OpenAI is quietly assembling a robotics team, marking a strategic return to physical AI systems after shuttering its robotics division three years ago to focus on large language models.

The artificial intelligence (AI) leader has been recruiting researchers specializing in humanoid robot systems and algorithms that enable machines to navigate and perform tasks in the physical world, according to sources familiar with the company’s efforts and recent job postings reviewed by Wired.

Among the notable hires is Chengshu Li, who joined OpenAI in June from Stanford University, where he developed benchmarks for measuring humanoid robot capabilities in household tasks. His doctoral research focused on robots with human-like upper bodies but wheels instead of legs.

Multiple other researchers from leading robotics laboratories have also joined the company, according to their LinkedIn profiles, while professors report students being recruited for OpenAI’s expanding robotics initiative.

Job listings reveal the company’s technical approach. One opening seeks expertise in teleoperation—a method where human operators control robot limbs while algorithms learn to replicate their movements. The position also requires familiarity with simulation tools like Nvidia Isaac, widely used for training humanoid systems in virtual environments.

Another recent posting called for a mechanical engineer with experience in “high volume” manufacturing systems exceeding one million units, suggesting potential mass production plans or deployment in manufacturing settings.

“Our robotics team is focused on unlocking general-purpose robotics and pushing towards AGI-level intelligence in dynamic, real-world settings,” states the language common across OpenAI’s robot-related job postings.

The pivot represents a significant shift for OpenAI, which abandoned robotics research in 2021 to concentrate on the language models that powered ChatGPT’s breakthrough success. The company had previously achieved notable robotics milestones, including developing an algorithm that solved Rubik’s cubes using a humanlike robotic hand in 2019.

The return to robotics comes as OpenAI faces questions about the trajectory of its flagship models, with reports suggesting GPT-5 has not delivered expected improvements. This broader industry challenge has researchers reconsidering approaches to artificial general intelligence.

“They’ve asymptoted on GPT-5,” said Stefanie Tellex, a roboticist at Brown University. “They need to move towards the physical world.”

OpenAI enters an increasingly crowded field. Humanoid startups including Figure, Agility, and Apptronik have raised significant funding, while tech giants Tesla and Google are also investing heavily in humanoid development.

Venture capitalists have poured more than $5 billion into humanoid startups since early 2024, and Morgan Stanley projects the industry could reach $5 trillion in value by 2050.

Despite the enthusiasm, current humanoid robots remain limited by their inability to operate reliably in unpredictable environments. Success will require AI models capable of processing high-dimensional sensory input and producing precise physical outputs—capabilities that extend far beyond current language model architectures.

“I don’t see them having any magical advantage over anyone else,” Tellex noted, referring to OpenAI’s competitive position in robotics.

It remains unclear whether OpenAI plans to manufacture its own hardware, partner with existing robotics companies, or focus solely on the AI software that powers robotic systems. However, job postings for mechanical engineers with prototyping experience suggest the company may be considering hardware development.

The robotics push reflects a growing belief within OpenAI that achieving artificial general intelligence that matches or exceeds human cognitive abilities may require algorithms capable of meaningful interaction with the physical world, not just processing text and images.

As the AI industry grapples with diminishing returns from scaling language models, OpenAI’s return to robotics signals a potential new frontier in the race toward more capable artificial intelligence.

OpenAI’s push into robotics comes as China makes and installs factory robots at a dizzying pace — two million worked in Chinese factories last year, according to the nonprofit trade group International Federation of Robotics. Nearly 300,000 new robots were installed in China last year, more than the rest of the world combined. The U.S., by comparison, installed 34,000 in 2024.