Elon Musk moves thousands of Nvidia's AI accelerators from Tesla to X and xAI

Nvidia GPUs ordered by Tesla for AI training are now to be used for AI at X. This infuriates Tesla investors, but Musk has a simple explanation.

Save to Pocket listen Print view
Fahrer eines Tesla ohne Hände am Steuer auf der Straße

(Image: Beach Media/Shutterstock.com)

4 min. read
By
  • Frank Schräer
This article was originally published in German and has been automatically translated.

Tesla had ordered thousands of AI accelerators from Nvidia to use for training artificial intelligence in robots and for autonomous driving. However, according to internal emails at Nvidia, these will not arrive at the car manufacturer. Elon Musk has therefore redirected the orders to the short message service X, formerly Twitter, and the AI developers at xAI. Tesla investors are not ecstatic about this measure, as it would delay AI development - after all, AI chips have long delivery times due to high demand. However, according to Musk, the Nvidia GPUs could not have been used initially because the corresponding data center has not yet been completed.

Tesla originally wanted to buy 35,000 to 85,000 of Nvidia's H100 GPUs for AI training over the course of this year. At the end of April, Elon Musk explained at X that Tesla will invest around 10 billion US dollars in 2024 to improve the software for autonomous driving and robotaxis. However, internal emails from Nvidia employees suggest that Musk was exaggerating here. Many of the AI accelerators ordered would not be delivered to Tesla, but to X and primarily its AI subsidiary xAI, writes CNBC, which has the emails.

"Elon is prioritizing X over Tesla for the provision of the H100 GPU cluster by redirecting 12,000 of the H100 GPUs originally intended for Tesla to X instead", Nvidia says. "In turn, the original X orders of 12,000 H100s planned for January and June should be redirected to Tesla." In response, Nvidia employees commented that Musk's previous statements "contradict the bookings".

This is further fueling Musk's conflict with Tesla shareholders. They are not happy about Tesla's latest development anyway and criticize Elon Musk's corporate strategy and his involvement in his other companies such as X, SpaceX, Neuralink and the Boring Company's tunnel drillers. Tesla is suffering from declining sales figures due to an older vehicle range, increasing competition and lower subsidies from many countries for the purchase of electric cars. Tesla's share price has fallen by around 29 percent over the course of this year.

The fact that Tesla is now foregoing the AI accelerators it ordered and will only receive them several months later is another point of criticism from investors. After all, Nvidia is barely keeping up with demand with the production of its GPUs used for AI, meaning that the time between ordering and delivery is getting longer. This is one of the reasons why Nvidia is matching the profits of Google's parent company Alphabet.

For Tesla, Elon Musk had promised a supercomputer called "Dojo" in the US state of New York for half a billion dollars as well as a water-cooled supercomputer cluster in Texas. This will be used to develop technologies such as recognition processes and language models for robots and autonomous vehicles. Tesla's Gigafactory in Texas is currently getting a southern extension for a data center.

However, it is not yet ready for the delivery of AI accelerators, explains Elon Musk at X in a comment on the CNBC report. "Tesla had no place to send the Nvidia chips to turn them on, so they would just sit in storage", Musk writes. "The south expansion of Giga Texas is almost complete. It will house 50,000 H100s for FSD training." This refers to AI training for autonomous driving (Full Self Driving).

Shortly afterwards, Musk also specified Tesla's investment of 10 billion dollars in AI for this year. This includes not only the purchase from Nvidia, but also internal investments such as the AI computer and sensor technology developed by Tesla itself and in Dojo, Musk explained at X. Only two-thirds of the costs for the AI supercluster would involve Nvidia hardware. Musk's estimate is that Tesla will purchase three to four billion dollars from Nvidia in 2024.

(fds)