Tesla Reportedly Plans ‘Robot School’ for Optimus

leaderobot.com reported on October 22, 2026, that Tesla CEO Elon Musk outlined plans to build a “robot school” for Optimus during the company’s third-quarter earnings call. According to the report, Musk also said it could take more than a year to reach the goal of producing 1 million units annually. He was also said to have indicated that 10,000 to 20,000 initial units would be needed to secure training data.

If Tesla plans to use 10,000 to 20,000 initial units for training before reaching its target of producing 1 million units annually, data acquisition could be a bottleneck for its robotics business ahead of sales. The reported plan suggests that operating systems to support repeated learning in real-world environments, as well as manufacturing capacity, could become a competitive factor. This raises the question not only of the robots’ capabilities, but also of how quickly data from real machines can be fed back into model improvements. Based on the account of Tesla’s statements, progress on the production plan should be assessed separately from the development of the training infrastructure.

What is new here is that Tesla’s humanoid robot plans put the development of a training infrastructure ahead of mass production. However, the concept itself extends an existing approach: using real machines to collect operational data. In other words, the focus is less on manufacturing facilities or sales plans than on first increasing the number of machines and establishing a learning cycle. In related coverage, Physical AI News has reported on developments such as AgiBot to start ‘AI WAVE’ on October 22; six G2 units to operate around the clock at convenience stores from November 2 and AgiBot launches a full-size humanoid robot smart retail warehouse at Wuhan Jiangtan, reflecting a continuing trend of using real-world operating environments for training and deployment. Tesla’s “robot school” appears to frame that trend as a larger-scale self-learning system. However, this article reports on remarks and does not provide specific details about the facility or how it would operate; readers should consult the original source for further information. At the industry level, the issues extend beyond robot development to the number of machines available to generate training data, the sites for repeated testing, and the operational systems needed to feed results back into model updates. If the figure of 10,000 to 20,000 initial units is accurate, the machines could take on a stronger role as data-generation tools, moving beyond the prototype stage. At the same time, the target of 1 million units annually remains a distant one. A key question is when the training fleet will be separated from the production line for commercial units. Further details are awaited on production volumes, training-data collection methods, safety assessments and which functions will be commercialized first.

For Japanese companies, areas such as component supply during production ramp-up, testing equipment and maintenance could become relatively more important alongside the robots themselves. If the plan involves operating large numbers of real machines to collect data, supply capacity for precision components, manufacturing equipment and maintenance parts could become a competitive issue.