Guotai Haitong has released a research report noting that the production timeline for Tesla's Optimus is now largely settled. The key variable determining whether it moves from "capable of mass production" to "capable of doing real work" is shifting toward the speed of building real-machine data collection and the data flywheel. The firm believes the demand elasticity in the data collection segment is more likely to materialize first on the side of domestic humanoid robot manufacturers and data collection centers.
The main points from Guotai Haitong Securities Co., Ltd. are as follows. With the production timeline settled, the data pathway becomes the next key variable. On October 21, 2026, Tesla held its third-quarter 2026 earnings call. The company disclosed that early production of Optimus at its Fremont factory has already begun, and most of the equipment for a production line planned at one million units of annual capacity is already in place. Musk expects it will take at least a year, possibly longer, to ramp up to roughly 20,000 units per week, mainly because some designs are still being iterated. External deliveries are expected to begin in about a year, with the first batch targeting dangerous and repetitive industrial scenarios. In addition, construction has started on a second Optimus factory in Austin.
On the data pathway, Musk proposed the concept of an "Optimus Academy" in an interview in February 2026: deploying at least 10,000 to 20,000 Optimus robots to repeatedly practice in the real world and conduct "self-play." The company also disclosed on its second-quarter 2026 earnings call that early output from Fremont will be assigned to the Optimus Academy to accelerate the collection of training data. The firm believes the production timeline locks in delivery certainty, while whether reusable generalization capability can be formed after the robots are delivered depends on how quickly the data flywheel starts.
Why data collection is the key constraint: real-machine data can only be actively "manufactured." Data sources for embodied intelligence can be roughly divided into three layers. The top layer is real-machine data, which has the highest precision and also the highest cost. The middle layer is simulation data, which is low-cost and scalable but suffers from a simulation-to-reality gap (Sim2Real gap). The bottom layer is internet video and human behavior data, which has broad sources and strong generalization but lacks 3D motion and force-sensing annotations. Autonomous driving can rely on on-road fleets to naturally feed back data, but humanoid robots have no such existing fleet and can only actively build data sources: first build scenarios, then build data, and only then discuss external sales. At the same time, the task space for humanoid robots is open-ended. Data demand therefore is not a one-time collection effort but continues to expand with task complexity.
Dual-wheel drive of "robot components plus domestic data collection infrastructure." On the robot side, benefiting from the certainty of production ramp-up, dexterous hands, precision transmission, and force/tactile sensing remain the core variables determining ramp-up speed. On the data collection side, demand-side participants need to be differentiated. Tesla mainly relies on self-play by its own robots in the real world, and its training focus has already shifted toward human video, so its dependence on external data collection equipment is relatively limited. Therefore, order elasticity in the data collection segment is more likely to materialize first in China. Domestic robot manufacturers and local data collection centers and training grounds are in the stage of "first build scenarios, then build data," and segments such as motion capture and teleoperation equipment, data annotation and simulation synthesis, and force/tactile and 3D vision are expected to enter a procurement cycle before large-scale external sales of complete machines. In addition, if AI5 is introduced into Optimus around April 2027 as planned, it will be able to support higher degrees of freedom and more complex tasks, while also raising requirements for data quality.
Tracking points: first, the pace of weekly production ramp-up and the rhythm of design freeze at Fremont; second, the deployment scale and training progress of the Optimus Academy; third, the structure of the first batch of external industrial customers in 2027; fourth, the timing of AI5 being introduced into Optimus; and fifth, the tendering and order implementation of domestic data collection centers. Risk warning: embodied intelligence/AI development may fall short of expectations; geopolitical risks.
Comments