The Anthropic-Physical Intelligence Rumor Explained: Robotics, IPO Pressure, and AI Strategy

What the Rumor Got Right

Tech blogger Robert Scoble posted over the weekend that Anthropic had acquired Physical Intelligence, a San Francisco-based robotics AI company. The claim spread quickly. Physical Intelligence’s CEO, Karol Hausman, told employees via Slack that the reports were not true — the denial delivered, notably, via a GIF of a character from The Office shaking her head.

That is not exactly a vigorous rebuttal.

According to reporting from The Information, Anthropic and Physical Intelligence did hold acquisition talks this spring. Scoble may have gotten the specifics wrong, but the underlying signal — that something had happened between the two companies — appears to have been accurate.

—

Who Is Physical Intelligence?

Physical Intelligence, often abbreviated as π, was founded roughly two years ago by co-founder Lachy Groom alongside former Google researchers and professors from Stanford and Berkeley. It has raised more than $1 billion in funding and was reportedly in talks earlier this year for an additional $1 billion round at an $11 billion valuation.

Its P0.5 model is considered one of the more widely used robot foundation models in academic and applied robotics research. That is not a minor detail — it means Physical Intelligence has already achieved meaningful distribution in the field, not just lab credibility.

—

Why Robotics, Why Now

The timing is not coincidental. Both Anthropic and OpenAI confidentially filed for IPOs within a week of each other in early June, setting up what could be two of the largest U.S. stock market debuts in recent history. Both companies are under pressure to demonstrate not just current capability, but a credible long-term roadmap.

Physical-world understanding is increasingly viewed as a prerequisite for the next tier of AI capability. The argument is straightforward: no amount of internet text can substitute for the kind of grounded, embodied reasoning that robotics demands.

OpenAI’s own trajectory illustrates the point. It built a robotic hand capable of solving a Rubik’s Cube, shut the entire robotics group down in 2021, then quietly rebuilt a humanoid robotics lab in San Francisco before CEO Sam Altman made it official in late May — announcing “OpenAI Robotics” with a near-term focus on infrastructure robots and a stated long-term goal of a personal robot for every individual.

Anthropic has taken a different path. Rather than building hardware infrastructure, it has published research stress-testing frontier model capabilities in physical contexts. Project Fetch, run internally last November, tested how effectively 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 prior year.

That is meaningful progress. It is also not the same as having a dedicated robotics team, years of embodied AI research, and a deployed model already used across the research community.

—

The Complication: OpenAI Is Already a Stakeholder

Here is where the situation becomes structurally interesting.

Physical Intelligence’s early investor base includes Khosla Ventures and Thrive Capital — both of which are also significant OpenAI investors. Founders Fund, another major OpenAI backer, was reportedly involved in Physical Intelligence’s most recent funding round. And OpenAI itself is a direct investor in Physical Intelligence.

That means OpenAI is not a peripheral observer in this situation. It is a shareholder in a company that its primary competitor was reportedly in acquisition talks with.

Strategic investors sometimes negotiate protective provisions — information rights, rights of first refusal on a sale to a competitor — precisely for scenarios like this one. Whether such provisions exist in OpenAI’s investment terms with Physical Intelligence is not publicly known. OpenAI did not respond to questions on the matter.

What is clear is that if Physical Intelligence is genuinely in play, OpenAI has structural proximity to the outcome that Anthropic does not.

—

Acquisition Pace as a Signal

Anthropic has made four known acquisitions in 2024. OpenAI has made at least 17 since 2023. The gap in acquisition velocity reflects a broader difference in how the two companies are building — OpenAI moving faster and wider, Anthropic more selectively.

Buying Physical Intelligence would represent a significant acceleration for Anthropic. It would acquire not just a model, but a team with deep robotics expertise, an existing research footprint, and a product already embedded in the research community. That is years of compounding work that cannot be replicated quickly from scratch.

—

What to Watch

The acquisition talks reportedly took place this spring. Whether they are ongoing, concluded, or superseded by other developments is not confirmed. Physical Intelligence’s response to the rumor was ambiguous enough to leave the question open.

A few things are worth tracking:

• Whether OpenAI moves to acquire or deepen its relationship with Physical Intelligence, given its existing stake and competitive incentive.
• How Anthropic’s robotics research trajectory develops in the absence of an acquisition — Project Fetch’s second-phase results suggest Claude‘s physical-world utility is improving rapidly.
• IPO implications: Both companies are now operating under the scrutiny of prospective public investors. A major robotics acquisition, or a failed attempt at one, will be read as a signal about strategic direction.

—

The core tension here is not really about one rumor or one company. It is about whether physical-world AI capability becomes a decisive differentiator in the next phase of the frontier model race — and which lab gets there first, through research, acquisition, or both.

Anthropic appears to be asking that question seriously. Whether it can act on the answer is a different matter.