unconventional layouts, whole-chip optimization
The founders say AI is already challenging traditional design habits by discovering layouts no human team would try. Goldie pointed to AlphaChip, the AI layout system the pair previously built, which once produced a circular, donut-shaped arrangement of chip components that surprised engineers but worked. At Ricursive, they are now optimizing entire chips at once rather than splitting them into dozens of smaller blocks, an approach that lets the AI generate intricate wave-like patterns they plan to describe in more detail later.
From years of design to a week
The bigger promise is speed. Goldie said designing a chip in about a week — instead of six months to two years — would change the economics of custom silicon. Chips for space, AR and VR, and biological computing could suddenly be worth building, she said, and the resulting hardware would be more energy efficient.
Why it matters
Goldie and Mirhoseini are not newcomers to the idea. Before founding Ricursive Intelligence, they co-led AlphaChip at Google, a system that generated chip layouts in hours instead of months and fed multiple generations of Google's Tensor Processing Units. The open question now is whether that approach works beyond one company's walls — for outside chipmakers, at commercial scale, and on fabricated silicon rather than faster layouts alone.
Sources:
- Ricursive Intelligence is betting on AI-designed chips to help solve the compute crisis | TechCrunch
- When Will AI Start Designing Its Own Chips? | AI Daily Post
