A weekend rumor that Anthropic is buying Physical Intelligence swept through AI circles faster than the company's denial could catch it. New reporting indicates the two sides did discuss a deal this spring, and OpenAI's stake in the startup turns a simple rumor into a three-way chess problem.
The AI industry spent this week chasing a rumor with an unusual amount of life in it. Over the past weekend, tech blogger Robert Scoble posted on X that Anthropic is acquiring Physical Intelligence, the San Francisco robotics startup behind the widely used π0 family of robot models. The claim raced across the industry within hours, and it kept spreading even after the startup's chief executive told his own staff it was false. By Monday, The Information had reported the part that gave the story legs: Anthropic and Physical Intelligence did hold acquisition talks this spring.
No deal has been announced, and the official word from Physical Intelligence remains a denial.
How a Single Post Lit the Fuse
Scoble's post attributed the acquisition claim to an unnamed investor, and it offered no documents or deal terms. In a calmer market it might have faded by Sunday. Instead it collided with a moment when both Anthropic and OpenAI are buying companies at a pace the industry can barely track, which made a robotics purchase sound less like a stretch and more like the logical next headline.
Physical Intelligence pushed back through internal channels rather than a press release. According to The Information, CEO Karol Hausman told employees over Slack that the reports were untrue, punctuating the message with a head-shaking gif from The Office. Co-founder Lachy Groom did not respond when TechCrunch requested comment on Monday night, and Anthropic has not publicly addressed the reports.
The Information's account suggests Scoble may have gotten the specifics wrong while still touching something real. Talks between the two companies took place earlier this year, even if no agreement came out of them. That distinction matters: a dead rumor and a dormant negotiation are different stories, and the second kind tends to resurface.
Why Physical Intelligence Is a Prize
Physical Intelligence is barely two years old, yet it sits near the center of the robot-learning world. The company was founded in San Francisco in 2024 by Hausman, a former Google robotics researcher, alongside Berkeley's Sergey Levine, Stanford's Chelsea Finn, and Groom, an investor-operator who has become one of Silicon Valley's most closely watched dealmakers. Its stated goal is a general-purpose foundation model for robots, software that lets machines fold laundry, clear tables, or handle unfamiliar homes without task-specific programming.
Investors bought into that vision early and expensively. The startup raised 400 million dollars in late 2024 at a 2.4 billion dollar valuation, with backers including Jeff Bezos, and its total funding has since climbed past the 1 billion dollar mark. This spring, reports surfaced of talks to raise roughly another billion at a valuation near 11 billion dollars, a round in which Founders Fund was said to be involved.
The technical asset underneath those numbers is the π model line. Its latest public release, π0.5, has become one of the most widely used control models in robotics research, which means a buyer would acquire mindshare across academic and industrial labs along with the team itself.
Two Buying Sprees on the Road to Wall Street
The rumor also landed on fertile ground because of what both frontier labs have been doing all year. By TechCrunch's tally, Anthropic has made four known acquisitions in 2026, while OpenAI has absorbed at least 17 companies since 2023, with targets ranging from developer tooling shops to product-testing startups. The pattern behind the shopping is consistent: converting raw model capability into products and enterprise revenue faster than the rival can.
Hovering over all of it is the IPO race. Anthropic filed confidentially for a public offering on June 1, days after closing a 65 billion dollar funding round that valued it at 965 billion dollars. OpenAI filed its own confidential paperwork about a week later, carrying a private valuation of 852 billion dollars from a March round. The two listings could rank among the largest stock-market debuts in US history, which raises the stakes of every strategic move either company makes between now and the opening bell.
An acquisition of an 11 billion dollar robotics startup weeks before going public would be a bold stroke even by 2026 standards. It would also be a legible one.
Why Frontier Labs Suddenly Want Robots
The strategic logic circulating in AI research circles is simple to state: intelligence trained only on internet text may hit a ceiling, because the physical world contains information no document captures. Under that view, embodiment is a requirement for the superintelligent systems both labs say they are building, and robotics stops being a side quest.
OpenAI's Second Run at Hardware
OpenAI has already lived one full cycle of this argument. Its early robotics team built Dactyl, a robotic hand that solved a Rubik's Cube one-handed in 2019, before the company disbanded the group in 2021 to concentrate on language models. Co-founder Wojciech Zaremba later said the approach lacked ingredients needed for genuine superintelligence. In the interim, OpenAI wrote checks to humanoid startups such as Figure and 1X instead of building in-house.
Then it reversed course. A humanoid effort took quiet shape inside the company starting in 2024, and on May 31 of this year Sam Altman made it official, announcing that OpenAI Robotics had grown out of the company's world-simulation research under Aditya Ramesh and was hiring engineers in San Francisco. The near-term mission is robots that support skilled workers building the physical backbone of the AI boom, its data centers and power grids. The long-term ambition, in Altman's words, is "everyone having a personal robot doing anything they need."
Anthropic's Paper Trail Instead of a Hardware Lab
Anthropic has built nothing comparable. What it has produced is research about robots, published through an internal group that stress-tests frontier capabilities for safety purposes. The most visible example is Project Fetch from last November, an experiment in which Anthropic staff with no robotics background tried to program a robot dog, some with Claude's help and some without, to measure how much the model accelerates novices in the physical world.
A second phase followed this June. According to Anthropic, a newer model completed the same robot-dog tasks roughly 20 times faster than the best human-plus-Claude team from the original study, a data point the company frames as evidence of how quickly AI systems are gaining physical-world competence.
Reading those papers next to the acquisition reports suggests a company that has measured the trend line and knows it lacks the team to act on it. Buying Physical Intelligence outright would collapse years of hiring and engineering into a single transaction.
The OpenAI Complication
OpenAI is more than a bystander in this story. It owns a piece of the company its chief rival reportedly tried to buy, having invested in Physical Intelligence's 2024 funding round.
The entanglement runs deeper than one check. Physical Intelligence's investor roster overlaps heavily with OpenAI's own, including Khosla Ventures and Thrive Capital, and Founders Fund, one of OpenAI's major backers, was reportedly part of the startup's newest round this year. Groom himself is known to be close to OpenAI's orbit.
That web raises a question the public record cannot yet answer: whether OpenAI's early investment came with protective provisions, such as information rights or a right of first refusal over a sale to a competitor. Strategic investors negotiate exactly those terms for exactly this scenario, and if such a clause exists, any Anthropic bid would have to go through OpenAI to succeed.
TechCrunch put those questions to OpenAI on Monday. The company did not respond.
What a Gif Denial Leaves Open
Where things stand is genuinely unresolved. Physical Intelligence says it is not being acquired, and nothing contradicts that as a statement about the present tense. At the same time, the reporting establishes that acquisition talks happened this spring, that a massive new funding round was under discussion in the same window, and that the startup's future is a live topic inside at least two of the most valuable private companies on earth.
The rumor will fade. The structural situation it exposed will not, because if Physical Intelligence ever is in play, the shareholder with the inside position is OpenAI, and nobody outside these companies yet knows what rights that position carries.
