The end of Tesla ‘s own artificial intelligence supercomputer efforts — when it was developing its own chips — means Elon Musk’s company will become an even bigger customer for major chipmakers, according to Wells Fargo. Analyst Aaron Rakers highlighted Nvidia and Advanced Micro Devices as beneficiaries of Tesla’s decision to shut down its custom-built AI training supercomputer Dojo team, which was first reported by Bloomberg on Thursday. Pete Bannon, Tesla hardware design engineering vice president, is leaving the company after joining from Apple in 2016, CNBC confirmed. Remaining employees are being reassigned to Tesla, Bloomberg’s report said, citing people familiar with the matter. Dojo’s end is “an indication that the company is stepping back from its internal AI-optimized silicon efforts,” said Rakers. “This should be viewed as a positive for NVIDIA and AMD as this likely increases Tesla usage of general purpose GPUs for AI.” Rakers highlighted that Tesla has been notably expanding its graphics processing units, or GPU, infrastructure. Tesla has bought several thousands of Nvidia GPUs such as the H100, in an effort to scale its training compute power. “During the 2Q25 earnings call, Tesla disclosed that the company has deployed a ~50k NVIDIA H100 training cluster Cortex. The same week Elon Musk posted that Tesla was deploying an add’l 16k NVIDIA H200 GPUs at its Texas Gigafactory; running at ~67k NVIDIA H100 equivalents,” Rakers said. Musk said in early 2024 that while a Dojo supercomputer would cost $500 million to build, “Tesla will spend more than that on Nvidia hardware this year.” Nvidia GPUs were being used for Tesla’s other efforts beyond Dojo, such as training and inference of its Grok AI chatbot. TSLA 1Y mountain Tesla stock performance over the past year. The analyst acknowledged that the Bloomberg report is markedly different from Musk’s comments during Tesla’s second-quarter earnings call, during which he said that the Dojo 2 supercomputer is expected to operate at scale next year and that the Dojo 3 and the Ai6 inference chip would find convergence, meaning they would integrate in some way and be the same chip. Tesla’s custom-designed Ai6 AI chip was intended for in-vehicle inference for Tesla’s Full Self-Driving system. The Dojo supercomputer was designed to train AI models on data captured by Tesla vehicles, and has been part of Musk’s heavy emphasis to investors over the past year that Tesla is an AI and robotics company, not just an electric vehicle company. Tesla shares edged slightly lower before Friday’s market open. The stock is down about 20% year to date.
