OpenAI has restarted its robotics business after a six-year hiatus, betting on assistive robots in the short term.
On June 1st, Sam Altman, the CEO of OpenAI, posted a recruitment notice on a social platform, officially announcing OpenAI's entry into the physical robotics field. Altman stated that the company is forming a new team called "OpenAI Robotics" and is publicly recruiting full - stack hardware, operations, systems, and machine learning engineers. The goal is to "program and build robots that are truly useful to society together."
According to Altman's description, OpenAI's robotics strategy has both short - term and long - term goals. In the short term, OpenAI focuses on developing robots that can assist skilled workers in building future infrastructure. In the long run, the company envisions that everyone will have a personal robot that can meet various needs in the future.
Altman revealed that the decision to enter the robotics field is based on the rapid development of an internal research project at OpenAI called "Worldsim". This project evolved into OpenAI Robotics in the past year and is led by Aditya Ramesh, the vice - president of research at OpenAI and a core developer of the text - to - image model DALL·E and the video - generation model Sora. The foundation of this project lies in the in - depth integration and collaborative design of robotics hardware research and machine learning research.
OpenAI's return to the robotics field is actually a "homecoming". In the early days of the company's establishment, robotics technology was an important direction for its exploration of Artificial General Intelligence (AGI). From 2016 to 2019, OpenAI successively launched the reinforcement learning benchmark environment OpenAI Gym and the open - source robot simulation platform Roboschool, and successfully developed a dexterous robotic hand named Dactyl.
In 2019, OpenAI trained an AI system using reinforcement learning and "Automatic Domain Randomization" (ADR) technology, enabling a humanoid robotic hand to successfully solve a Rubik's Cube. This research proved the feasibility of the technical route of training in a simulation environment and then transferring the capabilities to real robots. However, due to the scarcity of robot training data and slow iteration at that time, while text and image data on the Internet were abundant and easy to obtain, OpenAI made a strategic decision around 2020: to disband the robotics team and concentrate resources on the research and development of large language models represented by the GPT series. This decision ultimately led to the birth of ChatGPT.
In the following years, OpenAI triggered a global frenzy for large models with its ChatGPT series of products and became the world's most highly valued AI unicorn. According to multiple media reports, OpenAI secretly submitted a draft IPO prospectus on May 22nd and plans to go public as early as September 2026. In the latest round of financing completed in March this year, OpenAI's valuation reached $852 billion. Institutions such as Deutsche Bank predict that its valuation at the time of listing may exceed $1 trillion, and the fundraising scale may reach $60 billion, potentially becoming one of the largest technology IPOs in the history of the U.S. public market.
However, OpenAI also faces huge loss pressure. The company expects to incur a loss of approximately $14 billion in 2026, and its cash consumption will further increase. It is not expected to achieve positive cash flow until 2030 at the earliest. Its gross profit margin is only about 33%, and the high cost of AI model inference is the main reason for eroding profits.
During the years after disbanding its in - house robotics team, OpenAI did not completely abandon the robotics field. Instead, it adopted a "diversified investment" strategy through its venture capital fund and invested in several robotics startups, including the Norwegian humanoid robotics company 1X Technologies, the American humanoid robotics star company Figure AI, and Physical Intelligence.
The most notable cooperation was with Figure AI in February 2024. At that time, OpenAI not only participated in Figure AI's Series B financing of $675 million but also announced the development of a dedicated multimodal AI model for Figure's humanoid robots. Just 13 days after the cooperation began, the Figure 01 humanoid robot equipped with the OpenAI model demonstrated smooth natural language interaction, object recognition, and autonomous operation capabilities.
However, this cooperation lasted for less than a year. In February 2025, Brett Adcock, the founder of Figure AI, officially announced the termination of cooperation with OpenAI and decided to independently develop an end - to - end robotic AI model. The main reason for the breakdown of the cooperation was the divergence in technical routes. Figure believes that general large models cannot meet the hardware requirements of robots, and an end - to - end model with vertical integration must be built. This also prompted OpenAI to "revive" its robotics team after six years and upgrade robotics from an "investment" to an "internal strategic business".
Meanwhile, this is also OpenAI's way of depicting a new growth curve for the capital market before its IPO. It shows investors the company's grand vision of expanding from pure software to a combination of software and hardware and from the virtual world to the physical world, hoping to use the story of "embodied intelligence" to alleviate the market's concerns about its business model's sustainability and huge losses.
OpenAI's advantage in entering the robotics field lies in its globally leading large AI model capabilities, especially the "world model" for understanding and simulating the physical world. Its technical path may differ from that of many companies that start with the hardware itself. Instead, it follows the logic of "build the brain first, then grow the body", that is, first let the AI understand physical laws through a powerful world model, and then transfer its capabilities to physical robots. If this idea of defining hardware with software and algorithms succeeds, it may reshape the R & D model of the robotics industry.
This article is from "Jiemian News", author: Li Kefeng, published by 36Kr with authorization.