Sam Altman has confirmed, on the podcast "Sources," that OpenAI will build a humanoid robot. His reasoning is straightforward: the human form factor makes sense because the entire physical world, doorknobs, stairs, tools and workspaces, is already built for people. Altman expects "a cool demo" in 2027, though he was clear that humanoids won't be walking down British or American streets for considerably longer than that.
Brain over body
Altman's framing is that the "brain," meaning perception, reasoning and control, matters far more than the physical design of the body. That is a pointed comment in a field where much of the public attention goes to how a robot looks and moves rather than how well it understands what it is doing. OpenAI has posted 27 robotics job listings covering motion control, state estimation, actuators and manufacturing supply chain, suggesting the company is building out hardware capability in earnest, not just talking about it.
The company's stated plan has two horizons. In the short term, general-purpose robots to help skilled workers build infrastructure. In the long term, a personal home robot. That puts OpenAI in direct competition with Figure AI and 1X Technologies, two companies it has previously backed and funded, an awkward position that says something about how central robotics has become to OpenAI's long-term strategy.
A crowded, fast-moving field
OpenAI is a late entrant to a field that is moving quickly. Global humanoid shipments hit roughly 19,100 units in the first half of 2026, up 272% year on year, driven largely by automotive and logistics buyers. China's Ministry of Industry and Information Technology reports more than 400 humanoid models and over 40,000 units produced domestically in the same period, forecasting six figures for the full year. Tesla told shareholders in an October filing that Optimus V3 is "ready for mass production," with the first lines under construction. Figure's Figure 03 robot completed a nine-day, 348-mile treadmill endurance test. And SoftBank is reportedly close to acquiring Marc Raibert's Robotics and AI Institute from Hyundai, adding to its existing Boston Dynamics and ABB Robotics holdings.
Our take
Altman's "brain, not body" comment is the most useful part of this story, because it is an implicit dig at a field that has produced plenty of viral demo videos and comparatively few robots doing genuinely useful, repeatable work. If OpenAI's bet is that reasoning and real-world understanding are the harder, more valuable problem, that lines up with what has actually slowed humanoid robotics down: not the hardware, which companies like Tesla and Unitree have shown can be built, but the judgement needed to operate reliably outside a lab. A 2027 demo is a long way from a product, but the fact that OpenAI felt it needed to say this publicly tells you how seriously the rest of the industry is taking humanoid robots right now.