For most of its history, OpenAI has been a software company. ChatGPT, its API, and its coding agents live entirely in the digital world — no motors, no metal, no moving parts. That is now changing. OpenAI has confirmed it is moving beyond simply supplying the “brain” for other companies’ robots and is building a humanoid robot of its own.
From Investor to Builder
OpenAI’s involvement in robotics did not start overnight. The company had previously taken a hands-off approach, investing in humanoid robotics startups such as Figure AI and 1X Technologies and licensing its AI models to power their machines. A 2024 partnership with Figure AI paired OpenAI’s models with Figure’s hardware, and by 2026 Figure’s robots were running pilot programs in BMW factories using OpenAI’s language models for instruction-following.
That arrangement has now shifted. Rather than only providing the intelligence behind someone else’s hardware, OpenAI has decided to build and ship a physical robot under its own name — a much bigger bet that puts it in direct competition with the very companies it once backed.
Straight From the CEO
The clearest confirmation came from CEO Sam Altman himself. Speaking on the Sources podcast on September 2, 2026, Altman was asked whether OpenAI’s robotics push was aimed at a humanoid robot or a data-center machine. His answer: “Both.” He added plainly, “We will definitely do a humanoid. We will do other form factors as well.”
Altman also offered a glimpse of his long-term vision, suggesting that eventually “everyone should have a personal robot” at home. But he was careful to frame near-term efforts around industrial and infrastructure work — helping skilled workers build things — while positioning household robots as a longer-term goal rather than an immediate product.
Notably, Altman suggested that building the physical robot body may be the easier half of the problem. He emphasized that figuring out the “brain” that actually makes a robot capable and useful matters more than the hardware itself.
What’s Actually Confirmed vs. What’s Still Unknown
It’s worth separating hype from substance here:
Confirmed:
- OpenAI intends to design and build its own humanoid robot hardware, not just license AI to others
- The company has been hiring for hardware-focused roles covering custom actuators, firmware, thermal systems, and manufacturing
- Near-term ambitions are centered on industrial and infrastructure applications
Still unknown:
- A ship date or launch timeline
- Manufacturing partners
- Pricing
- The final physical form factor of the robot
Reporting also indicates some internal turbulence: former Meta hardware executive Caitlin Kalinowski, who had been leading early robotics efforts and partnerships at OpenAI, reportedly resigned in March 2026 amid concerns about the company’s work with the Pentagon and questions around guardrails for surveillance and autonomous weapons use — a reminder that embodied AI raises governance questions that pure software products don’t.
Why This Matters — and Why Now
The timing isn’t random. The humanoid robotics field has moved rapidly from research demonstrations to real-world deployment in 2026:
- Tesla’s Optimus is being used inside Tesla’s own factories, trained on motion-capture data from human workers
- Figure 02 is running production pilots at BMW plants
- Boston Dynamics’ electric Atlas is operational inside Hyundai plants
- 1X’s Neo and Agility’s Digit are advancing their own commercial pushes
- Chinese manufacturers now account for the overwhelming majority of humanoid robot shipments — global shipments hit roughly 19,100 units in just the first half of 2026, up 272% year-over-year
Google DeepMind (with its Gemini Robotics work), Nvidia (building tools for “physical AI”), and Tesla are all competing to define this space. Analysts see the humanoid robotics market potentially reaching trillions of dollars by mid-century, and the strategic logic for OpenAI is straightforward: owning both the AI software and the physical hardware means capturing more value, controlling the full feedback loop of real-world data, and avoiding dependence on hardware partners.
The Risk
Building robots is a fundamentally different business from building software. OpenAI would be stepping into a space where well-funded hardware efforts — including Google’s now-abandoned Boston Dynamics investment and Amazon’s discontinued Astro home robot — have struggled or failed outright. Manufacturing, supply chains, safety certification, and physical reliability are challenges that don’t map neatly onto OpenAI’s existing expertise in training large language models.
Still, with Altman’s direct confirmation, OpenAI has signaled that it sees the “brain” of a robot — the reasoning, perception, and instruction-following capability — as inseparable from the body it inhabits. Whether that bet pays off will likely become clearer over the next one to two years, as OpenAI’s plans move from hiring pages and podcast comments toward an actual product.
This article reflects publicly available reporting and statements as of September 2026. OpenAI has not announced a launch date, price, or final design for its humanoid robot.