Tesla AI5 chips will be used for volume production in 2026, while Teslabot will start production in 2025. The cost of a humanoid robot is estimated to be around $20,000. Tesla FSD for Cybertruck is expected to be released in mid-August.
The power demand issues with the GPU and Dojo AI training clusters require significant electrical power, dropping up to 50 megawatts in milliseconds. This means more computing power is needed during training to compensate for the lower compute power in cars.
When Tesla FSD enables unsupervised driving, it will enable cars to go from 10 hours of use per week to 50-100 hours per week. The value of 10 billion robots at $20,000 each would be approximately $200 trillion.
For AI robot training, Tesla uses sensors attached to vehicles and requires millions of cars to train its FSD. Waymo/Cruise are struggling against FSD due to the vast amount of training data available. However, unless Tesla trains robots en masse using human volunteers or paid trainers, it may lose the battle to a competitor willing to do so.
Tesla’s vehicle robots are the foundation for building humanoid robots, but the situations aren’t perfectly analogous. While vehicles need to handle nearly every possible scenario, humanoids can simply stop if they’re confused and don’t require individual knowledge of most tasks.
Humanoid training will mainly involve dozens of humans teleoperating robots for tasks like doing chores or buying groceries. The car can act as a Starlink hub for giving the robot full connectivity.
The key to Tesla’s valuation is its software expertise, which may be its biggest advantage over non-Chinese companies and competitors.
Source: https://www.nextbigfuture.com/2024/07/elon-musk-talks-teslabot-being-part-of-a-200-trillion-market-and-more-updates.html