Anthropic-Physical Intelligence Acquisition Rumor Shakes AI Twitter

Over the weekend, a rumor that Anthropic was acquiring robotics startup Physical Intelligence spread quickly through the AI community. Physical Intelligence CEO Karol Hausman denied it via a Slack message, but according to The Information, the two companies did in fact hold acquisition talks this past spring.

Physical Intelligence was co-founded by former Google researchers and Stanford/Berkeley faculty, has raised over $1 billion, and was in talks to raise an additional $1 billion at an $11 billion valuation. Its robot control model π0.5 is one of the most widely used models in robotics research.

Behind Anthropic's interest in robotics is the view that understanding the physical world is a prerequisite for superintelligent systems. Last November, Anthropic ran an experiment called Project Fetch, in which Claude helped program a robot dog, and in a follow-up study this past June, it reported that its latest model completed the same task about 20x faster than last year's human+Claude team.

Meanwhile, OpenAI is also an investor in Physical Intelligence, creating a complicated situation in which a rival company allegedly sought to acquire one of OpenAI's portfolio companies. Anthropic has made 4 acquisitions this year, while OpenAI has acquired at least 17 companies since 2023. Both companies are preparing for IPOs, with Anthropic having confidentially filed for one on June 1.