Physical Intelligence, co-founded by a Russian-born robotics professor, has raised more than $2.1 billion in two years. Its investors include billionaire Jeff Bezos and even, to some extent, rival OpenAI. What makes the company so interesting?
Expanding its footprint
What do short-term rental provider Airbnb and Physical Intelligence have in common? The answer may come as a surprise: a former industrial building on Brannan Street in San Francisco. In the early 2010s, Airbnb’s headquarters were located there; just days ago, PI subleased part of the office space. The area covers 80,964 square feet, although earlier this year the space under discussion was around 60,000 square feet.
At the same time, Physical Intelligence expanded south of San Francisco, leasing 233,500 square feet in a business complex near Google’s headquarters. According to the Silicon Valley Business Journal, the space is intended for offices and R&D.
Why might the company need so much space all at once if it does not engage in physical manufacturing, despite its name, Physical Intelligence? The answer likely lies in the nature of its technology: Physical Intelligence is trying to create a kind of universal “brain” for real-world robots—a software layer for machines powered by AI. The company may therefore need computing infrastructure and a large amount of space where it can train real-world robots, test models of their “brains,” and refine them.
What Physical Intelligence does
The company develops artificial intelligence models for robotics—vision-language-action (VLA) models that can perceive their surroundings, understand instructions, and control a machine’s actions.
The latest such model is π0.7, introduced in April this year. The company says the model can transfer skills across robots and tasks: for example, it can work with a second machine and fold clothes without task-specific data, as well as learn to operate a household appliance through language training.
The company’s materials also describe experiments in real-world settings, not just in labs: robots powered by its models are operating at the facilities of active businesses, performing tasks such as packing orders.
According to Revenue Memo, Physical Intelligence had around 80 employees in January 2026.
“I’m not very good at business…”
“…And I’m not very knowledgeable about investing, so I can’t claim to know how to do it,” Sergey Levine, one of Physical Intelligence’s founders and a professor of robotics at the University of California, Berkeley, said in a recent episode of The Peterman Post podcast. He has previously worked at Google and received the U.S. Presidential Early Career Award for Scientists and Engineers (PECASE).
Cade Metz’s book Genius Makers says that the professor grew up in Moscow and that his parents were engineers who worked on the Soviet Buran space shuttle program.
Physical Intelligence was co-founded by Levine, along with Google DeepMind researcher and Stanford lecturer Karol Hausman, Stanford professor Chelsea Finn, former Google Brain and Google DeepMind researcher Brian Ichter, Adnan Esmail, who previously led engineering at Anduril after working at Tesla, and Lachy Groom, a former Stripe employee and prominent angel investor who has invested in companies including Figma, Notion, and Ramp.
The company does not disclose financial statements. Even so, Physical Intelligence has already raised around $2.1 billion from investors. According to Revenue Memo, they include OpenAI, Jeff Bezos, Alphabet’s CapitalG fund, and Thrive Capital. OpenAI participated in funding rounds both directly—in November 2024 and November 2025—and through the OpenAI Startup Fund in March 2024. The company does not provide a breakdown of the exact investment amounts.
Bezos took part in two funding rounds and was one of the lead investors in the $400 million Series A in November 2024, then participated in the $600 million Series B in November 2025. The size of his investments was not disclosed.
In July, the Forge platform included Physical Intelligence among the ten most sought-after private companies of the second quarter of 2026 and listed its valuation at $11.19 billion. According to AXEVIL’s estimates, the company’s shares may now be worth $633.50 each. Nasdaq Private Market, by contrast, valued a share at $500.59 on September 16.
Physical Intelligence’s strengths and weaknesses
The company’s key risk is that it must turn its pilot projects and research into full-fledged paid contracts within a relatively short time, according to a Startup Diligence report. It says that a lack of commercial revenue by the end of 2026, coupled with no compelling customer pipeline, could become a “tipping point” for its investment appeal.
The Startup Diligence report also discusses competition from Skild AI and Google DeepMind, as well as Physical Intelligence’s reliance on Google technology: the company’s first VLA model is built on PaliGemma, a foundational model developed by Google DeepMind.
The Startup Diligence report also notes the risk of dependence on a few key people: if Sergey Levine or Chelsea Finn were to leave before the company begins generating commercial revenue, it could undermine its investment appeal.
But the team itself is also considered one of Physical Intelligence’s strengths. Its researchers come from different fields, giving the company the expertise to train models on different types of robots. This could create a “data flywheel” effect: the more real-world robots use the model, the more data the company gets to further train its robots.
Sergey Levine previously said that robotics is still at the stage of searching for a technology that can be scaled reliably. Unlike large language models used in AI, where increasing computing resources, data volume, and model size already leads to a fairly predictable improvement in capabilities, this relationship has not yet been established for robotics.
Representatives of the company, as well as Levine himself, did not respond to questions about plans to raise funding or go public.ㅤ