OpenAI Robotics Offers $500K Salary for Data Systems

OpenAI is offering robotics engineers base salaries up to $500,000 annually, with the highest-paid role focused on distributed data systems for training robots, not direct robot construction [1]. This move signals OpenAI's intensified push into "physical AI" and custom robot design, potentially putting it in direct competition with companies like Tesla and Figure AI [1].

Expanding Robotics Ambitions

OpenAI's current robotics team, led by DALL·E and Sora creator Aditya Ramesh, has grown to 27 job listings, up from 11 in May [1]. These roles span hardware design, software development, data collection, and prototyping, indicating a comprehensive effort to build custom robotic systems [1]. The company is hiring for actuator design engineers, who develop robot motors and joint mechanisms, alongside software, firmware, and machine-learning engineers [1]. One role also seeks a lab technician to "develop, build, test, and iterate on robotic systems" [1].

The highest advertised base salary, up to $500,000, is for a machine-learning engineer specializing in distributed data systems, which manage the processing and movement of large volumes of robotics training data [1]. This focus on data infrastructure addresses a key bottleneck in the robotics industry, where companies often need to generate their own training data, unlike large language models that can leverage existing internet data [1].

OpenAI CEO Sam Altman has stated the company "will definitely do a humanoid" robot, alongside other form factors [1]. The company's long-term vision includes "general-purpose robotics" and "AGI-level intelligence," with the goal of "everyone having a personal robot doing anything they need" [1]. This marks a return to robotics for OpenAI, which previously closed a robotic hand project in 2020 [1, 2].

Data Collection and Competitive Landscape

OpenAI has quietly built a humanoid robotics lab in San Francisco, which has quadrupled in size since February 2025 and employs around 100 data collectors [2]. The lab focuses on teaching robotic arms, specifically Franka robots with pincers, to perform household tasks using 3-D printed GELLO controllers [2]. This approach differs from competitors like Tesla and Figure AI, which often use full humanoid robots, motion capture suits, and virtual reality headsets for training [2]. OpenAI's method, which mirrors a 2023 UC Berkeley study, is described as potentially cheaper and more effective for teaching specific human movements to robots [2].

The company's data collection program, which began by teaching a robot arm to put a rubber duck in a cup, has since moved to more complex tasks like folding laundry [2]. OpenAI runs three shifts and dozens of workstations around the clock in San Francisco, with cameras recording both operators and robots, and workers rated on "good hours" of functional training data [2]. The company plans to open a second lab in Richmond, California [2].

OpenAI's venture into building its own humanoid hardware could place it in direct competition with Tesla's Optimus and Figure AI, though OpenAI Startup Fund is an investor in Figure [1]. OpenAI also previously partnered with Figure on robot AI models, a deal Figure CEO Brett Adcock said the company was exiting in February 2025 [1, 2].

What to watch

• Hardware Development: Monitor OpenAI's progress in designing and building custom robot hardware, particularly for actuators and overall system integration, as job postings indicate a focus beyond just software [1].
• Data Collection Scaling: Observe how OpenAI expands its data collection facilities and strategies, as this is a critical bottleneck for training physical AI models [1, 2].
• Competitive Responses: Watch for reactions and new developments from established robotics companies like Tesla and Figure AI as OpenAI escalates its direct hardware and data-driven robotics efforts [1, 2].

OpenAI's significant investment in high-salary robotics roles, particularly for data infrastructure and custom hardware, signals a serious commitment to developing general-purpose physical AI, shifting from its prior focus solely on software models.