Physical Intelligence is reportedly raising $1B at a $20B valuation — with no revenue yet

Physical Intelligence, a San Francisco startup that writes software for robots rather than building robots itself, is in talks to raise at least $1 billion at a valuation of roughly $20 billion, according to a Bloomberg report published on October 27. Existing investors Avenir and Lux Capital are reportedly in talks to lead it, according to people familiar with the matter.

If it closes on those terms, it would nearly double the company’s value in under half a year — for a business that, by the same accounts, has not yet started bringing in revenue. Investors are not paying for sales; they are paying for a position in what they believe the next platform will be.

The numbers, and how fast they moved#

The trajectory, as reported, is steep. Physical Intelligence raised $400 million at $2.4 billion in November 2024; a round earlier in 2026, described as an unannounced Series C in May, valued it at about $11.5 billion. Secondary trades had put it near $11.8 billion by the end of September, before the new talks surfaced.

All of this comes with the same caveat: these are reported talks and reported prior rounds, not a company announcement. Physical Intelligence, Avenir and Lux Capital did not respond to requests for comment, and a negotiation at $20 billion can still close lower, shrink, or fall apart.

What the company actually sells#

Founded in 2024 by former Google DeepMind researchers and academics from Stanford and UC Berkeley, the company does not sell a robot you can buy. It sells the intelligence that could run inside robots built by someone else: its flagship model, pi-zero, turns images and language into physical actions, trained on data from real robot arms doing real tasks.

Demonstrations show the same model, fine-tuned, doing ordinary household work: folding laundry, making coffee, and clearing tables, including sorting dishes from rubbish. The pitch is generality: one robot model, given enough data and compute, could transfer across different robot bodies and jobs, rather than needing a bespoke program for each machine and task.

That pitch got its first outside test earlier this month, when the company opened its latest model to external developers in beta. A model other developers can build on is the start of a platform — but for now it is only a beta, and the step from developers experimenting to customers paying is exactly what the company has not yet shown.

Why investors are paying frontier-lab prices#

The bet is that robotics is where the language-model playbook works next. Language models scaled on internet text; robot models must learn from physical interaction, which is slower and costlier to collect and nobody has yet harvested at scale. Whoever assembles the most diverse robot experience, and the model that learns most generally from it, could own a layer every robot maker needs.

Physical Intelligence is not alone in being priced on that thesis. Skild AI, building a similarly hardware-agnostic robot brain, was valued above $14 billion in January and was reported in September to have passed $100 million in annual recurring revenue — traction Physical Intelligence has yet to match on current reporting. Figure AI, which builds its own humanoids, closed a round above $1 billion in September 2025 at a reported $39 billion. The market is pricing all three as if at least one will become infrastructure.

Its existing backers read like a bet on that outcome too: Alphabet’s CapitalG, Jeff Bezos and Thrive Capital are among its investors. A round led by existing investors would also say something — that the people who know the company best are willing to defend its price, even at double the May mark.

What to watch#

• Whether the round closes, and at what price: talks are not a term sheet, and a $20 billion discussion can settle lower. The first confirmation will matter more than the first report.
• Who leads, and who joins: a round led by Avenir and Lux Capital alone looks different from one that pulls in a new large investor at the higher price.
• What the beta developers do next: if outside developers keep building on the model and report real results, the platform claim strengthens; if usage fades after the novelty, it weakens.
• The first revenue: the most important number in the company’s next chapter is not its valuation but its first dollar of recurring revenue — and what customers are paying for.
• Competition: OpenAI, Tesla and Skild AI are all pursuing versions of the same prize, and a rival with revenue already on the books sets the bar Physical Intelligence must clear.

A $20 billion tag on zero revenue is either early recognition of the next platform or a warning about how much of the future is already priced in. That will not be settled by any funding announcement, but by robots doing useful jobs, reliably, for paying customers, on hardware the company did not build.