Wells Fargo names two beneficiaries of Tesla's refusal to develop its own AI chips

Tesla's refusal to create its own supercomputer Dojo and specialized AI chips for its work will lead to an increase in the company's purchases from leading manufacturers of graphics processors - Nvidia and AMD, predicted Wells Fargo. The shutdown of the Dojo project was reported on August 8 by Bloomberg, citing sources inside the company, and company CEO Ilon Musk actually confirmed it.

Tesla's refusal to develop the Dojo supercomputer and AI chips for it will benefit AMD and Nvidia, as Tesla will become an even more active buyer of them. Wells Fargo analyst Aaron Rakers wrote this in an Aug. 8 note, he quoted by CNBC.

"This should be seen as a positive for Nvidia and AMD, as it will likely lead to Tesla expanding Tesla's use of their general-purpose GPUs for artificial intelligence," he said.

The Wells Fargo analyst pointed out that Tesla had been steadily expanding its GPU infrastructure even before the Dojo project was shut down, including buying several tens of thousands of units from Nvidia, such as the H100 model. 

"During the Q2 earnings call, Tesla disclosed that it has deployed approximately 50k Nvidia H100 chips in the Cortex training cluster (located at Tesla's headquarters in Austin, Texas. - Oninvest). "In the same week, Ilon Musk announced that Tesla was deploying an additional 16k Nvidia H200 chips at its 'gigafactory' in Texas: translating to about 67k H100 chips," Rakers wrote.

That said, Tesla's decision is markedly different from what Musk said during a call after the report, a Wells Fargo analyst noted. The top executive claimed that the company expects to fully launch the Dojo 2 supercomputer next year. He also said that Dojo 3 will be integrated with Tesla's AI chip AI6, CNBC explained.

Tesla decided to wind down the team involved in building its own Dojo supercomputer and the specialized chips to run it, reported Bloomberg on August 8. According to the agency's sources, Vice President of Hardware Engineering Pete Bannon is leaving the company, and the remaining engineers will be reassigned to other projects.

Following Bloomberg's publication, Tesla CEO Ilon Musk wrote on social network X that it was difficult for the company to spend resources on developing multiple chips and would focus on releasing in-car AI5 and AI6 artificial intelligence processors.

"It doesn't make sense for Tesla to share resources and [parallel] scale two rather different AI chip architectures. Tesla's AI5, AI6 and beyond chips will excel for inferencing (AI's ability to produce new information after learning. - Oninvest)  and at least good enough for learning. All efforts are focused on this direction," Musk stated.

Bloomberg sources named Samsung, with which Tesla in July contracted a $16.5 billion deal, as the beneficiary of the Dojo project closure. The Korean manufacturer will supply semiconductors for Tesla's artificial intelligence systems until 2033. In addition, the agreement includes the construction of a Tesla AI6 chip plant in Texas, which will allow the company to diversify its supply chain and reduce its dependence on the world's leading manufacturer, TSMC, Bloomberg noted. 

Dojo is a Tesla-developed supercomputer used to train machine learning models underlying the Autopilot and Full Self-Driving systems, as well as the humanoid robot Optimus, Bloomberg wrote. And the AI6 chip was developed for AI work directly in cars as part of the Full Self-Driving system, the agency explained.

The supercomputer processes video data from the cars and quickly analyzes it to improve the company's algorithms. Analysts have noted that Dojo could become an important competitive advantage: in 2023, Morgan Stanley predicted that the supercomputer could potentially add as much as $500 billion to Tesla's market capitalization.

This article was AI-translated and verified by a human editor