The co-founders of Ricursive Intelligence appear to have been fated to work together. After declining other offers, the duo, who previously collaborated at Google Brain and were early team members at Anthropic, made a significant impact at Google. They developed the Alpha Chip, an AI system capable of producing high-quality semiconductor layouts in mere hours—a task that traditionally occupies human engineers for over a year. This technology contributed to the design of three iterations of Google's Tensor Processing Units.

This impressive background clarifies how, merely four months after founding Ricursive, the company secured a $300 million Series A funding round at a $4 billion valuation, led by Lightspeed. This milestone followed a $35 million seed round led by Sequoia by just a few months. Ricursive is focused on creating AI software for chip design, not manufacturing the physical chips themselves. This distinction sets them apart from most other AI chip ventures; they are not positioning themselves as a competitor to Nvidia. In fact, Nvidia is an investor. The startup's intended clients are GPU manufacturers and other chipmakers, including AMD and Intel.

"We aim to enable the automated and highly accelerated construction of any chip, whether custom or more traditional," explained one founder.

Their partnership began at Stanford University, where Azalia Mirhoseini was teaching computer science as Anna Goldie completed her PhD. Their professional journeys have been remarkably synchronized ever since.

"We began 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 returned to Google on the same day, and then we left Google again on the same day. Then we started this company together on the same day," Goldie recalled.

Their closeness at Google extended to shared circuit training workouts. This routine inspired a pun from their collaborator, the renowned Google engineer Jeff Dean, who dubbed their Alpha Chip initiative "chip circuit training." Internally, the pair became known as "A&A."

While the Alpha Chip brought them industry recognition, it also sparked controversy. In 2022, a colleague at Google was reportedly dismissed after a prolonged campaign to undermine A&A and their semiconductor work, despite its critical role in developing some of Google's most vital AI chips.

The Alpha Chip project at Google Brain validated the core idea behind Ricursive: using artificial intelligence to radically speed up chip design.

**Designing chips is a formidable challenge.**

The complexity arises because modern computer chips integrate millions to billions of microscopic logic gates onto a silicon wafer. Engineers can spend a year or more meticulously arranging these components to optimize performance, power efficiency, and other design requirements. Achieving precise digital placement for components at such a minute scale is, unsurprisingly, extremely difficult.

"The Alpha Chip could produce a very high-quality layout in about six hours. The remarkable aspect of this method is that it actually learns from experience," Goldie stated.

Their AI design approach is based on a "reward signal" that evaluates the quality of a design. The AI agent uses this feedback to "update the parameters of its deep neural network to improve," Goldie explained. After iterating through thousands of designs, the agent became highly proficient and increasingly faster.

Ricursive's platform aims to advance this concept further. The AI designer they are building will "learn across different chips," according to Goldie. This means each design it completes enhances its capability for all subsequent projects. The platform will also leverage Large Language Models (LLMs) and manage the entire process from component placement to final design verification.

Their potential customer base includes any company that manufactures electronics and requires semiconductors. If successful, Ricursive could contribute to the ambitious pursuit of Artificial General Intelligence (AGI). Their long-term vision involves AI systems designing the very chips that power them, essentially allowing AI to engineer its own computational hardware.

"Chips are the fuel for AI," Goldie said. "I believe building more powerful chips is the most effective way to push that boundary forward."

Mirhoseini adds that the slow pace of chip design is a bottleneck for AI progress. "We believe we can enable a rapid co-evolution of the AI models and the chips that power them," she noted, which would allow AI to advance more swiftly.

For those concerned about AI autonomously designing its own hardware at accelerating speeds—evoking dystopian scenarios—the founders highlight a more immediate and positive outcome: dramatic gains in hardware efficiency. If AI labs can design vastly more efficient chips and underlying hardware, the expansion of AI won't demand such an immense consumption of global resources.

"We could design a computer architecture uniquely suited to a specific model and achieve nearly a tenfold improvement in performance per total cost of ownership," Goldie said.

While the young company does not disclose its early clients, the founders report hearing from every major chipmaker imaginable. Consequently, they have the luxury of selecting their preferred development partners.