OpenAI Disbands Its Robotics Team for the Second Time

By Elena Vasquez

OpenAI has disbanded its dedicated robotics team, the company confirmed to Fortune, after rebuilding the group in 2024. Fortune reported the exclusive story. OpenAI confirmed that it had dispersed the team’s researchers among other departments. OpenAI first shut the team down in 2020 and rebuilt it in 2024 when the company partnered with Figure. Guanya Shi, an assistant professor at Carnegie Mellon University’s Robotics Institute, told Fortune he believes the reorganization means OpenAI now views robotics as “no longer a special problem.” Fortune linked the move to an industry shift toward general-purpose AI models that control robots, rather than specialized models designed for narrow tasks. An OpenAI spokesperson told Fortune the company remains committed to general-purpose robotics despite the shutdown.

What does the reorganization change?

OpenAI’s confirmation ends a standalone team dedicated to robotics, but it does not amount to an exit from the field. The company’s spokesperson told Fortune that OpenAI remains committed to general-purpose robotics. Researchers from the dissolved team will work in other departments, according to the same report.

That distinction sets the immediate stakes. OpenAI is changing where robotic research sits inside the company, not publicly abandoning the broader goal. Fortune’s account did not identify the receiving departments, the number of affected researchers, or any canceled robotics program. It also did not announce a new model, robot, commercial partnership, or launch date.

According to Fortune, the shutdown completes a three-stage sequence. OpenAI first shut the team in 2020. The company rebuilt it in 2024, when it partnered with Figure. Now, the latest reorganization has disbanded the group again and dispersed its researchers.

That history shows how fluid the company’s organizational approach has been. The 2024 rebuild associated robotics with a named industry partner, Figure, while the latest move folds the work into a broader structure. Fortune framed the change as a response to an industry shift toward general-purpose AI models for robot control.

The newest arrangement also preserves continuity in one important sense. Fortune’s description puts the former team’s researchers inside OpenAI’s broader organization, even without a standalone robotics unit. Their expertise can now inform robotics as part of other departments.

The company has not said those other departments will abandon robotics. Instead, its stated commitment points to a more distributed model. Under Fortune’s account, the reorganization shifts the burden onto OpenAI’s distributed structure to sustain a robotics focus without preserving a dedicated group.

Why are general-purpose models driving the change?

Fortune described an industry-wide move away from creating specialized models for robots and toward using general-purpose AI systems to control them. The report said researchers had already achieved a surprising level of success with that approach. In Fortune’s account, those results helped make a consolidation of effort worthwhile.

The strategic logic is straightforward. A general-purpose model can treat robot control as another application for a broader AI system, rather than requiring a separate model for each kind of robot or task. That is the operating logic Fortune described.

Shi’s interpretation is more pointed. A dedicated team suggests a separate technical problem; a dispersed team suggests the company expects broader systems to absorb robotics as one use case. Shi’s phrase captures that organizational judgment without claiming that robots have become simple to build or deploy.

That reading fits the shift from a standalone team to shared models and departments. If general-purpose systems can handle robotic control, robotics may no longer need a separate technical track within OpenAI, Shi told Fortune. The company’s reorg makes that view visible in its structure.

Fortune’s report offered no benchmark scores, task lists, or named robotics models to measure the claimed success. It also did not identify the researchers who produced the reported results. The report therefore offers organizational evidence and expert reaction, rather than a detailed technical comparison.

Even so, the reorganization carries weight. OpenAI rebuilt a dedicated team in 2024, then dissolved it as the industry shifted toward general-purpose models for robot control. Fortune treated that move as an indication of how successful and powerful LLM-based AI has become.

The caveat is that Fortune does not define what consolidation means beyond dispersing the team. The report gives no account of whether general-purpose models make specialized work unnecessary or simply move it into shared departments. Fortune centered its report on OpenAI’s decision, the team’s earlier history, and the broader technical direction.

What still limits cloud-based robot control?

Shi agrees that the approach has advanced, but he warns that LLM-based AI has a real-world dependency: it runs through the cloud. Under that model, a robot powered by the AI must always remain connected to the network, Shi told Fortune.

By contrast, smaller specialized AI models can run locally on the robot itself, Shi said. The contrast places a basic architecture choice at the center of the debate. General-purpose models favor a broad, shared system; specialized models can keep computation on the device.

That difference matters because the two routes carry different operating requirements. The general-purpose route depends on continuous connectivity for the robot’s AI control. The local route can use the robot’s own hardware without the same cloud dependency for model execution.

Shi’s warning narrows what OpenAI’s reorganization proves. The move suggests confidence in general-purpose models, according to Fortune. It does not erase the connectivity constraint that Shi identified for robots controlled by cloud-based systems.

Fortune did not provide figures for latency, connection reliability, energy use, or local-computing capacity. It also did not say that all LLM-based robots face an outage or that every specialized model avoids broader capability limits. Shi’s point was narrower: cloud dependence creates a practical constraint, while smaller specialized models can operate locally.

In Fortune’s account, the research stakes extend beyond model quality. The source’s central technical debate also concerns where the model runs and whether a robot can function without a network link. OpenAI’s dispersed team now works within a field where general-purpose progress and a basic deployment constraint coexist.

The report offered no evidence on how those two paths compare in production. Its strongest technical claim remains Shi’s warning that LLM-based robot control depends on cloud connectivity, against his contrasting point that smaller specialized models can execute locally.

What does the move mean for robotics?

“There may be different starting points for how to grow intelligence,” Pete Florence, cofounder of Generalist, told Fortune. His quote captures the source’s most balanced conclusion: the industry has more than one route toward capable robotic systems, even as OpenAI consolidates around general-purpose models.

OpenAI’s own sequence reinforces that uncertainty. The company shut down its robotics team in 2020, rebuilt it in 2024 during its partnership with Figure, and then disbanded it again as the industry shifted toward general-purpose models. OpenAI is still committed to general-purpose robotics, its spokesperson said, but the work will no longer sit in a dedicated team.

The next test is practical and technical: whether OpenAI’s dispersed researchers can turn general-purpose models into reliable robot control while addressing the cloud dependency Shi raised. The company has confirmed the organizational change and stated its commitment, but Fortune’s report leaves the model results, deployment details, and future team structure unspecified.