Tesla activates Fremont line for Optimus robot assembly

Tesla has begun assembling its Optimus humanoid robots on a high-volume production line within its Fremont, California facility. The company confirmed in its third-quarter shareholder letter that these units represent the first batch to meet commercial product standards.

The new line occupies the floor space previously dedicated to Model S and Model X vehicles. Tesla completed the conversion of that area in 46 days during the spring. The infrastructure is currently sized to support annual output of one million robots. Elon Musk stated on Wednesday's earnings call that the necessary equipment for that capacity is already installed in the plant.

Production is expected to climb toward 20,000 units per week as the line ramps up. Musk described this growth trajectory as a classic S curve, noting that the initial slope will be shallow. The engineering team is currently refining the robot's artificial intelligence, electromechanical components, and service life. Tesla plans to finalize this work by the end of next year, at which point it will begin delivering Optimus units to external companies for repetitive and hazardous tasks.

A key operational challenge for Tesla is generating sufficient training data for the robots. Unlike its automotive fleet, which provides continuous data from millions of vehicles, Optimus does not yet have a large deployed base. To address this gap, the company is implementing a program called the Optimus Academy. This system will deploy between 10,000 and 20,000 robots to practice diverse tasks. The data collected from these units will feed directly into the next iteration of the model's software.

Every robot produced on the Fremont line will serve as a source for this training data. Tesla also plans to upgrade the hardware by integrating its AI5 chip into Optimus units next year. Musk indicated that a new video demonstrating the robot performing tasks driven by AI, rather than preprogrammed motions, should be released within one to two months. He contrasted this approach with rival humanoid demonstrations, arguing that scripted routines often fail in real-world conditions.