What just happened? Elon Musk has confirmed reports that 12,000 Nvidia H100 GPUs intended for Tesla's EV production were diverted to X. Musk says logistical challenges were behind the diversion as Tesla "had no place to send the Nvidia chips to turn them on, so they would have just sat in a warehouse."
On Tuesday, CNBC reported that Musk had diverted the 12,000 H100s from Tesla to X, citing a memo from Nvidia staff. The publication wrote that by allowing privately held X to jump the line for the in-demand chips ahead of Tesla, he has potentially pushed back the EV maker's acquisition of $500 million worth of GPUs by months.
Musk said during Tesla's first-quarter earnings call in April that the company would be increasing its number of active H100s from 35,000 to 85,000 by the end of this year in order to transform the automaker into "a leader in AI and robotics." He added that Tesla would spend $10 billion "in combined training and inference AI, the latter being primarily in car."
But the emails from Nvidia employees paint a different picture. They point out that Musk's comments during the earnings call and on X "conflicts with bookings."
"Elon prioritizing X H100 GPU cluster deployment at X versus Tesla by redirecting 12k of shipped H100 GPUs originally slated for Tesla to X instead," an Nvidia memo from December reads. "In exchange, original X orders of 12k H100 slated for Jan and June to be redirected to Tesla."
Musk responded to the story in an X post, claiming that Tesla could not accept the Nvidia chips due to a lack of capacity as the Texas factory is incomplete. He added that the South extension of Giga Texas will house 50,000 H100 chips for FSD (Full Self-Driving) training.
Tesla had no place to send the Nvidia chips to turn them on, so they would have just sat in a warehouse.
– Elon Musk (@elonmusk) June 4, 2024
The south extension of Giga Texas is almost complete. This will house 50k H100s for FSD training.
In a separate post, Musk wrote that of the roughly $10 billion of AI-related expenditures Tesla will make this year, about half is internal, primarily for the Tesla-designed AI inference computer and sensors present in all its cars, as well as the Dojo supercomputer it is building. He added that the company will purchase $3-4 billion of Nvidia hardware this year.
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.
– Elon Musk (@elonmusk) June 4, 2024
For building the AI training superclusters, NVidia hardware is about…
Musk gave Tesla an ultimatum in January: increase his ownership in the company to 25% or he will cut back on its development of AI and robotics. He explained that without owning a quarter of Tesla, which he says is enough to be influential but not so much that he can't be overturned, he would prefer to build products outside of the automaker.