Physical Intelligence Valuation Nears $11B as Robot Brains Win

A valuation that nearly doubled in four months

Physical Intelligence, the San Francisco startup building general-purpose AI models to control robots, is reportedly in talks to raise roughly $1 billion at an $11 billion valuation. Bloomberg and TechCrunch reported the talks on March 27, 2026 4. If the round closes on those terms, it would nearly double the company's $5.6 billion valuation from about four months earlier 4. That earlier figure came with the company passing $1 billion in total funding, from backers including Khosla Ventures, Sequoia Capital, and Thrive Capital 3. Sacra lists the company's 2025 funding at $1.07 billion against the $5.6 billion valuation 1.

The company was founded in 2024 and is led by CEO Karol Hausman 1. It reportedly has fewer than 80 employees 4. Co-founder Sergey Levine, a UC Berkeley associate professor, pitches the product simply: think of it as ChatGPT, but for robots 3.

What the money is buying

The technical case rests on a run of model releases. The first, π0, arrived in October 2024 and could control seven different robot types across more than 50 tasks 1. One account puts the number at 68 distinct tasks 4. A later version, π*0.6, adds reinforcement learning from both the robot's own autonomous experience and human corrections. According to Sacra, that roughly doubled completion rates on demanding manipulation jobs such as making espresso and folding laundry 1.

The company has also open-sourced π0's weights. Developers can fine-tune the model with as little as 1 to 20 hours of their own robot data 1. The hardware is deliberately modest. The arms in its lab cost around $3,500 each, and materials would run under $1,000 if built in-house 3. That is a bet that intelligence, not expensive mechanics, is the bottleneck.

The revenue question: conflicting pictures

On the business model, the accounts disagree. Sacra describes a B2B software-as-a-service model for robotics companies, manufacturers, and integrators. Customers pay a $300 monthly subscription per connected robot, so revenue grows with fleet size 1. Dealroom says the company licenses its models to multiple robot makers, including Ultra, on a non-exclusive basis 5.

Other coverage is blunter. One report frames the company as having no revenue plan yet, with co-founder Lachy Groom declining to tell backers when they will see returns 3. Another states flatly that the company has no revenue 4.

The likeliest reconciliation is that a per-robot price and early licensing deals exist, but the revenue is still negligible next to the valuation. Leadership's public emphasis remains research over commercialization. Either way, investors are not paying $11 billion for current sales. They are paying for the chance that one model becomes the default brain across many manufacturers' machines, which one outlet calls an "operating system" for robotics 2.

Brains over bodies

The deal sits inside a broader funding surge. Robotics startups raised nearly $14 billion in 2025, above even the 2021 venture peak, according to Crunchbase data 2. That money splits between humanoid hardware makers such as Figure, Apptronik, and 1X, and foundation-model labs such as Physical Intelligence, Skild AI, and Generalist 2. Figure reached a $39 billion post-money valuation, and Skild AI tripled to $14 billion in seven months 2. Those figures suggest the market is pricing physical AI like frontier AI rather than industrial automation.

Dealroom's comparison is more pointed. Ultra, a robots-as-a-service operator that depends on Physical Intelligence's models, raised $62 million without disclosing a valuation, while its AI supplier commanded a multibillion-dollar price 5. Dealroom also cites XPeng's robotics unit raising over $900 million at a $6.3 billion-plus valuation and chipmaker SiMa.ai raising $150 million at $1.45 billion 5. Its read: investors are paying the steepest premiums for the intelligence and compute layers, not for the companies deploying robots 5.

Our read

The $300-per-robot figure helps explain the logic. A hardware-agnostic subscription has software-like margins and grows with every robot deployed, whoever built the machine. Hardware and deployment businesses carry inventory, field service, and thinner margins. That asymmetry is why capital is flowing upstream.

The risk is just as clear. Open-sourcing π0 builds an ecosystem, but it also commoditizes the baseline. Well-funded rivals like Skild AI are chasing the same position 2. A near-doubling in four months, with little or no reported revenue, prices in a winner-take-most outcome that has not yet been shown. For founders, the lesson is that the model layer carries the pricing power for now. For investors, the open question is whether robot brains will consolidate like operating systems or fragment like the language-model market.