Elon Musk admitted to transferring Nvidia’s latest artificial intelligence chips originally destined for Tesla to his latest startup xAI and social media company X.
Musk confirmed the CNBC report on Tuesday and explained the move, saying the chips used to train neural networks were better used elsewhere at the time.
“Tesla has no place to send Nvidia chips to power them up, so they sit in a warehouse,” he posted.
That may not seem all that controversial for the electric car maker, but in the face of low-cost competition, chief executives are gradually abandoning their goal of growing annual vehicle sales more than tenfold to the promised 20 million mark.
Instead, Musk has been methodically reinventing Tesla as an artificial intelligence and robotics company, pledging to invest $10 billion this year alone to bolster its artificial intelligence training and inference computing. In April this year, he revealed plans to increase chip procurement to the equivalent of 85,000 Nvidia H100 chips by the end of this year, which will be used only for artificial intelligence training, that is, not including inference. His score at the end of the first quarter was around 35,000.
Tesla has no place to send Nvidia chips to power them, so they sit in a warehouse.
The southern extension of Giga Texas is nearly complete. It will accommodate 50,000 H100 units for FSD training.
— Elon Musk (@elonmusk) June 4, 2024
The timing of the revelation could also be damaging, as Musk awaits the outcome of a tense June 13 vote on his record-breaking pay package, which has been described as “the largest in human history.” Early approval of the program (in 2018) was ruled invalid by a Delaware court because of governance failures. Tesla shareholders may be concerned that they could be at a disadvantage if Musk decides that xAI, SpaceX or X need the chips more.
Musk threatened earlier this year that if investors didn’t grant him enough shares to increase his stake from 12% to 25% (another 9 are held indirectly in the form of options due to the controversial Delaware ruling percentage).
Those concerns intensified after his xAI startup began hiring employees, including from Tesla, allegedly because they wanted to leave the newly branded artificial intelligence and robotics company anyway.
Musk’s conflicts of interest
On Tuesday, after a CNBC report cited Nvidia warning that shipments of benchmark artificial intelligence chips to Tesla could be further delayed, Musk went on to say that the expansion of the electric car maker’s Texas factory is almost complete to accommodate A new data center.
“This will accommodate 50,000 H100s for FSD training,” the CEO said, a number that accounts for the shortfall in remaining wafers to be purchased this year. He went on to add: “How difficult it is to train 50,000 H100s into a coherent system. No company on the planet has yet been able to achieve this.
It’s unclear why Tesla ordered the chips in the first place if it didn’t have a functioning data center infrastructure that could use them, and Musk didn’t elaborate.
One reason may be that Nvidia’s baseline H100 training chips – so advanced that the U.S. banned their export to China – are in such high demand that it can’t afford to risk giving up the potential delivery slots that Jensen Huang’s company has allocated.
However, he has complained in the past about not getting the Nvidia GPU clusters he needed, explaining that his planned Dojo supercomputer would use Tesla’s custom-designed chips simply to prevent further delays.
It’s unclear whether moving some chips from Tesla will actually cause any material harm to the automaker. The company said it had invested $1 billion in the first quarter to more than double its computing power in the three months to the end of December.
At the end of March, Musk pointed out in an article that Tesla is no longer limited by artificial intelligence training calculations. This may indicate that the period when the infrastructure capabilities were up and running, but there were not enough wafers to run them, is now over.
It’s possible that no damage was done to the company, and given the scale of Musk’s financial stakes, it seems logical that he wouldn’t want to jeopardize his only proven profitable endeavor.
However, his admission that moving Tesla’s AI chips to his other companies highlights the fraught nature of running a public company with several private companies, especially when they compete for the same resources. Musk doesn’t see a conflict of interest and isn’t shy about diverting resources as he sees fit.
During the trial of Musk’s compensation package, the CEO confirmed that he sent 50 so-called volunteer Tesla engineers to assist him on Twitter. Tesla director James Murdoch, who sits on the Audit Committee and Governance Committee of Tesla’s board of directors, couldn’t even provide a “ballpark” figure when asked about the decision, although his testimony was limited to It wasn’t available until a few weeks later.
