Email Tesla CEO Elon Musk has ordered the shutdown of the Dojo supercomputer team, according to a Bloomberg report. The team leader Peter Bannon departed from the team. The Dojo supercomputer was designed to process vast amounts of data and videos from Tesla’s EVs, to train its automated driving software. Around 20 Dojo team members also left for a new startup called DensityAI, while the remaining staff were reassigned to other projects. This comes weeks after Tesla’s second-quarter earnings call, when Musk painted a glowing picture of the Dojo project. Dojo was planned to be a custom-built supercomputer designed to train the neural networks for Full Self-Driving (FSD) and Tesla’s humanoid robot, Optimus. He called the next generation of Dojo “really spectacular,” and spoke of a new “AI factory” with “a lot of potential,” and laid out plans for Dojo 3. READ: Elon Musk’s Tesla enters India with $70,000 model; Set to compete with BMW, Mercedes-Benz (July 15, 2025) The Dojo project had already faced several challenges. Widely seen as an attempt by Tesla to reduce reliance on Nvidia by building its own custom chips and training hardware, its strategy was considered “risky” by experts. On the July call, Musk responded to a question about whether his AI startup xAI might use Dojo by saying “we expect Dojo 2 to be operating at scale within the next year. At about 100 Kh, 100 equivalents.” He then added, “And then AI Five, which is also really spectacular. I don’t use those words lightly. We hope to have the AI factory in production towards the end of next year. But that has a lot of potential.” Musk even discussed an “intuitive” convergence between Dojo 3 and chips used in Tesla’s cars and Optimus robots: “Thinking about Dojo 3 and the AI six in the first chip, it seems intuitively that we want to try to find convergence there. It’s basically the same chip that’s used when you say two of them in a car or Optimus and maybe a larger number on a five-twelve count on a card or something like that, if you want high-bandwidth communication between chips. That seems sort of intuitively the way to go.”