Anthropic and a Billion-Dollar Robot Startup Held Talks This Spring. Here Is What We Know.

Key points

• Anthropic and Physical Intelligence, a San Francisco robotics startup valued at roughly $11 billion, held acquisition talks in spring 2025, according to The Information.
• Physical Intelligence has raised more than $1 billion and its AI model for controlling robots, called pi0.5, is widely used in academic robotics research.
• OpenAI is already an investor in Physical Intelligence, which could give it legal rights to block or match any sale to a rival.
• Anthropic confidentially filed for a stock-market debut on 1 June 2025; OpenAI followed one week later.
• OpenAI has acquired at least 17 companies since 2023; Anthropic has made four known acquisitions this year.

A social-media post from tech blogger Robert Scoble lit up AI Twitter over the weekend with a claim that Anthropic was buying Physical Intelligence. Physical Intelligence's CEO Karol Hausman swiftly denied it, but his denial, sent to staff via a Slack message with a gif of a character from "The Office" shaking her head, was not exactly a full-throated rebuttal.

As first reported by TechCrunch AI, citing The Information, the rumour had roots in reality. The two companies did hold acquisition talks this spring. The deal did not close, but something clearly happened.

So what is Physical Intelligence, and why does it matter? Founded roughly two years ago in San Francisco by investor Lachy Groom, former Google researchers, and professors from Stanford and Berkeley, the company builds AI software that tells robots how to move and interact with physical objects. Think of it as a brain for robot arms. Its pi0.5 model, a system that lets robots learn tasks from demonstrations rather than hard-coded instructions, has become one of the more commonly used tools in robotics research labs. The company has raised more than $1 billion and was reportedly in talks earlier this year for a further $1 billion round at an $11 billion valuation.

Why would an AI lab want a robotics company?

The honest answer is that text on the internet only teaches so much. Researchers increasingly believe that AI systems will need to understand the physical world, weight, friction, cause and effect, to reach what the field calls superintelligence, meaning AI that matches or exceeds human performance across almost any task. Acquiring a team that already solves those problems saves years of work.

OpenAI knows this firsthand. It built a robotic hand that could solve a Rubik's Cube, shut the whole effort down in 2021, then quietly restarted a humanoid robotics lab in San Francisco in 2024. CEO Sam Altman made it official in late May 2025, announcing that OpenAI Robotics was hiring.

Here is the complicating detail for any Anthropic deal: OpenAI is already a shareholder in Physical Intelligence. Large investors often negotiate rights that let them match a competitor's purchase offer, or at least receive advance notice of one. OpenAI did not respond to questions about whether it holds those rights. But if Physical Intelligence is genuinely for sale, OpenAI, with existing shares, existing relationships, and its own robotics ambitions, may have a stronger claim than any outside buyer.

For ordinary readers, the immediate takeaway is straightforward. Both Anthropic and OpenAI are preparing to sell shares to the public for the first time, and both are racing to show investors they can do more than generate text. Robotics is part of that story, and the competition for the companies that know how to do it is already fierce.

One honest takeaway: if you work in a field that involves physical, hands-on tasks, robots are coming for parts of that work. Watching which AI labs acquire which robotics firms is a practical early-warning signal, not just tech gossip.