Sam Altman has confirmed OpenAI will build a humanoid robot. Speaking to Fortune, he said the human form makes sense because, in his words, the world is very much designed for people, and that he expects a demo in 2027. He added a caveat in the same breath: he does not think there will be robots walking the streets for a few years longer than that.
His framing of the problem is the interesting part. Altman said the exact physical design matters less than the brain that controls it, which is a direct inversion of how most humanoid programmes are pitched. OpenAI is building hardware regardless - the careers page lists 27 robotics roles covering custom actuators, sensing, thermal systems, manufacturing and supply chain - but the stated bottleneck is control, not mechanics.
There is precedent inside the company. OpenAI's Dactyl research trained a robotic hand entirely in simulation and later applied the approach to solving a Rubik's Cube, before the robotics team was wound down. Restarting that work puts OpenAI against Tesla's Optimus and XPENG's IRON, both of which have been demonstrating hardware publicly for longer. No partners or funding figures have been disclosed, and OpenAI has not said what the 2027 demo will show or whether anyone outside the company will get access to it.