In an unexpected move, Elon Musk has ordered Nvidia to send thousands of GPUs, originally reserved for Tesla, to their X and xAI projects. According to internal emails from Nvidia reviewed by CNBC, this decision has sparked controversy within the technology company, with employees noting that Musk’s comments on GPU investment during the first-quarter earnings call “conflict with reservations.”
Nvidia has declined to comment on the matter, while Musk did not respond to requests for comments. However, on X (formerly Twitter), Musk explained that “Tesla had no place to send Nvidia chips to activate them, so they would have remained in a warehouse. The south extension of Giga Texas is nearly complete. This will house 50,000 H100s for FSD training.”
During the earnings call in April, Musk stated that Tesla was currently operating 35,000 H100 GPUs and expected to increase that number to 85,000 by year-end. Days later, on X, he mentioned that Tesla would spend $10 billion this year on “combined AI training and inference.”
Nevertheless, Nvidia memos indicated that these statements did not align with what they had sold to Tesla. Instead, the company redirected over $500 million worth of GPU reservations to X.
“Elon prioritizes the deployment of the H100 GPU cluster in X over Tesla, redirecting 12,000 H100 GPUs originally slated for Tesla to X,” a December Nvidia memo stated.
“In exchange, X’s original orders of 12,000 H100 scheduled for January and June will be redirected to Tesla.”
Another email, post-earnings call, noted that the $10 billion figure “conflicts with reservations and fiscal year 2025 forecasts,” and referenced drastic job cuts at the electric vehicle company, which could further delay the H100-based supercomputer at the Texas Gigafactory.
Musk has asserted that autonomous driving is key to Tesla’s future, but to achieve it, they will need massive GPU capabilities. These decisions could slow down that transition, though they may help with short-term cash flow as the company’s finances continue to struggle.
The renowned CEO is also trying to persuade shareholders to reinstate the largest executive compensation package in corporate history, a $56 billion compensation agreement that was nullified by a Delaware judge. If reinstated, Musk would have a 20.5% stake in Tesla. If not, he mentioned he would “prefer to build products outside of Tesla.”
Musk has previously expressed that he felt “uncomfortable growing Tesla to be a leader in AI and robotics without ~25% voting control. Enough to be influential, but not so much as to be irrevocable.”
Several AI researchers have left Tesla to join xAI.
In a post on X following CNBC’s report, Musk clarified: “Of the approximately $10 billion in AI-related expenses I said Tesla would make this year, about half is internal, mainly Tesla-designed AI inference computer and sensors in all our cars, in addition to Dojo. To build the AI training superclusters, Nvidia hardware represents about 2/3 of the cost. My current best estimate for Tesla’s Nvidia purchases this year is $3 billion to $4 billion.”
This week, Musk also stated that xAI will deploy a 300,000 GPU B200 supercomputer next summer and launch a data center with 100,000 H100 GPUs soon. None of these projects have been independently verified.
It is believed that xAI currently rents around 16,000 H100 GPUs from Nvidia at Oracle Cloud, in addition to utilizing Amazon Web Services and excess capacity in X data centers.