The key facts
- The founders predict demand for computing will grow at least 10,000-fold over the next five years, calling the estimate "potentially conservative."
- The startup optimizes chips as a whole rather than splitting them into blocks, as is traditionally done.
- Fast design (about a week) could make custom chips for AR, VR, space, and biocomputing economically viable.
New Startup from Former AlphaChip Creators
Ricursive Intelligence founders Anna Goldie and Azalia Mirhoseini spoke at the TechCrunch Disrupt conference on Wednesday, presenting a company that develops AI for chip design. The system is meant to create designs, learn from each iteration, and improve subsequent versions, automating and speeding up the development process. The founders previously created the AlphaChip AI system.
Motivation: Growing Demand for Computing
Goldie said from the stage that, according to common belief, demand for computing will grow at least 10,000-fold over the next five years, and that estimate is "potentially conservative." According to her, instead of building 10,000 times more data centers, chips need to be made more efficient — using data center bandwidth, wafers, and fabs more rationally. She added that even if data centers are built in space, "you won't see the sun anymore."
The Role of Humans and AI in Design
Addressing possible concerns about AI designing its own chips, Goldie noted that handing design over to AI will allow people to focus on developing applications and exploring new possibilities. "People bring their great ideas and creative thoughts, and the execution is done by AI," she said, adding that this happens in chip design, software development, and other fields.
Unconventional AI Solutions
The founders said that AI finds unconventional layouts that can outperform traditional ones. AlphaChip, for example, created a circular "donut"-shaped layout of chip components that initially surprised engineers but proved successful. Ricursive Intelligence optimizes chips as a whole rather than splitting them into dozens of blocks as is traditionally done — this allows AI to create complex patterns, including wavy designs, which the founders plan to discuss in more detail later.
Prospects for Custom Chips
According to the founders, AI design will make possible chips that previously could not be created. Goldie explained: if designing a new chip takes six months, a year, or two, you have to be extremely selective about what to design. But if a chip can be designed in about a week, it becomes financially viable to create custom chips for various applications — space, AR, VR, and biological computing — with greater energy efficiency.
