By Hamza Faisal
OpenAI is building a humanoid robot. That is now confirmed, straight from the top: when asked on the Sources podcast whether the ChatGPT maker would actually construct one, CEO Sam Altman gave the shortest possible answer — "We will." He then went further in an interview with Fortune, setting a rough timeline: "I think we'll have a cool demo in 2027 ... I don't think there will be robots walking the streets for a few years longer."
The confirmation puts the world's most valuable startup squarely into the crowded humanoid race, alongside Tesla's Optimus, Figure, and 1X — and signals that OpenAI believes its frontier models are finally capable of becoming the software layer for general-purpose robots.
The world is designed for people
Altman's case for the human shape is practical rather than romantic. Doors, keyboards, tools, and factory equipment are all built around the human body, he argued — "the world is very much designed for people," so a robot that works in those same spaces should probably match that form factor. As he put it, "matching that form factor seems good."
OpenAI's ambitions may not stop at a single humanoid. Back in early September, Altman answered a question about whether the company's hardware push meant a humanoid or a data-center machine with a simple "Both," later clarifying: "We will definitely do a humanoid. We will do other form factors as well."
The brain, not the body
What the robot looks like is, in Altman's words, "less important than really figuring out ... the brain that makes the robot work." A useful humanoid has to read its cameras and sensors, interpret spoken instructions, plan a task, coordinate two arms and two legs, and recover when the real world throws something unexpected at it.
Picking up a coffee mug sounds trivial until you break it down: the machine must find the handle, estimate the grip, and apply exactly the right force — or you're left with a spill. OpenAI has wrestled with this software-physics gap before: its Dactyl research program trained a robotic hand to manipulate objects entirely in simulation, and the company later applied the approach to a Rubik's Cube, learning firsthand how hard it is to make learned behaviors survive contact with reality.
This is not OpenAI's first brush with robotics partnerships either. The company previously explored acquiring 1X before deciding to go in-house, and it dissolved its partnership with Figure over insufficient physical testing, according to published reports — moves that explain why third-party licensing deals were structurally insufficient and why OpenAI is now building its own hardware.
A hiring spree for hardware people
The operational evidence is already on OpenAI's careers page: the company currently lists around 27 open robotics roles spanning motion control, state estimation, actuator control, whole-robot behavior, and machine-learning infrastructure for physical AI systems. The listings point to custom electromechanical actuators — including motors, transmissions, sensing, and thermal systems — plus manufacturing and supply-chain roles needed to eventually produce hardware at scale.
That hiring pattern suggests OpenAI intends to own the full stack, not just supply the AI brain to someone else's body. No partner has been named, no price discussed, and no release date promised — for now, the 2027 demo is a CEO's target rather than a product plan.
Still, with the company already spending hundreds of billions on compute infrastructure and racing toward a possible IPO, a humanoid division is a natural next frontier: the same intelligence that writes your code may one day pick up your coffee. — Hamza Faisal
