Photo: igovar igovar / Pexels
The AI giant is rebuilding its robotics division with plans for construction-assisting robots by 2027 and consumer models further down the line
OpenAI CEO Sam Altman announced on May 31, 2026 that the company is launching a major push into robotics, with ambitions to build humanoid robots that can handle real-world tasks.
Altman laid out a two-phase vision: first, general-purpose robots designed to assist skilled workers in infrastructure and construction, and later, personal robots built for everyday consumer use.
From Dactyl to round two #
This isn’t OpenAI’s first attempt at robotics. The company previously ran a robotics division that produced Dactyl, a robotic hand capable of manipulating objects with impressive dexterity. That effort was shut down around 2020-2021 as the company shifted resources toward its language model work, the bet that eventually produced ChatGPT.
The rebuild started quietly in February 2025, when OpenAI began reassembling its robotics capabilities. The company established a new lab in San Francisco and is planning to expand with a second facility in Richmond, California. The lab has reportedly quadrupled in size since its inception.
Altman has signaled an aggressive hiring campaign across hardware engineering, operations, systems design, and machine learning.
The technical approach draws from OpenAI’s internal work on world simulation, essentially training AI systems to understand and predict physical environments. Altman links this shift in strategy to advancements in embodied intelligence and simulation technology, representing a meaningful evolution from the Dactyl era, when the focus was narrower and more task-specific.
The competitive landscape is getting crowded #
OpenAI is entering a robotics race that already has some formidable runners. Tesla’s Optimus humanoid project has been progressing through iterative prototypes. OpenAI itself holds investment stakes in Figure and 1X, two humanoid robotics startups that have attracted significant venture capital.
Altman suggested that robots capable of performing useful real-world tasks could emerge by 2027.
Why construction comes first #
The decision to target infrastructure and construction before consumer applications is pragmatic. Construction sites are controlled environments with defined tasks, high labor costs, and chronic worker shortages. A robot that can assist a skilled electrician or carry materials on a job site doesn’t need to navigate the unpredictable chaos of a family kitchen.
The broader humanoid robotics market has attracted projections ranging into multi-trillion-dollar territory over the coming decades.
Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our