More Tesla Profits From Distributed Computing Than Full Self Driving in 2025

by

Elon Musk said that the entire fleet of Tesla’s that are on the road can be combined into the largest AI inference supercomputer. Most Tesla EVs have hardware 3 with 144 Trillion operations per second of compute. Tesla will end 2023 with 5.5 million Tesla’s on the road. There will be about 8 million Tesla’s on the road at the end of 2024. Tesla started producing Tesla Model Y vehicles with HW4 in May 2023. HW4 should have two to three times the compute of HW3. The total fleet could have about 2000 ExaOPS of compute by the end of 2024. Each customer with HW3 could generate $400 of compute value each year and each HW4 customer could generate $1000 of compute value each year. The 3 million HW4 vehicles would make the equivalent of 7.5 million HW3 vehicles. Tesla would have the equivalent of 13 million HW3 vehicles. IF Tesla makes $400/year per HW3 car then this would be about $5 billion per year at the start of 2025 if the entire fleet could be used. This could be $7 billion for all of 2025.

Tesla should have 100 ExaFLOPS of Dojo and Nvidia training compute by the end of 2024.

IF we use the FSD subscription rate of $2400 per year. If 30% of the global fleet had purchased or subscribed to FSD then this would $720 of revenue per car.

Goldman Sachs is assuming that the $2400 per year FSD remains and there was fleetwide 45% adoption for about $1100 per year per car.