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

The co-founders of startup Ricursive Intelligence seemed destined to become co-founders.

Anna Goldie, CEO, and Azalia Mirhoseini, CTO, are so well known in the AI ​​community that they were among the AI ​​engineers who “got those weird emails from Zuckerberg with crazy offers to us,” Goldie told TechCrunch with a chuckle. (They didn’t take up the offers.) The pair worked together at Google Brain and were early employees at Anthropic.

They won high praise at Google for creating the Alpha Chip – an AI tool that could generate solid chip layouts in a matter of hours – a process that would normally take human designers a year or more. The tool helped design three generations of Google’s Tensor Processing Units.

That pedigree explains why, just four months after launching Ricursive, they announced a $300 million Series A round last month at a $4 billion valuation led by Lightspeed, just a few months after raising a $35 million seed round led by Sequoia.

Ricursive builds AI tools that design chips, not the chips themselves. That makes them fundamentally different from almost every other AI chip startup: they are not a wannabe Nvidia competitor. Nvidia is actually an investor. The GPU giant, together with AMD, Intel and all other chip manufacturers, is the target group of the startup.

“We want to make sure that any chip, like a custom chip or a more traditional chip, any kind of chip, can be built in an automated and highly accelerated way. We use AI to do that,” Mirhoseini told TechCrunch.

Their paths first crossed at Stanford, where Goldie received her PhD while Mirhoseini was teaching computer science classes. Since then their career has been gaining momentum. “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 joined Google again on the same day, and then we left Google on the same day. Then we started this company together on the same day,” says Goldie.

During their time at Google, the colleagues were so close that they even trained together, both enjoying circuit training. The pun was not lost on Jeff Dean, the famous Google engineer who was their employee. He called their Alpha Chip project “chip circuit training” – a play on their shared workout routine. The pair also received an internal nickname: A&A.

The Alpha Chip earned them attention in the industry, but also caused controversy. In 2022, one of their colleagues at Google was fired, Wired reportedafter spending years trying to discredit A&A and their chip work, even though that work was used to produce some of Google’s most important AI chips.

Their Alpha Chip project at Google Brain proved the concept that would become Ricursive: using AI to dramatically speed up chip design.

Designing chips is difficult

The problem is that computer chips have millions to billions of logic gate components integrated onto their silicon wafer. Human designers can spend a year or more putting these components on the chip to ensure performance, good power consumption, and other design needs. Accurately digitally determining the placement of such infinitesimally small components is, as you might expect, difficult.

Alpha Chip “could generate a very high quality layout in about six hours. And the nice thing about this approach was that there is actually learning from the experience,” Goldie said.

The starting point of their design work for AI chips is the use of ‘a reward signal’ that indicates how good the design is. The agent then uses that assessment to “update the parameters of its deep neural network to get better,” Goldie said. After completing thousands of designs, the agent got really good. It also got faster as it learned, the founders say.

Ricursive’s platform will take the concept further. The AI ​​chip designer they build will “learn about different chips,” Goldie said. So every chip he designs should help him become a better designer for each subsequent chip.

Ricursive’s platform also uses LLMs and will handle everything from component placement to design verification. Every company that makes electronics and needs chips is its target group.

If their platform proves itself, which it likely will, Ricursive could play a role in the moonshot goal of achieving artificial general intelligence (AGI). Their ultimate vision is to design AI chips, meaning the AI ​​will essentially design its own computer brain.

“Chips are the fuel for AI,” Goldie said. “I think building more powerful chips is the best way to push that limit.”

Mirhoseini adds that the lengthy chip design process limits the speed at which AI can develop. “We think we can also enable this rapid co-evolution of the models and the chips that actually power them,” she said. So AI can become smarter faster.

If the thought of AI designing its own brain at ever-increasing speeds brings to mind visions of Skynet and the Terminator, the founders point out that there is a more positive, more immediate and, they think, more likely benefit: hardware efficiency.

When AI Labs can design much more efficient chips (and eventually all underlying hardware), their growth won’t have to consume so much of the world’s resources.

“We could design a computer architecture ideally suited to that model, and we could improve performance almost tenfold per total cost of ownership,” Goldie said.

Although the young startup won’t name its first customers, the founders say they’ve heard of every major chip manufacturer you can think of. It is not surprising that they can also choose their first development partners.