Report: OpenAI signals timing for humanoid robot reveal

leaderobot.com reported on October 23, 2026, that OpenAI had signaled when it would unveil a humanoid robot. We have not been able to verify the original source, so please refer to it for details.

The report that OpenAI has indicated a timeline for a humanoid robot strengthens the view that established large-model companies are moving into embodied AI. What the report suggests is not whether a finished product exists, but that companies with computing and data infrastructure may expand into robot control.

More important than OpenAI's indication of a timeline for a humanoid robot is the renewed focus on the prospect of established large-model companies with existing infrastructure entering embodied AI. The primary source also cites speculation about GPT-6 Astra in September 2026 and OpenAI collecting tens of millions of hours of embodied-operation data through in-house efforts and external sourcing. This suggests the focus is shifting from hardware alone to the combination of models, data and control. A related article from this newspaper, Google Cloud AI Research releases Harness²; Google DeepMind research raises repair rate from 24.4% to 48.3%, showed that results depend not only on the model itself but also on how external components are assembled and execution systems are designed. The OpenAI report may likewise indicate that the competitive focus will include not just the robot's capabilities but also which data is used at each level of control. Related coverage from this newspaper, such as [Aishida officially launches humanoid robot ASD-D3 in Shanghai, demonstrating manufacturing capacity at the 10,000-unit scale](/news/ai-shi-tooru-shanhai-de-ningyou-robotto-asd-d3-wo-seishiki-k-b1f56c), has already highlighted mass production and supply capacity. If model companies such as OpenAI enter this arena, competitive advantage is unlikely to depend on software intelligence alone. In terms of industry structure, the roles of major companies with computing resources and data-collection capabilities and businesses strong in hardware design, mass production and operations could shift. How much embodied data companies collect in-house and how they supplement it externally may also affect training efficiency and reproducibility. The supplementary source's description of starting with applications for factories and data centers suggests that deployment in controlled environments, rather than in homes, will be the initial focus. Questions to watch include whether the 2027 indication refers to a prototype, the range of tasks that can be controlled and whether the methods for obtaining embodied data can support real-world operations.

In Japan, if the initial focus is on limited applications for factories and data centers, industries strong in hardware design, component supply and deployment operations could become key points of competition. Conversely, few companies can collect embodied data or develop model-based control in-house, raising the question of which role they will take on: software or hardware.