The headline fact is simple: Anthropic and Physical Intelligence held serious acquisition talks this spring. The rumor that spread online last weekend had the details wrong, but the underlying story was real. Physical Intelligence's CEO did deny it, though his message to employees was a Slack GIF of a character shaking her head, which is not quite the same as a formal statement.
So what is Physical Intelligence, and why does it matter enough to cause this kind of noise?
Physical Intelligence does not make robots. It makes the software that runs inside them. Its model, called π0 (pi-zero), takes plain-language instructions and translates them into physical movements that robotic arms can follow. The same software works across different robot hardware, which means a manufacturer does not have to rebuild everything from scratch each time they switch robot platforms. That flexibility is what the market is paying for.
The company was founded in early 2024 by researchers from Google DeepMind, Stanford, and UC Berkeley. It raised $70 million to start, then $400 million at a $2.4 billion valuation, then $600 million at $5.6 billion. That is three funding rounds in under two years, each at a higher price, for a company with no commercial product and no disclosed revenue. Investors are essentially paying for the team and the research.
The reason Anthropic was interested is straightforward. Both Anthropic and OpenAI believe that truly powerful AI will eventually need to understand and operate in the physical world, not just process text on a screen. OpenAI made this clear: it built an early robotic hand in 2019, shut the whole effort down, then quietly rebuilt a robotics lab starting in early 2025. That lab now employs roughly 100 people in San Francisco, and in late May, OpenAI officially announced a dedicated robotics division with the long-term goal of a personal robot for everyone.
Anthropic has no hardware lab. What it has done is run internal research tests on how well its AI models can guide robots, including a project called Project Fetch, where staff tested whether non-experts could use Claude to program a robot dog. A follow-up in June found that a newer model completed the same tasks around 20 times faster than the best results from the year before. That is promising research, but it is nowhere near the capabilities Physical Intelligence has already built.
Here is the problem for Anthropic. OpenAI is not just a competitor in this context; it is a shareholder. OpenAI invested in Physical Intelligence in both the Series A and Series B rounds. Strategic investors at that level often negotiate terms that give them the right to be informed of, or to match, any acquisition offer from a rival. If those terms are in place here, Anthropic cannot simply write a check and close the deal. OpenAI would have the ability to step in.
OpenAI did not respond to questions about this when asked.
For businesses that operate warehouses, manufacturing lines, logistics operations, or any facility that already uses or is considering robots, this matters. The race to build a universal software brain for robots is accelerating. If one of these two companies acquires Physical Intelligence, its technology is unlikely to remain a neutral, widely available product. It would become a proprietary asset tied to whichever AI platform the acquirer sells. Companies currently using or evaluating Physical Intelligence's technology should factor that dependency into their planning now.
The broader picture is that Anthropic and OpenAI are both preparing to go public, having filed confidentially for IPOs within a week of each other in June. Each needs to show investors a credible path into physical-world applications. Robotics is that path. The acquisition race that follows is not really about who gets Physical Intelligence. It is about who controls the infrastructure layer that runs inside the next generation of industrial machines.