Elon Musk admitted diverting Nvidia’s newest AI chips initially destined for Tesla to his newest startup xAI and social media firm X.
Confirming a report by CNBC on Tuesday, Musk defined the transfer by saying the chips, used for coaching neural networks, had been higher used elsewhere on the time.
“Tesla had no place to send the Nvidia chips to turn them on, so they would have just sat in a warehouse,” he posted.
This won’t appear so controversial for a producer of electrical automobiles, however it comes because the CEO has step by step pivoted away from the purpose of rising annual automobile gross sales greater than tenfold to the promised 20 million mark within the face of low-cost competitors out of China.
As an alternative, Musk has been methodically rebranding Tesla as an AI and robotics firm, pledging to make investments $10 billion this yr alone to beef up its AI coaching and inference compute. In April, he revealed plans to ramp up chip procurement to the equal of 85,000 Nvidia H100s by the tip of this yr only for AI coaching, i.e. not together with inference. He had round 35,000 on the finish of the primary quarter.
Tesla had no place to ship the Nvidia chips to show them on, so they might have simply sat in a warehouse.
The south extension of Giga Texas is nearly full. This may home 50k H100s for FSD coaching.
— Elon Musk (@elonmusk) June 4, 2024
The timing of the revelation may be damaging as Musk awaits the June 13 consequence of a tense vote over his report pay bundle, dubbed the “largest in human history”, whose earlier approval (in 2018) was dominated invalid by a Delaware courtroom due to governance failings. Tesla shareholders may worry they could get the brief finish of the stick, if Musk deems the chips are extra wanted at xAI, SpaceX or X.
Musk sparked fears he may breach his fiduciary duty to Tesla shareholders after threatening initially of this yr to additional transfer AI growth out of Tesla if traders didn’t grant him sufficient inventory to boost his stake from 12% to 25% (one other 9% held not directly within the type of choices granted as compensation is in dispute due the Delaware ruling).
These issues had been then exacerbated after his xAI startup started hiring staff, together with from Tesla, allegedly as a result of they had been seeking to depart the newly branded AI and robotics firm anyway.
Musk’s conflicting pursuits
On Tuesday, Musk went on to say an growth of the EV maker’s Texas web site to accommodate a brand new information heart is nearly full, after the CNBC report cited warnings of attainable additional delays to shipments of Nvidia’s benchmark AI chips to Tesla.
“This will house 50k H100s for FSD training,” the CEO stated, a quantity which accounts for the hole in remaining chips to be bought this yr. He then added: “I can’t overstate the difficulty of making 50k H100s train as a coherent system. No company on Earth has been able to achieve this yet.”
It’s unclear why Tesla ordered the chips within the first place if that they had no functioning information heart infrastructure that might use them on the time, and Musk didn’t elaborate.
One motive could be that Nvidia’s benchmark H100 coaching chips—so superior the U.S. banned their export to China—are in such demand that he couldn’t afford to take the danger of passing on a possible supply slot allotted by Jensen Huang’s firm.
Prior to now, nonetheless, he has complained about not getting as many Nvidia GPU clusters as he wanted, explaining that his deliberate Dojo supercomputer that makes use of customized Tesla-designed silicon was little greater than a hedge towards additional delays.
Whether or not diverting among the chips away from Tesla really harm the carmaker in any materials means can also be unclear. The corporate claims to have invested $1 billion to greater than double its compute capability within the first quarter over the earlier three months to end-December.
Late in March, Musk flagged in a put up that Tesla was not AI coaching compute constrained. This might counsel the interval by which it had infrastructure capability up and working however not sufficient chips to function them is now over.
It’s attainable that no injury was completed to the corporate, and given the dimensions of Musk’s financial curiosity it could appear logical he wouldn’t need to hurt his solely confirmed worthwhile endeavor.
Nevertheless, his admission of diverting Tesla’s AI chips to his different firms highlights the fraught nature of working a publicly traded firm alongside a number of privately-owned ones, particularly after they compete for a similar assets. Musk doesn’t see a battle of curiosity and isn’t shy about shifting assets as he sees match.
Throughout Musk’s pay bundle trial, the CEO confirmed he dispatched 50 so-called volunteer Tesla engineers to help him at Twitter. James Murdoch, a Tesla director on each the board’s audit and governance committees, was not in a position to present a lot as a ‘ballpark’ determine when requested in regards to the determination, although his testimony was offered just a few weeks later.
