Anthropic Held Secret Acquisition Talks with Robotics Startup Physical Intelligence
Key Takeaways
- ▸Anthropic and Physical Intelligence held acquisition talks in spring 2026, confirming that rumors swirling on social media had a factual basis, even if details were imprecise
- ▸Both Anthropic and OpenAI are on aggressive acquisition sprees while preparing for IPOs, signaling a race to acquire specialized technical talent and proven research capabilities
- ▸Industry focus on robotics and embodied AI reflects a growing consensus that superintelligent systems may require physical-world understanding beyond text-based learning
Summary
A weekend rumor claiming that Anthropic was acquiring robotics startup Physical Intelligence spread rapidly across AI Twitter, despite a denial from Physical Intelligence's CEO Karol Hausman, who responded via a casual Slack message with an Office gif. However, The Information has confirmed that the rumor wasn't entirely baseless—Anthropic and Physical Intelligence did actually hold acquisition talks this spring, though no deal materialized. The specifics may have been wrong, but something real did happen.
The acquisition chatter fits into a broader pattern of aggressive consolidation by both Anthropic and OpenAI as they prepare for major IPOs. Anthropic has announced four acquisitions this year, while OpenAI has acquired at least 17 companies since 2023. Both filed for IPOs in June, setting the stage for what could be transformational exits for the AI industry. Physical Intelligence itself is no minor player—it's raised over $1 billion, was valued at $11 billion in recent funding talks, and its π0.5 model is widely used in robotics research.
The underlying motivation appears to be a strategic bet that physical-world understanding is essential for achieving superintelligence, and that text-based models alone are insufficient. OpenAI has already built out a robotics lab in San Francisco with near-term ambitions in infrastructure robotics and long-term goals for personal robots. Anthropic, meanwhile, has focused on robotics research for safety stress-testing, including Project Fetch (testing Claude's ability to help non-experts program robots) and follow-up work showing a newer model completing tasks roughly 20 times faster than prior human-plus-Claude teams. Acquiring an established robotics team like Physical Intelligence would accelerate that capability-building significantly.
