Anthropic and Physical Intelligence held acquisition talks, sources confirm, reigniting AI rivalry over robotics
Anthropic and Physical Intelligence did hold acquisition talks this spring, according to The Information, even though the companies denied a completed deal. The incident reflects an escalating robotics competition between OpenAI and Anthropic as both prepare for IPOs.
A weekend rumor that Anthropic was acquiring Physical Intelligence spread rapidly across tech social media, only to be denied by Physical Intelligence's CEO via a Slack message with an Office gif. The specifics of the rumor appear to have missed the mark, but the underlying claim was grounded: Anthropic and Physical Intelligence did hold acquisition talks this spring, according to The Information.
The rumor originated from tech blogger Robert Scoble's post on X, which gained traction partly because Physical Intelligence is no lightweight player. The robotics startup, co-founded by investor-operator Lachy Groom and former Google researchers alongside Stanford and Berkeley professors, has raised over $1 billion and was reportedly seeking another $1 billion round at an $11 billion valuation earlier this year. Its π0.5 model counts among the more widely used robot brains in academic robotics research.
Why the acquisition interest matters has less to do with Physical Intelligence's current valuation and more to do with what both Anthropic and OpenAI are signaling about AI's future. Both companies are preparing for public offerings—Anthropic confidentially filed for an IPO in June, OpenAI the following week—yet they're simultaneously doubling down on robotics. The pattern suggests a shared conviction: understanding the physical world may be non-negotiable for building genuinely capable AI systems.
OpenAI offers a cautionary precedent. The company built an early robotic hand capable of solving a Rubik's Cube, then shut its robotics group in 2021. Co-founder Wojciech Zaremba later acknowledged the approach was missing critical pieces for achieving superintelligence. OpenAI revived the effort in 2024 by quietly establishing a humanoid robotics lab in San Francisco. CEO Sam Altman made it public in late May, announcing OpenAI Robotics would focus on infrastructure robots near-term and personal robots as a long-term ambition.
Anthropic's path has differed. Rather than building a hardware lab, the company has published research through an internal group testing Claude's capabilities for safety. In November, Anthropic's Project Fetch examined how well Claude could help non-experts program a robot dog. A second phase in June found a newer model completing identical tasks roughly 20 times faster than the best human-plus-Claude team from a year prior. Acquiring an established robotics team would bypass years of iterative work.
But a potential complication clouds the picture. Physical Intelligence's early backers include Khosla Ventures, Thrive Capital, and Founders Fund—the same investor constellation backing OpenAI. Critically, OpenAI itself is an investor in Physical Intelligence, meaning the company holds a stake in the very firm its main rival was recently negotiating to buy. Depending on the terms of OpenAI's investment, the company may have negotiated information rights or a right of first refusal on any sale to a competitor. Such provisions are standard for strategic investors precisely to forestall scenarios like this one.
Physical Intelligence CEO Karol Hausman's response to the rumor mill—a Slack gif and no further comment—fell short of categorical denial. Lachy Groom did not respond to requests for comment. The unfolding situation illustrates a widening tug-of-war between Anthropic and OpenAI as both approach public markets while racing to secure talent and IP in robotics, a domain neither wants ceded to the other.
