OpenAI has disbanded its stand-alone robotics team. Fortune’s Wen Shao reported it as an exclusive on Thursday and the company confirmed the reorganisation when asked: robotics researchers now sit inside the main research org under chief research officer Mark Chen, while applied robotics and hardware staff report to President Greg Brockman’s Applied organisation. No robotics team, as such, exists any more.
The company’s framing is acceleration, not retreat. A spokesperson told Fortune that its frontier models — including September’s GPT-6 Astra — have shown they can be applied directly to robots, and that this discovery is what motivated the consolidation. Altman’s target of a first public humanoid demonstration in 2027 stands, according to the company. The reorganisation arrived late enough that OpenAI had already registered a booth at this week’s CoRL conference in Austin; Fortune found booth #208 empty when the floor opened, though a few former team members still drew a crowd.
The bet, and the evidence behind it
The underlying claim is now the industry’s most consequential fault line: general-purpose models built for language, code, and vision can also command robots — embodied AI without a dedicated robotics brain team. As AIBase’s round-up of the reporting noted, the internal view is that robotics is simply a vertical application of the same general intelligence that already powers coding and legal tools.
Fortune’s evidence for the claim is a benchmark comparison: researchers gave Astra camera images, robot state, and a control interface, and an Astra-powered agent scored 74.8 per cent on a simulated manipulation leaderboard versus 51.0 per cent for the best specialised robotic models. Those are simulation numbers — Fortune is careful to say so — but the gap runs the “you need spatial-intelligence specialists” argument in reverse at a moment robotics startups have raised billions on that premise.
Not everyone is convinced the consolidation settles anything. Guanya Shi of Carnegie Mellon told Fortune he reads the move as a signal that OpenAI sees robotics as “no longer a special problem” — but he still sees a need for smaller, specialised models onboard the robot, because a cloud-hosted LLM is slow and needs constant connectivity. Russ Tedrake, the MIT professor who now runs Walden Robotics, offered the counterweight labs can’t buy: robots already gaining real work experience on factory floors, gathering deployment data that becomes more valuable as base models improve, not less.
This is the second time, and the pattern is the point
The history rhymes. OpenAI’s first robotics team wound down in 2020 — Robot Report archived cofounder Wojciech Zaremba’s reasoning that limited robot data made other routes to AGI more efficient. The rebuilt team of 2024 announced a Figure partnership that Figure terminated in February 2025 in favour of its own models. This week’s disbanding is the third act, and it is the first motivated by success rather than scarcity: in 2020 the team died for lack of data; in 2026 it died because the general model could already do the job.
That distinction is why this reorganisation matters more than a hiring shuffle. When we covered the GPT-6 Astra launch in September, the open question was whether a frontier LLM would stay a brain-on-call or displace the specialised policies robotics companies were building. Nvidia’s GR00T 2 work bet on dedicated physical-action models. OpenAI just answered with org charts: its robots will run on the same model family as everything else, managed by the same research org, with a 2027 demo as the public deadline.
The risk OpenAI is running has a name: cloud dependency. Shi’s point about connectivity is not a detail — it is the difference between a robot that folds laundry in a basement and one that waits on a WebSocket. Labs that control frontier models have every incentive to believe intelligence is enough. The people shipping robots into kitchens, warehouses, and (as of this week) stadium undercrofts in Haiti keep insisting that the last ten per cent of reliability lives at the edge. Both can be right, which is exactly why the next eighteen months of robot-brain architecture will be fought harder than any robotics funding round.
Sources: Fortune (primary, Wen Shao, 13 Nov 2026), AIBase, California Outlook, The Robot Report.
