OpenAI has officially disbanded its robotics team, just five months after relaunching it.
The company quietly dissolved its independent robotics division, dispersing its engineers and researchers across the company’s broader artificial intelligence (AI) groups.
The restructuring, confirmed by an OpenAI spokesperson to Fortune, represents a strategic shift toward using general-purpose AI models to control physical machines rather than building specialized robotic “brains” from scratch.
Under the new structure, OpenAI’s robotics researchers now report to Chief Research Officer Mark Chen within the company’s main research organization. Meanwhile, the applied robotics and hardware teams have been absorbed into the larger Applied division under OpenAI President Greg Brockman.
The move happened abruptly enough to leave a visible footprint at the Conference on Robot Learning (CoRL) in Austin, Texas, this month. Although OpenAI was listed on the conference floor plan, Fortune reported that booth #208 sat empty without a display or company logo, save for a few former team members who still drew crowds of curious attendees.
Despite the optics, OpenAI framed the decision as an acceleration of its hardware ambitions rather than a retreat. The company is still developing its own humanoid robot — with CEO Sam Altman targeting a public debut in 2027 — and stated that general-purpose robots remain central to its long-term vision.
This is the second time OpenAI has shuttered a dedicated robotics unit.
The company disbanded its original team in 2020 to focus on large language models, a pivot that eventually produced ChatGPT. It revived the group in 2024 alongside a partnership with robotics startup Figure, though Figure exited the deal in February 2025 to build its own in-house AI.
At the heart of the latest reorganization is a growing consensus in the AI sector: frontier models built for language, vision, and code may be capable of powering “embodied AI” directly.
Earlier this year, OpenAI researchers believed robotic progress depended on building upon video-generation systems like Sora, which require an implicit grasp of physics.
However, recent breakthroughs with OpenAI’s general-purpose model, GPT-6 Astra, changed that calculus.
In a recent NVIDIA Corp. study, an Astra-powered agent given camera feeds and a control interface achieved a 74.8% success rate on a simulated manipulation benchmark — outperforming the 51% score of top specialized robotic models.
“It signals that OpenAI sees robotics as no longer a special problem,” said Guanya Shi, an assistant professor at Carnegie Mellon University’s Robotics Institute, comparing robotics to downstream applications like coding or legal software.
The shift also aligns with mounting financial pressure on OpenAI to streamline operations and cut what Altman previously termed “side quests,” a push that already led the company to shelve Sora.
Facing reported quarterly operating losses of $10.9 billion ahead of a new funding round and a potential early 2027 IPO, OpenAI is consolidating computing power around core, potentially profitable products.
Still, robotics experts caution that general-purpose models will not immediately replace specialized engineering. Shi noted that large cloud-hosted models are slow and require constant internet connectivity, leaving a critical need for smaller, onboard models that can run locally on a robot’s hardware.
Russ Tedrake, an MIT professor and CEO of Walden Robotics, pointed out that dedicated robotics startups still hold a vital edge: real-world deployment experience on factory floors, generating site-specific data that foundational AI labs cannot easily replicate.


