Elon Musk has shared additional estimates on the future of Tesla’s supercomputing cluster, Dojo, and highlighted roughly how much the company expects to spend on Nvidia products in 2024.
Discussion about Nvidia’s supercomputers arose on Tuesday after Tesla needed to have some of the company’s chips to X and xAI locations, due to a lack of storage, according to Musk. Following the reports, Musk later went on to post on X about Tesla’s purchases from Nvidia this year, which he estimates to be between $3 billion and $4 billion, as well as other insights about the company’s AI supercomputing needs.
Musk also noted earlier in the day that Tesla’s Giga Texas south extension is nearing completion, at which point it will house 50,000 Nvidia H100 chips used for FSD training.
Of the roughly $10B in AI-related expenditures I said Tesla would make this year, about half is internal, primarily the Tesla-designed AI inference computer and sensors present in all of our cars, plus Dojo.
For building the AI training superclusters, NVidia hardware is about…
— Elon Musk (@elonmusk) June 4, 2024
In a follow-up conversation to the post, Musk also responded to one X user asking if there was a “realistic pathway” toward Tesla producing its Dojo supercomputers at enough volume to rely more heavily on these for training compute than on Nvidia hardware. Musk responded by clarifying that training compute at Tesla is actually a small fraction of what’s needed, especially compared to inference compute needed for the FSD system.
“Training compute for Tesla is relatively small compared to inference compute, as the latter scales linearly with size of fleet,” Musk wrote. “Perhaps the best way to think about it is in terms of power consumption.”
“When the Tesla fleet reaches 100M vehicles, peak power consumption of AI hardware in cars will be ~100GW. Training power consumption is probably <5GW. These are very rough guesses.
“Obviously, 5GW of AI training compute is enormous by current standards, but is only about 5% of total Tesla AI compute.
While Musk calls it a “long shot,” he says it could be possible for production of Tesla’s Dojo to someday exceed that of Nvidia’s.
“There is a path for Dojo to exceed Nvidia,” Musk added. “It is a long shot, as I’ve said before, but success is one of the possible outcomes.”
Tesla self-driving development gets huge compliment from NVIDIA CEO
What are your thoughts? Let me know at zach@teslarati.com, find me on X at @zacharyvisconti, or send us tips at tips@teslarati.com.
