Tesla hosted a session with Wall Street analysts this week, led by Investor Relations head Travis Axelrod at the company’s office in Palo Alto, California. During the session, Axelrod discussed more details about the upcoming Robotaxi service, progress on Optimus, and improvements in its Dojo supercomputing platform.
Robotaxi Launch
Tesla is moving forward with its long-awaited robotaxi service, with the initial rollout planned for Austin, Texas next month. However, according to Morgan Stanley analyst Adam Jonas, Tesla’s first robotaxi deployment will be open to a limited group of participants, operate with a small fleet, and rely on teleoperation support to ensure safety.
Jonas confirmed that the robotaxi pilot remains a “go,” but expectations are being set for a cautious and methodical start. Tesla plans to deploy 10 to 20 vehicles initially, with the service also initially launching as invite-only.
Jonas also confirmed Tesla is planning to use teleoperators—remote workers who can intervene when a vehicle encounters challenging driving situations. Jonas emphasized that there will be “plenty of tele-ops” involved in the Austin launch, with safety being the top priority. He quoted Tesla staff saying, “We can’t screw up,” reflecting the company’s understanding of the scrutiny this pilot will face.
Tesla started hiring for teleoperators, both for its robotaxi service and Optimus, last year.
Optimus Robot Production
Currently Tesla is building out a production line that would give the company the capacity to build 1,000 Optimus units per month. While the completion of that line is still several months away, according to CEO Elon Musk, Tesla is targeting substantial increases with second- and third-generation production lines. The second-gen line aims for 10,000 units per month, while the third-gen line is expected to deliver a tenfold increase over that, potentially scaling to 100,000 units per month.
Tesla hopes to build thousands of Optimus robots by the end of 2025, with plans to offer investor tours of low-scale manufacturing operations in the fourth quarter.
Interestingly, Tesla noted that the neural networks powering Optimus are “far larger” than those used in its vehicles. This highlights the robot’s increased complexity and the significant computing power required to operate in dynamic, human-centric environments.
Dojo Supercomputer
Tesla also provided an update on its Dojo supercomputer, which is central to training the neural nets for both autonomous driving and robotics. Dojo 2 (D2) is slated for release in 2026, promising greater efficiency and lower costs compared to Nvidia solutions, though not expected to match Nvidia’s top-end performance immediately.
Looking further ahead, Dojo 3 is planned for 2028, with Tesla aiming to narrow the performance gap and achieve significant cost advantages. Tesla believes scaling its computing capabilities by 10x and acquiring cheaper AI chips could be a key unlock for future growth.
