The OpenAI Humanoid Robot Is Finally Official
The OpenAI humanoid robot has moved from rumour to roadmap, and the confirmation came straight from the top. For more than a year the question was not whether OpenAI would build a machine, but when the company would stop pretending it wasn't. That pretence ended this week. Asked directly whether the firm would develop its own robot, Sam Altman gave a two word answer that removed all ambiguity. "We will."
Altman framed the decision around a simple observation about the physical world. The human form, he argued, makes sense because the world is very much designed for people, and matching that form factor seems good. Factories, kitchens, warehouses and offices are already shaped around arms, legs and hands, so a machine that fits those spaces inherits decades of built environment for free. That logic is why Tesla, Figure and a dozen venture backed startups have chased the same silhouette, and now the OpenAI humanoid robot joins them.
What makes this different is who is saying it. OpenAI spent years as an investor and software partner to the robotics industry rather than a competitor. The company now intends to design the hardware itself, and it has put a date on the calendar. Altman told Fortune he expects a cool demo in 2027, while cautioning that robots will not be walking city streets for a few years beyond that.
The Brain Matters More Than The Body
Altman was careful to redirect attention away from the metal. The body, he suggested, is the less important problem. The real challenge for any OpenAI humanoid robot is figuring out the brain that makes it work. It is a revealing emphasis from a company whose entire advantage sits in software rather than servos.
The point holds up under scrutiny. Two arms and two legs are worthless if the system does not know what to do with them. Humanoid hardware has improved rapidly, with several firms now walking machines off production lines, yet none has cracked general purpose autonomy in messy, unstructured environments. Picking up an unfamiliar object, recovering from a stumble, and reasoning about a half finished task all remain unsolved at scale. That is a perception and planning problem, and it is exactly the territory where OpenAI believes its models give it an edge.
This also explains the structure of the effort. The OpenAI humanoid robot programme is led by Aditya Ramesh, the researcher behind DALL-E and a contributor to the Sora video team. Handing a robotics division to a generative modelling veteran signals that OpenAI sees embodiment as a continuation of its model work, not a separate hardware gamble. The bet is that intelligence, not actuators, is the moat.
A Hiring Spree Built For Hardware
The careers page tells the rest of the story. OpenAI currently lists 27 robotics roles, and the job descriptions read like the staffing plan for a manufacturer, not a lab. The company wants engineers for custom electromechanical actuators, covering motors, transmissions, sensing and thermal systems. It is recruiting for motion control, state estimation, whole robot behaviour and the machine learning infrastructure that physical AI demands.
Crucially, several listings reference manufacturing and supply chain work for eventual large scale production. That is not the language of a research prototype. Building the OpenAI humanoid robot in house means owning the hardest parts of the stack. Actuators alone can account for 40 to 60 percent of a humanoid's bill of materials, and custom actuator development typically takes two to four years.
The money reflects the ambition. Reported pay runs to around 500,000 dollars for machine learning infrastructure engineers and between 255,000 and 325,000 dollars for robotics software engineers. The division, announced in May 2026, operates from a 202,000 square foot facility in Richmond, California, wired with 14,000 amperes of power and capable of running dozens of robot workstations that use teleoperation to generate training data.
From Backer To Rival
The strategic shift is sharpest when measured against OpenAI's own history. The company was an early supporter of the humanoid field before deciding to compete with it. Figure AI ended its collaboration with OpenAI in early 2025, with founder Brett Adcock arguing that large language models were becoming smarter yet more commoditised, which pushed Figure to build its own Helix foundation model. Figure has since raised enormous sums and reached a valuation in the tens of billions.
OpenAI also led a funding round in 1X Technologies, whose NEO robot is aimed squarely at the home. Those relationships now look less like partnerships and more like scouting trips. Having studied the field from the inside, the makers of the OpenAI humanoid robot concluded they would rather own the full stack, from the reasoning model down to the limbs.
That decision drops a heavyweight into an already crowded arena. Tesla has worked on its Optimus programme for years, XPENG has shown a humanoid walking off its assembly line, and Chinese manufacturers are pushing aggressively on cost. The arrival of the OpenAI humanoid robot turns a competitive market into a referendum on whether software intelligence or hardware mastery wins the physical AI race. For more coverage of that contest, see Blumefield.
Why The Stakes Run Higher Than A Demo
A flashy 2027 demonstration will be the easy part. The harder test is whether the OpenAI humanoid robot can become something people trust in their homes and workplaces. Altman has sketched a two stage ambition. In the near term, the goal is general purpose robots that help skilled workers build infrastructure. In the longer term, he imagines a personal robot for almost anyone, an idea he has floated in his writing on what he calls the gentle singularity.
There are reasons for caution beyond engineering. OpenAI's hardware chief departed in early 2026 over concerns tied to defence work, and the company has not named a successor. The OpenAI humanoid robot still has no disclosed design, no name, no manufacturing partner and no deployment site. Safety questions around autonomous machines operating near people remain largely unanswered, and regulators have barely begun to engage.
Still, the signal matters. When the most closely watched company in artificial intelligence commits to a body as well as a brain, the rest of the industry recalibrates. The OpenAI humanoid robot may not reach a factory floor until the end of the decade, yet the race it accelerates is already reshaping where the next hundred billion dollars of AI investment will flow. Readers tracking that money can follow Blumefield's ongoing analysis.