On June 1, OpenAI CEO Sam Altman dropped a job posting on social media that officially put the company back in the physical robotics game. Altman said the company is building a new team called “OpenAI Robotics” and is looking for full-stack hardware, operations, systems, and machine learning engineers—people who want to “code and build robots that are actually useful for society.”
According to Altman, OpenAI’s robot strategy has two horizons: near-term and long-term. In the short run, they’re focused on creating robots that can help skilled workers build tomorrow’s infrastructure. Way down the line, the vision is that everyone will have a personal robot that can do pretty much whatever they need.
Altman shared that this move into robotics came out of a sudden acceleration in an internal project called “Worldsim.” Over the last year, that effort evolved into OpenAI Robotics, led by Aditya Ramesh—the company’s VP of research and a core developer behind DALL·E and Sora. The whole thing rests on a deep integration of hardware robotics research and machine learning, with both sides talking to each other from day one.
Actually, this isn’t OpenAI’s first time in robotics. It’s a comeback. Way back when the company was just starting out, robotics was a big part of their journey toward artificial general intelligence (AGI). Between 2016 and 2019, they released the OpenAI Gym (a reinforcement learning benchmark), the open-source robot simulation platform Roboschool, and built a clever robotic hand called Dactyl.
In 2019, OpenAI trained an AI system using reinforcement learning and Automated Domain Randomization (ADR), which allowed a human-like robotic hand to solve a Rubik’s Cube. That research showed it’s possible to train in simulation and then transfer those skills to a real robot. But back then, robot training data was scarce and slow to gather, while text and image data from the internet was abundant and easy to get. So around 2020, OpenAI made a tough call: they disbanded the robotics team and poured all their resources into developing large language models—the GPT series. That decision eventually gave us ChatGPT.
Fast forward a few years, and ChatGPT has turned OpenAI into the most valuable AI unicorn on the planet. Multiple news outlets report that on May 22, OpenAI confidentially filed a draft IPO prospectus, with plans to go public as early as September 2026. In the latest funding round back in March, the company was valued at $852 billion. Some banks like Deutsche Bank predict the IPO valuation could top $1 trillion, with a potential raise of $60 billion—making it one of the biggest tech IPOs in U.S. history.
But it’s not all sunshine and roses. OpenAI is bleeding cash—expecting a loss of about $14 billion in 2026, with cash burn getting worse. They don’t expect to turn cash-flow positive until 2030 at the earliest. Their gross margin is only about 33%, mainly because running those big AI models costs a fortune.
Even after shutting down their in-house robotics team, OpenAI never really left the robotics space. They just switched to an investment strategy through their startup fund, placing bets on several robotics companies: Norway’s 1X Technologies, the U.S. humanoid robot star Figure AI, Physical Intelligence, and others.
The most buzzworthy deal was with Figure AI in February 2024. OpenAI not only joined Figure’s $675 million Series B round but also agreed to build a custom multimodal AI model for Figure’s humanoid robot. Just 13 days later, the Figure 01 robot—powered by OpenAI’s model—showed off smooth natural language conversations, object recognition, and autonomous manipulation skills.
But that partnership lasted less than a year. In February 2025, Figure AI founder Brett Adcock called it quits, deciding to build their own end-to-end robot AI model instead. The split reportedly came down to technical disagreements: Figure thought general-purpose large models couldn’t match the specific hardware needs of robots, and that a vertically integrated end-to-end model was the only way to go. That breakup is what pushed OpenAI to revive its robotics team after six years, moving robots from “investment bet” to “internal strategic priority.”
This move also helps OpenAI paint a fresh growth story for the capital markets ahead of its IPO. It shows investors the company’s grand vision—going from pure software to a mix of hardware and software, from the virtual world to the physical one. With the “embodied intelligence” narrative, they hope to ease concerns about business model sustainability and those massive losses.
OpenAI’s edge in robotics comes from its world-leading AI models, especially its ability to understand and simulate the physical world—the “world model.” Their approach might be pretty different from companies that start with hardware. Think of it as “build the brain first, then grow the body”: first use a powerful world model to make AI understand physics, then pour that intelligence into a real robot. If this software-and-algorithm-first strategy works, it could totally reshape how the robotics industry builds things.
