How Ricursive Intelligence raised $335M at a $4B valuation in 4 months

The co-founders of startup Ricursive Intelligence always looked like the kind of pair who would end up building something together.

CEO Anna Goldie and CTO Azalia Mirhoseini are well-known enough in the AI world that they were among the engineers who, as Goldie put it with a laugh, "got those weird emails from Zuckerberg making crazy offers to us." They didn't accept. Both worked at Google Brain, and both were early employees at Anthropic.

They became widely recognised at Google for building Alpha Chip—an AI system that could produce high-quality chip layouts in hours, a task that typically takes human designers a year or longer. The project contributed to the design of three generations of Google's Tensor Processing Units.

That background helps explain why Ricursive moved so fast. Just four months after launching, the company announced last month that it had closed a $300 million Series A at a $4 billion valuation led by Lightspeed, only a couple of months after raising a $35 million seed round led by Sequoia. In total, that's $335 million in four months.

Ricursive isn't making chips. It's building AI software that helps others design chips. That position sets it apart from most AI chip startups: it's not aiming to become the next Nvidia. NVIDIA is actually an investor. The company's intended customers are the GPU heavyweight, AMD, Intel, and essentially every major chip maker.

"We want to enable any chip, like a custom chip or a more traditional chip, any chip, to be built in an automated and very accelerated way. We're using AI to do that," Mirhoseini said.

Their connection goes back to Stanford, where Goldie earned her PhD, and Mirhoseini taught computer science. From there, the timing of their careers remained almost perfectly synchronised. "We started at Google Brain on the same day. We left Google Brain on the same day. We joined Anthropic on the same day. We left Anthropic on the same day. We rejoined Google the same day and left the same day again. Then we started this company together on the same day," Goldie recalled.

At Google, they were close enough to train together; both were fans of circuit workouts. Jeff Dean, the celebrated Google engineer who collaborated with them, enjoyed the coincidence so much that he jokingly labelled their Alpha Chip effort "chip circuit training," a reference to their shared fitness routine. Inside the company, they were also known by a simple nickname: A&A.

Alpha Chip brought them major attention — and also controversy. In 2022, Wired reported that a Google colleague was fired after spending years attempting to undermine A&A and their work on chip design, even though their work was used to help build some of Google's most critical, high-stakes AI chips.

What they demonstrated at Google Brain has become the foundation for what Ricursive is now pursuing: using AI to compress chip design timelines dramatically.

Designing chips is hard.

The challenge begins with what chips actually are. A modern processor can contain millions to billions of logic gate components packed onto a silicon wafer. Human designers can spend a year or more deciding where all those pieces should go to hit performance targets, keep power usage efficient, and satisfy countless other design constraints. Getting the placement right for unimaginably small components—and doing it at scale—is, unsurprisingly, extremely difficult.

Alpha Chip, Goldie said, "could generate a very high-quality layout in, like, six hours. And the cool thing about this approach was that it actually learns from experience."

Their approach relies on a "reward signal" that scores the quality of a design. The system then uses that score to "update the parameters of its deep neural network to get better," Goldie explained. After running through thousands of designs, the agent became highly capable — and the founders say it also became faster as it learned.

Ricursive plans to push the concept further. The chip-design AI platform the startup is building will "learn across different chips," Goldie said, meaning eeachnew project should strengthen the system for the next. 

The platform also incorporates LLMs and is designed to cover the full pipeline—from component placement to design verification. Any electronics company that needs chips is positioned as a potential customer.

If the platform performs as the founders expect, Ricursive could become tied to one of the field's biggest ambitions: achieving artificial general intelligence. Their longer-term vision is to help design the chips used for AI itself— enabling AI systems to help create the hardware that runs them.

"Chips are the fuel for AI," Goldie said. "I think by building more powerful chips, that's the best way to advance that frontier."

Mirhoseini argues that today's slow chip-design cycles are effectively limiting how fast AI can advance. "We think we can also enable this fast co-evolution of the models and the chips that basically power them," she said, making it possible for AI capabilities to improve more quickly.

If the idea of AI designing its own "brains" at increasing speed triggers Skynet-style fears, the founders point to a more immediate upside they believe is both practical and likely: efficiency.

If AI labs can design far more efficient chips — and eventually optimise more of the underlying hardware stack — growth in AI won't need to consume such a large share of global resources.

"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," Goldie said.

Ricursive won't identify its early customers yet. Still, the founders say they've heard from essentially every major chip company you'd expect — and, unsurprisingly, they say they can be selective about their first development partners as well.