Anthropic, OpenAI in robotics race; acquisition talks with Physical Intelligence surface

The weekend rumor that Anthropic was acquiring Physical Intelligence, a San Francisco robotics startup, spread rapidly across social media and tech Twitter. Physical Intelligence CEO Karol Hausman initially dismissed the reports via a Slack message to employees that contained a gif of a character from "The Office" shaking her head no—a response that The Information characterized as notably lacking vigor. However, subsequent reporting by The Information revealed that the rumor was not entirely spurious: Anthropic and Physical Intelligence did hold acquisition talks this spring, though no deal materialized. Tech blogger Robert Scoble, whose weekend post on X sparked the frenzy, may have gotten specifics wrong without being entirely incorrect that something had happened.

Physical Intelligence is not an obscure player. Co-founded by investor-operator Lachy Groom roughly two years ago alongside former Google researchers and professors from Stanford and Berkeley, the company has raised more than $1 billion（約1600億円） and was reportedly in talks this spring for another $1 billion（約1600億円） round at an $11 billion（約1.8兆円） valuation. Its π0.5 model is among the more widely used robot brains in robotics research. Groom's star has been rising in Silicon Valley, and the company's early investor base overlaps significantly with OpenAI's own, including Khosla Ventures, Thrive Capital, and Founders Fund. Notably, OpenAI itself is an investor in Physical Intelligence, a fact that complicates the acquisition picture.

The broader context is a race between Anthropic and OpenAI to build robotics capabilities. OpenAI built an early robotic hand capable of solving a Rubik's Cube before shutting down its entire robotics group in 2021; co-founder Wojciech Zaremba later explained the shutdown by saying the approach was missing pieces needed for real superintelligence. The company reversed course in 2024, quietly establishing a humanoid robotics lab in San Francisco. In late May, CEO Sam Altman made the effort official, announcing "OpenAI Robotics" was hiring and describing a near-term focus on robots for infrastructure work and a long-term goal of a personal robot for everyone. Anthropic has not built a hardware lab comparable to OpenAI's but has published research through an internal group that stress-tests frontier capabilities. Project Fetch, released last November, tested how much Claude could help non-experts program a robot dog. A second phase in June found that a newer model completed the same tasks roughly 20 times faster than the best human-plus-Claude team from the year before.

Both companies are also preparing for major public debuts. Anthropic confidentially filed for an IPO on June 1, followed by OpenAI a week later. The acquisition activity—Anthropic has made four known acquisitions this year while OpenAI has acquired at least 17 companies since 2023—suggests both are racing to consolidate robotics expertise before going public. However, OpenAI's existing stake in Physical Intelligence raises questions about whether it negotiated information rights or a right of first refusal on a sale to a competitor, the kind of protective provisions large strategic investors sometimes seek. If Physical Intelligence is in play, that stake could give OpenAI leverage in any bidding war with Anthropic.

The Anthropic-Physical Intelligence acquisition rumor, though unconfirmed, illuminates a strategic pivot happening across the AI industry's largest players. Both Anthropic and OpenAI have shifted from pure language models toward embodied AI—systems that interact with the physical world. OpenAI's path is instructive: it built an early robotic hand that solved a Rubik's Cube, then abandoned robotics in 2021 because co-founder Wojciech Zaremba believed the approach was missing pieces needed for superintelligence. That conviction reversed in 2024, when OpenAI quietly built a humanoid robotics lab in San Francisco and made it public in late May. Anthropic's approach has been more cautious; rather than building hardware, it has stress-tested its Claude model's ability to help humans program robots through internal research—notably Project Fetch last November and a June follow-up showing vastly faster task completion.

The timing of these acquisition sprees matters. Both companies filed for IPO within days of each other (Anthropic on June 1, OpenAI a week later), positioning themselves for two of the largest U.S. stock debuts in history. The race to secure robotics expertise and teams—whether through acquisition or internal build—reflects a belief that physical-world understanding is essential to the next phase of AI capability. For investors, the Financial Intelligence stake in both companies underscores how intertwined the AI ecosystem has become and raises questions about whether early strategic investments, like OpenAI's stake in Physical Intelligence, carry protective provisions that could influence future M&A outcomes.