Physical Intelligence raises $1 billion for robot foundation models

San Francisco-based startup Physical Intelligence has secured over $1 billion in funding at a $5.6 billion valuation. The company is developing robotic foundation models—AI models for robot control—designed to operate across different robot hardware. Backed by investors including Khossla Ventures, Sequoia Capital, and Thrive Capital, the company is prioritizing long-term research over immediate commercialization. Co-founder Lachy Groom, a 31-year-old former Stripe employee who spent roughly five years as an angel investor before making his first robotics investment in Standard Bots in 2021, does not provide timelines for revenue generation. “I don’t give investors answers on commercialization,” Groom said, highlighting a strategy that focuses on research before selling products.

This research-first approach stands in contrast to its competitors. In Pittsburgh, rival startup Skild AI, founded in 2023, has taken a more commercial path. Skild AI recently raised $1.4 billion at a $14 billion valuation and has already deployed its system commercially, generating $30 million in revenue last year across security, warehouses, and manufacturing. While Skild AI operates on the premise that commercial deployment creates a data feedback loop, Physical Intelligence is betting that resisting near-term commercialization will enable it to produce superior general intelligence.

Physical Intelligence’s technical strategy relies on cross-embodiment learning, which involves training AI models across different robot hardware. This approach allows the software to transfer knowledge to new physical platforms, lowering the marginal cost of onboarding autonomy. Co-founder Sergey Levine, an associate professor at UC Berkeley, describes the vision as being similar to ChatGPT, but for robots. The startup’s founding team also includes Stanford lab head Chelsea Finn, former Google DeepMind researchers Karol Hausman and Quan Vuong, and Groom, who started his first company in Australia at age 13.

The company, which has about 80 employees, conducts its testing using robot arms that cost about $3,500 each, though manufacturing them in-house could bring material costs below $1,000. While the team initially set a 5- to 10-year roadmap for what they believed was possible, Groom noted they had progressed through it by month 18. However, physical deployment remains a bottleneck. Groom acknowledged the friction of working with physical systems, noting that hardware is difficult and everything they do is harder than running a software company.

Why it matters

Physical Intelligence is a well-capitalized startup attempting to build general-purpose robotic foundation models, positioning itself against competitors like Skild AI by prioritizing research over near-term commercialization.