OpenAI confirms it will build its own humanoid robot
OpenAI plans to build a humanoid robot as well as the physical AI intended to control it, CEO Sam Altman confirmed during an appearance on the Sources podcast. Asked whether the company was developing humanoids or machines for data centers, Altman answered, “Both.”
“We will definitely do a humanoid. We will do other form factors as well.”Sam Altman, OpenAI CEO
The statement confirms that the OpenAI humanoid robot program includes hardware, rather than only a general purpose robotics model intended for third party platforms. It is not yet a product announcement. According to Forbes, OpenAI has disclosed no prototype, delivery schedule, production target or manufacturing partner.
Altman argued that intelligence remains the harder problem. He also gave the standard rationale for a human form: homes, workplaces, tools and equipment are largely designed around human reach, movement and manipulation. He said industrial infrastructure and manufacturing would come before household adoption, although he expects personal robots eventually to become widely owned.
Hiring points to custom humanoid hardware
OpenAI’s recruitment activity indicates a substantial internal hardware effort. Forbes found 19 open robotics positions in San Francisco, including four focused on actuators and a technical program manager responsible for actuator work. Other listings cover PCB layout, thermal simulation, firmware, prototyping, inventory and commodity management.
One actuator design position calls for responsibility across motors, transmissions, sensing and thermal architecture, with an emphasis on torque density and manufacturability. The engineer would also work with manufacturing partners on production ready designs. That is deeper hardware development than would be required to place an OpenAI model on an existing commercial chassis.
Actuators are among the most difficult parts of a humanoid to source and manufacture at scale. Citing McKinsey, Forbes reported that actuators can account for 40 to 60 percent of a humanoid robot’s bill of materials. Precision components including harmonic drives and roller screws also face concentrated supply, while rare earth magnets remain heavily dependent on Chinese processing.
Owning the body creates a robotics data loop
Building and operating robots would give OpenAI direct access to physical interaction data, including what a robot attempted, how hard it gripped and what moved as a result. Forbes reported that insufficient data contributed to OpenAI ending its previous robotics project in 2021. A company controlled robot fleet could provide the failures, corrections and sensor records needed to train embodied models.
The strategy also places OpenAI in closer competition with vertically integrated humanoid developers. OpenAI participated in Figure AI’s $675 million funding round in 2024, but Figure ended the collaboration in February 2025 and chose to develop its robot intelligence internally. Figure CEO Brett Adcock said at the time that embodied AI required integration of both hardware and AI.
Manufacturing remains a separate test. Forbes noted that Tesla began producing Optimus units roughly five years after announcing the program, yet had not shipped customer robots and missed its target of about 10,000 units in 2025. Elon Musk has described Optimus as Tesla’s hardest product to scale because its components and supply chain are largely new.
OpenAI’s actuator recruitment suggests the company understands that constraint, even if Altman describes the robot brain as the central challenge. OpenAI has not disclosed the humanoid’s design, intended first application or whether final assembly will be handled internally or by a manufacturing partner.
Sources: Forbes, Gadget Review
