OpenAI announced on May 31, 2026, that it is launching a dedicated robotics division. CEO Sam Altman confirmed that the company's world simulation research program — led by DALL-E creator Aditya Ramesh — has evolved into a standalone unit called OpenAI Robotics. The company is hiring for 11 specialized roles and intends to build robots from mechanics to software.
Key takeaways
- OpenAI Robotics evolved from an internal world simulation research program led by Aditya Ramesh, creator of the DALL-E model series
- 11 open roles in San Francisco include Actuator Design Engineer, Simulation Realism Engineer, and Operations Manager for data acquisition
- Short-term goal: robots that support skilled workers in building infrastructure
- Long-term goal: personal robots accessible to everyone for general-purpose tasks
- The announcement follows the split with Figure AI in March 2026 — Brett Adcock claimed OpenAI lacked commitment to physical hardware
From world simulation to physical hardware
OpenAI has spent years building brains for other companies' robots. The lab partnered with humanoid manufacturers, supplied language models, and supported the ecosystem through investments. But the partnership model had a ceiling: without direct contact with hardware, it is difficult to understand where simulation ends and reality begins.
The strategy shift is decisive. According to Altman, the world simulation research program — developed internally over the past year — has grown into a full-fledged robotics division. Altman described progress as "rapid" and grounded in "co-design between robotics hardware and ML research."
Aditya Ramesh, previously associated exclusively with generative AI through the DALL-E model series, now leads a unit working with metal and actuators. This signals that OpenAI intends to treat embodied AI — robots operating in the physical world — as the next frontier after LLMs, and wants to define it on its own terms.
What the job listings reveal
Altman's public statements are broad — the open roles say far more about what OpenAI is actually building. Three key positions define the scope of ambition:
The Actuator Design Engineer will lead development of custom electromechanical actuators from early architecture through prototype validation. Required expertise spans torque density, efficiency, bandwidth, and thermal architecture trade studies. This is not off-the-shelf integration — it is designing robot mechanics from scratch.
The Simulation Realism Engineer must close the sim-to-real gap: the divergence between a model's behavior in simulation (MuJoCo, Isaac Sim) and its performance on physical hardware. The role demands deep knowledge of rigid and soft-body dynamics and contact mechanics.
The Operations Manager for Data Acquisition is responsible for managing a large field workforce and fleet infrastructure across multiple sites to collect real-world physical interaction data. The scale described in the listing suggests OpenAI plans to build its own robotics data pipeline at massive scale.
Other open roles include a 3D Printing Lab Technician, DAQ Station Engineer, Electrical Engineer, and distributed ML and data systems engineering positions.
Context: the Figure AI split
In March 2026, Figure AI CEO Brett Adcock publicly described the breakdown of the partnership with OpenAI. According to Adcock, Figure's internal AI teams had "run circles around" OpenAI, and the San Francisco lab could not maintain the "daily, weekly" presence required on physical hardware — preferring simulation instead.
The inflection point came when Sam Altman called Adcock to disclose that OpenAI was contemplating an internal robotics project. Adcock took that as a signal to end the partnership. Months later, Altman's May 31 announcement confirms that call was not speculation.
The two sides tell very different stories. Figure claims it won the technical race inside the partnership. OpenAI — through its new hires in actuator design and physical prototyping — signals that it absorbed the lessons from that collaboration and is now building an integrated hardware-software loop in-house.
Open questions
OpenAI's entry into physical manufacturing raises serious operational questions. Scaling hardware production lines is fundamentally different from scaling compute clusters for model training — it requires supply chain management, spare parts logistics, and field service operations.
The job listings mention "a broad range of robotic form factors" without specifying whether OpenAI intends to build a classic bipedal humanoid, a wheeled mobile manipulator, or an entirely novel industrial architecture. The choice of form factor will determine the entire distribution and monetization strategy.
What is certain: OpenAI wants to own the physical stack from the ground up. That puts it on a collision course with companies like Figure AI, Boston Dynamics, and Agility Robotics — the very same partners it previously funded or supported.
Why it matters
OpenAI is the company that defined modern LLMs and mainstreamed chat AI. Now it is declaring that it wants to do the same for robotics — and this time, it wants to control both the brain and the body of the machine. If the plan succeeds, OpenAI will not be one of many AI suppliers for robots. It will be a robot manufacturer with its own AI stack.
That reshapes market dynamics. Robotics startups that built products on top of OpenAI models may suddenly find themselves in direct competition. Investors exposed to both ecosystems will need to reconsider their risk positions. Academic labs and research teams that today rely on OpenAI's API may gain an alternative in the form of a full physical AI platform.
The question is no longer whether OpenAI will enter hardware — it has. The question is how quickly it can close the experience gap separating a new entrant from companies that have spent years working with physical metal.
What's next?
- Hiring for 11 roles in San Francisco is active — early prototypes will be the first test of the declared "rapid progress"
- OpenAI has given no timeline or date for a first physical prototype demonstration — the industry will watch Altman and Ramesh's announcements at upcoming conferences
- The target robot form factor (humanoid, wheeled manipulator, other) remains unspecified — the choice will define the competitive strategy against Figure AI, Boston Dynamics, and Agility Robotics
