When AI Designs Its Own Brain Chips

$4 Billion in Four Months

Anna Goldie and Azalia Mirhoseini have synchronized careers down to the day. Same day joining Google Brain. Same day leaving. Same day at Anthropic. Same day out. Even their workout routine—circuit training—became the inspiration for their breakthrough Alpha Chip project's internal nickname: "chip circuit training."

Now, just four months after launching Ricursive Intelligence, this Google alum duo has raised a $300 million Series A at a $4 billion valuation. Their pitch? AI that can design computer chips in 6 hours instead of the year-plus it takes human experts.

But here's the twist: they're not trying to compete with Nvidia. Nvidia is actually an investor. So are AMD and Intel. Because Ricursive isn't building chips—they're building the AI that designs chips for everyone else.

The Million-Component Problem

Designing a computer chip means placing millions to billions of logic gates on a silicon wafer with microscopic precision. It's like solving a jigsaw puzzle where every piece affects every other piece, and getting it wrong means your chip doesn't work, overheats, or burns through power.

Human designers spend over a year on this process, carefully optimizing for performance, power efficiency, and manufacturing constraints. Ricursive's AI does it in hours, and—crucially—it gets better with each design.

"The cool thing about this approach was that it actually learns from experience," Goldie explains. Their system uses reward signals to rate design quality, then updates its neural network parameters to improve. After thousands of designs, it doesn't just get good—it gets fast.

Beyond Speed: The 10x Vision

Sponsored

Advertise with Us

[email protected]

Sponsored

The immediate benefit is obvious: faster time-to-market for chip companies. But Goldie and Mirhoseini see a bigger picture. "We could design a computer architecture that's uniquely suited to that model, and we could achieve almost a 10x improvement in performance per total cost of ownership."

This isn't just about making existing chips faster. It's about AI designing chips specifically optimized for AI workloads—essentially, AI designing its own brain. As AI models evolve, the chips powering them could co-evolve in real-time.

"Chips are the fuel for AI," Goldie says. "By building more powerful chips, that's the best way to advance that frontier."

The Efficiency Paradox

Here's where it gets interesting for the broader AI debate. Current AI development consumes massive amounts of energy and computing resources. But if AI can design chips 10x more efficient for specific AI workloads, the resource consumption problem might solve itself.

Mirhoseini puts it this way: "We think we can enable this fast co-evolution of the models and the chips that basically power them." Instead of AI labs being constrained by existing hardware, they could have custom silicon designed for their exact needs.

For investors and tech executives, this represents a fundamental shift in the AI value chain. Instead of competing for limited GPU capacity from Nvidia, companies could potentially have AI-designed, custom chips that give them unique advantages.

The A&A Controversy

The duo's work hasn't been without controversy. At Google, they were nicknamed "A&A" and their Alpha Chip work was so groundbreaking it reportedly led to a colleague's firing after he spent years trying to discredit their research—even though that same research helped design Google's most important AI chips.

Now, with backing from Lightspeed, Sequoia, and major chip companies, they're taking their vision mainstream. Every major chip maker has apparently reached out, giving them their pick of development partners.