Anthropic-Physical Intelligence acquisition talks resurface amid robotics push
Anthropic and Physical Intelligence held acquisition talks this spring, rekindling debate over AI labs' pivot toward embodied reasoning.
Last verified:
Acquisition Talks Confirmed, Deal Did Not Close
Anthropic and Physical Intelligence held acquisition negotiations in spring 2026, according to The Information, though the discussions did not result in a merger. The rumor surfaced publicly over the weekend when tech blogger Robert Scoble posted about the potential deal on X, sparking rapid spread across AI social media despite Physical Intelligence’s denial.
Physical Intelligence CEO Karol Hausman responded to the reports with a notably oblique acknowledgment—a Slack message to employees containing a gif of a character shaking her head no, per The Information—that notably avoided a categorical denial. Lachy Groom, Physical Intelligence’s co-founder and an influential Silicon Valley investor-operator, did not respond to TechCrunch’s request for comment.
Why Physical Intelligence Matters to the AI Giants
Physical Intelligence is not a fringe robotics shop. The startup has raised over $1 billion in funding and was reportedly in talks this spring for an additional $1 billion round at an $11 billion valuation, according to The Information. Its π0.5 model is among the more widely used robot control systems in robotics research, giving the startup outsized influence in embodied AI development.
Anthropic has completed four known acquisitions in 2026 alone, prioritizing developer tools and research capabilities. OpenAI has been significantly more aggressive, acquiring at least 17 companies since 2023. Both companies are now preparing for public offerings: Anthropic confidentially filed for an IPO on June 1, according to The Information, followed by OpenAI a week later.
The Robotics Race as IPO Preparation
The pursuit of robotics expertise aligns with a broader industry belief that physical-world understanding may be prerequisite for advanced AI systems. Pure text-based training, by this logic, cannot substitute for embodied reasoning—the capacity to understand physics, causality, and manipulation through interaction with real environments.
TechCrunch reports that OpenAI previously built a robotic hand capable of solving a Rubik’s Cube before shutting down its robotics division in 2021. The group has since quietly resumed operations with a new humanoid robotics lab in San Francisco. Both acquisitions and internal research suggest that embodied AI capabilities are now viewed as competitive moats worth significant capital deployment.
Why This Matters
Enterprise and government buyers evaluating AI platform choices will likely begin scoring robotics integration—or a credible robotics roadmap—as a differentiator within the next 12-18 months. Teams building specialized robotics applications face a narrowing vendor field: Anthropic and OpenAI’s documented interest in the space signals that mid-market robotics startups will face pressure to either join one of the AI giants or raise substantially larger independent rounds to remain competitive. For investors in Physical Intelligence and peer robotics firms, the failed Anthropic deal does not signal retreat from the sector—it reflects hard-fought negotiations over valuation, not capability.
Frequently Asked Questions
Did Anthropic acquire Physical Intelligence?
No. Anthropic and Physical Intelligence held acquisition talks in spring 2026, according to The Information, but no deal closed. Physical Intelligence CEO Karol Hausman denied the reports via a Slack message to employees.
What is Physical Intelligence's π0.5 model?
π0.5 is a robot control model developed by Physical Intelligence that is reportedly widely used in robotics research. The company raised over $1 billion and was reportedly in talks for an additional $1 billion round at an $11 billion valuation this spring.
Why are AI labs acquiring robotics startups?
Both Anthropic and OpenAI view physical-world understanding as potentially foundational to advanced AI systems. Internet-text training alone may be insufficient for developing superintelligent systems that operate in the physical world.