In a year when AI giants have turned acquisitions into a background hum, one weekend rumor managed to cut through the noise. Posts on X claimed Anthropic was buying Physical Intelligence, a well-funded robotics startup whose software powers a growing share of experimental robot fleets. The story spread rapidly across AI Twitter, even after Physical Intelligence’s chief executive pushed back internally.

The frenzy reflected how unusual this particular target would be. Physical Intelligence is no anonymous lab. Co-founded by investor Lachy Groom alongside former Google researchers and academics from Stanford and Berkeley, the company has reportedly raised more than $1 billion, with talks for another massive round at an $11 billion valuation. Its π0.5 model is widely cited as one of the most capable “robot brains” available to researchers, a kind of operating system for embodied AI.

The rumor was not entirely baseless. According to reporting from The Information, Anthropic and Physical Intelligence did hold acquisition talks earlier this year. Tech commentator Robert Scoble appears to have overstated the outcome while correctly sensing that something significant had been in motion.

Physical Intelligence’s internal response did little to calm speculation. Chief executive Karol Hausman told staff the reports were untrue via Slack, attaching a gif from the sitcom The Office of a character shaking her head no. The denial was read by some as more playful than definitive, especially given the high strategic stakes.

Those stakes are rising fast. Anthropic has already logged several acquisitions, while OpenAI has snapped up more than a dozen companies as both race toward public offerings and seek to turn cutting-edge models into durable revenue. Robotics is emerging as the next contested frontier: OpenAI disbanded its first robotics group in 2021, only to reconstitute a humanoid robotics lab and formally launch OpenAI Robotics with ambitions that range from infrastructure work to eventual personal robots.

Anthropic, by contrast, has focused on safety research, including Project Fetch, which tested how well its Claude models could help non-experts program robot dogs. Results suggested rapid gains in capability, but no substitute for a seasoned robotics team.

Acquiring Physical Intelligence would let Anthropic vault years ahead. Yet there is a twist: OpenAI is already an investor in Physical Intelligence, alongside overlapping venture backers such as Khosla Ventures, Thrive Capital, and Founders Fund. That raises thorny questions about information rights or rights of first refusal that might give OpenAI leverage if the startup ever formally goes up for sale.

For now, the only certainty is that embodied AI has moved from side project to strategic prize. The rumor may have been premature, but it exposed how central control of robot brains has become to the next phase of the AI arms race.