OpenAI starts hiring for its first robotics hardware team

OpenAI isn't content just living on cloud servers anymore.

After months of vague hardware rumors and partnerships with companies like Figure, the AI giant is officially assembling an in-house robotics hardware crew. Caitlin Kalinowski, who joined OpenAI about two months ago after leading Meta’s AR glasses division, announced on X that the company is hiring its very first hardware robotics engineering roles.

Building eyes, motors, and labs

These aren't junior positions. Kalinowski posted openings for two senior technical lead roles alongside a TPM manager to run operations.

First up is an electrical sensing engineer tasked with designing the proprietary sensor suites for OpenAI's upcoming machines. The second is a mechanical product design engineer focused on the nitty-gritty physical hardware: gears, actuators, motors, and mechanical linkages.

Meanwhile, the TPM manager will be throwing together a brand-new data-gathering lab. According to the job descriptions, this role includes overseeing logistics and making sure contractors have the setups required to pull real-world physical data from active test environments.

The listings spell out OpenAI's ambition in plain English: unlock general-purpose robotics, chase AGI in dynamic real-world settings, and explore diverse robotic form factors across the entire model stack.

Moving past partnerships

Up until now, OpenAI's physical footprint was strictly indirect. The company made headlines by powering humanoid robots from startup Figure with multimodal models, and rumors have circulated about consumer hardware ventures alongside ex-Apple designer Jony Ive.

Bringing actuator and motor design under OpenAI's own roof signals a massive escalation. It indicates that relying on third-party hardware to collect physical interaction data is no longer enough. Notably, building native robotics hardware will almost certainly put OpenAI on a collision course with partners like Figure, mirroring its complicated frenemy relationship with Microsoft.

Why it matters

We think this move shows OpenAI hitting the hard limits of synthetic and web-scraped data. If you want true general-purpose intelligence, your model has to touch, drop, bump, and manipulate the physical world, and you cannot get that data purely from YouTube transcripts.

Controlling both the neural networks and the mechanical actuators gives OpenAI the kind of end-to-end vertical integration Apple mastered in consumer tech. If their software knows the exact physical feedback curve of an in-house motor, the intelligence loop closes faster.

The catch? Hardware is notoriously brutal. Precision manufacturing, supply chains, and component wear-and-tear are entirely different beasts from running GPU clusters. For robotics startups that assumed OpenAI would remain a friendly model provider, this is a clear wake-up call. The giant is stepping out of the cloud and straight into the workshop.