- Published on
OpenAI just proved the idea is the cheap part
- Authors

- Name
- AbnAsia.org
- @steven_n_t
Listen to this article · OpenAI just proved the idea is the cheap part
0:00 / 0:00
Watch this article · OpenAI just proved the idea is the cheap part
OpenAI has disbanded its stand-alone robotics team. The researchers moved to the main research organisation under chief research officer Mark Chen; the applied robotics and hardware teams went to the larger Applied group under president Greg Brockman. The company told Fortune the change accelerates robotics rather than abandoning it.
The detail that makes it concrete: OpenAI had already booked booth #208 at CoRL, the robotics conference in Austin this week. When Fortune visited, the booth was empty, with no logo and no demo.
What the reorganisation is actually claiming
The claim is not that robots matter less. It is that building a separate team to make specialised robot models was the wrong shape of work — because the general models already do it.
Camera in
Researchers gave the model images from the robot’s camera and information about the machine’s state.
Control out
A control interface lets the general model command actions directly — no robotics-specific training run in between.
Result
In an Nvidia study, an Astra-powered agent reached 74.8% success on a simulated manipulation benchmark, against 51.0% for the best specialised models on that leaderboard.
Guanya Shi of Carnegie Mellon's Robotics Institute reads the move as OpenAI deciding robotics is "no longer a special problem" — a downstream application of general intelligence, like coding tools or legal research.
The counter-argument is about execution
Russ Tedrake of MIT, who also runs Walden Robotics, gave Fortune the opposing case: robotics companies hold something the labs do not — robots already working in real places, on real factory floors, generating deployment data. And better foundation models make that site-specific data more valuable, not less, because the robot learns from it faster.
Pete Florence of Generalist added the honest version of the same point: it is not clear that any existing model wins robotics in the long run. Shi raised a practical constraint — large models are slow and cloud-hosted, so a robot needs a connection unless a smaller model runs on board.
Our view: the idea is cheap, execution decides
A good idea is worth having. It is also the cheapest thing in the building.
Anyone can copy an idea — read the paper, watch the demo, hire two engineers who were there. What nobody can copy is the part that takes years: the deployed fleet, the failure logs, the iteration on real hardware, the boring discipline of shipping a thing that keeps working when the lighting changes and the bottle moves.
That is exactly what this reorganisation shows, from both directions. OpenAI is arguing that the idea of a separate robotics team no longer earns its keep, because the general model already carries it. Tedrake is arguing that the execution — the fleet in the field — is the asset that compounds. Both are saying the same thing about where value sits: not in the concept, in the carrying out.
The temptation, when a demo looks like this, is to conclude the hard part is over. It is the opposite. When a capability becomes generally available, execution stops being an advantage and becomes the only thing left to compete on.
What to watch
Altman has targeted a first public humanoid demonstration in 2027. Between now and then, the questions that decide this are not about model benchmarks: whether OpenAI ships hardware itself or licenses, whether the robotics startups' deployment data turns into a lead the labs cannot buy, and whether the cost discipline behind this reorganisation — $10.9 billion of reported quarterly losses, a funding round, a possible IPO in early 2027 — survives contact with a hardware programme.
Source: Fortune, 13 November 2026, reporting OpenAI's confirmation of the reorganisation, the CoRL booth, the Nvidia study figures, and interviews with Guanya Shi, Russ Tedrake and Pete Florence. The 2020 wind-down, the 2024 team rebuild, the Figure partnership ending in February 2025, the 2027 demonstration target and the reported quarterly loss are as described in that report. The section marked "Our view" is our editorial position.
Contact: [email protected]