The Zhitong Finance App learned that CITIC Securities released a research report saying that Fortune reports that OpenAI has reorganized an independent robotics team and merged into two major departments of basic model research and application. Recent industrial progress in the brain-building industry is the core driving force for OpenAI to make this adjustment, and this adjustment also continues to reflect the industry trend of cutting-edge AI laboratories speeding up the physical AI layout. From a technical architecture perspective, as a number of cutting-edge pedestal models such as GPT6-Astra and Fable 5.2 show outstanding performance in the embodying task set, the integration trend of the S1/S2 layer of the embodied model may accelerate. From the domestic hardware supply chain level, it is recommended to focus on: 1) motors; 2) speed reducers; 3) screws; 4) robotic arms; 5) assemblies.
CITIC Securities's main views are as follows:
OpenAI reorganizes the robot team organization structure: the universal cutting-edge model can now directly assume the robot's “brain” function.
On November 13, Fortune reported that OpenAI had reorganized its independent robotics team. OpenAI confirmed the adjustments and emphasized that robots are a core component of the company's long-term vision. The restructuring aims to accelerate general robot research and development. The direct basis for this decision is that since mid-2026, general-purpose cutting-edge models such as the GPT-6 Astra have shown the ability to be directly applied to robot control: agents built on GPT-6Astra have achieved a 74.8% success rate on simulated operation benchmarks, while the best dedicated robot model on this list is only 51.0%. The bank believes that this indicates that the paradigm of “embodied intelligence as a downstream application of general intelligence” has been verified by leading global model manufacturers at the organizational level, and its significance is similar to the reconfiguration path of the general model for vertical tools such as programming, law, and finance. In terms of overall industry trends, the trend of cutting-edge AI laboratories entering the field of embodied intelligence was further determined and strengthened.
OpenAI completely integrates robot capabilities into the two main lines of “general model research” and “application implementation.”
According to Fortune, the robotics research team was merged into the company's core research team and reported to Chief Research Officer Mark Chen; the applied robotics and hardware team was merged into the larger Applied organization and reported to the company's president Greg Brockman. This is the third time in six years that OpenAI has adjusted its robotics strategy. Its repeated trajectory of “build-withdrawal-reconstruction-re-integration” is what the bank believes is essentially a necessary organizational adjustment to follow the pace of development of the robotics industry as a whole. Looking at history, OpenAI had a robotics team in the early days. It was abolished in 2020. Co-founder Wojciech Zaremba explained at the time that robot training data was scarce and that AGI was more efficient through other paths; in 2024, the company rebuilt an independent robotics team and announced cooperation with humanoid robot startup Figure to develop robot AI models, but the collaboration ended in February 2025, and Figure switched to self-research. In 2026, OpenAI once again formed a robotics team. Beginning at the end of May, OpenAI began recruiting a large number of robot-related positions, covering the complete hardware stack of actuator design, simulation accuracy, data acquisition and operation, and electrical engineering. Overall, the bank believes that OpenAI has made this organizational adjustment in the context of seeing the rapid development of overall industry trends since this year.
Frontier AI laboratories have entered into embodied intelligence to be further established, and the trend of integration at the S1 and S2 levels has further emerged.
In addition to OpenAI, the bank also saw the recent acceleration of cutting-edge AI laboratories entering the physical field. At the same time, Google DeepMind released a dedicated vision-language-action model Gemini Robotics 2.5 to partners, while Anthropic Fable 5.2 also performed well on Robodojo; Nvidia released a next-generation open source personalization model GR00T-2 during the CORL period. Architectural level: The embedded intelligent system is traditionally built in layers between System 1 (fast execution) and System 2 (slow thinking), and this round of changes shows that the two layers are moving closer in both directions — from the top down, generic models such as Astra do not require task-specific fine-tuning, and are penetrating S1 functions downward; from the bottom up, vFlash recently released by Nvidia is accelerating the speed of planning and generation at the S2 level. The bank believes that in the short term, the general model and the dedicated model are more likely to form division of labor and collaboration, and the pace of integration will depend on the speed of convergence of the general model in the three dimensions of delay, accuracy, and long-range task performance, and the economic progress of deployment in actual scenarios.
Risk factors:
The improvement in motion accuracy and reasoning efficiency of the general model fell short of expectations; the synergy between the general model and the special strategy fell short of expectations; the migration of simulation results to the real machine fell short of expectations; the reduction in robot deployment costs and the progress of commercialization fell short of expectations.