Physical Intelligence, the San Francisco robotics startup founded by former Google DeepMind researchers, is in talks to raise approximately $1 billion at a valuation exceeding $11 billion. The deal would double the company’s valuation in just four months.
The Numbers
Physical Intelligence closed a $600 million round in late 2025 at a $5.6 billion valuation. Founders Fund is expected to participate in the new round, with Lightspeed Venture Partners also in discussions. Returning investors Thrive Capital and Lux Capital are expected to follow on.
The company has now raised over $1.6 billion total since its 2024 founding—putting it among the most capital-intensive AI startups outside the large language model players.
What They’re Actually Building
Unlike most robotics companies that build specialized machines for specific tasks, Physical Intelligence is creating what co-founder Sergey Levine calls “ChatGPT for robots”—a foundation model approach that can power any robot across any task.
Their flagship model, π0, is a 3-billion parameter transformer trained on over 10,000 hours of real-world robot data spanning seven robot types and 68 different tasks. The model has demonstrated capabilities across unstructured environments: cleaning kitchens, folding laundry, making beds, and following complex natural language instructions.
In February 2026, Physical Intelligence open-sourced the π0 code and weights, a move that signals confidence in their ability to stay ahead through execution rather than secrecy.
The Founding Team
The company’s leadership reads like a robotics all-star roster:
- Karol Hausman (CEO): Former Staff Research Scientist at Google DeepMind, adjunct professor at Stanford. Over a decade focused on robot learning that transfers across environments.
- Sergey Levine (Chief Scientist): UC Berkeley associate professor whose lab pioneered deep reinforcement learning for robotic manipulation.
- Chelsea Finn: Stanford assistant professor specializing in machine learning for robotics.
- Brian Ichter: Former Google DeepMind/Brain researcher with a PhD from Stanford in aeronautics.
- Adnan Esmail: Engineering background from Anduril Industries and Tesla hardware.
- Lachy Groom: Early Stripe employee turned investor with bets on Figma, Notion, and Ramp.
The Real Play
Physical Intelligence is betting that robotics will follow the same scaling dynamics that transformed language models. Collect enough diverse real-world data, train large enough models, and robots should generalize across tasks the way GPT generalizes across text.
The company resists early specialization despite commercial pressure. They’re not building kitchen robots or warehouse robots—they’re building the intelligence layer that could power all of them.
This approach requires enormous capital and patience. The $11 billion valuation reflects investor confidence that the foundation model thesis will work for physical manipulation the way it worked for language and vision.
Who Wins, Who Loses
If Physical Intelligence succeeds, they become the infrastructure layer for an entire industry. Every robotics company that can’t match their data flywheel and model scale becomes a hardware vendor licensing Pi’s brains.
Traditional robotics players like Boston Dynamics, Fanuc, and ABB face a strategic choice: build their own foundation models (expensive and unproven), license from the new entrants (margin compression), or hope the foundation model approach doesn’t work (risky bet).
The wildcard is timing. Foundation models for robotics remain unproven at scale. The demos are impressive, but commercial deployments with reliable error rates are still nascent. Physical Intelligence is raising enormous sums on the promise that scaling laws transfer to the physical world—a thesis that’s intuitive but unconfirmed.
At $11 billion, investors are paying for that promise. The next year will show whether it’s earned.