(Credit: Robert Wray via Getty Images)
Just as the success of its self-driving AI looks increasingly integral to Tesla's future, the company is dissolving the team behind its Dojo supercomputer—an in-house chip development and supercomputer project that was meant to process vast amounts of the company's vehicle video data to train its AI.
The project’s lead, Pete Bannon, who joined from Apple in 2016, will leave the company. According to Bloomberg, which first reported the news, Tesla will now lean on technology from third-party chipmakers such as Nvidia and Advanced Micro Devices (AMD) for its compute hardware, and Samsung Electronics for chip manufacturing. Employees from the Dojo team are set to be reassigned to other data center and compute projects within the company.
CEO Elon Musk had certainly talked a big game about the project's potential. In a 2024 earnings call, he said he saw "a path to being competitive with Nvidia with Dojo," though he also called it a "long shot" in another call with investors. (Nvidia is now a multi-trillion-dollar company.)
In a post on X earlier this week, Musk explained that: "It doesn't make sense for Tesla to divide its resources and scale two quite different AI chip designs.” Musk added that “all effort” would be focused on its “AI5, AI6 and subsequent chips.”
The move follows a wave of executive departures from the project in recent months. Bloomberg reported in August that roughly 20 former Dojo team members, including senior-level staff, joined AI startup DensityAI, founded by Ganesh Venkataramanan, the former head of the Dojo team. The new firm will work on chips, hardware, and software for AI data centers used in robotics, AI agents, and automotive applications.
Tesla has already signed deals to outsource parts of its AI work. In July, it reached an agreement worth up to $16.5 billion with Samsung to manufacture its next-generation AI6 chip, meant to power everything from its driver-assistance system, its Full Self-Driving (Supervised) feature, to the Optimus humanoid robot, as well as AI training in its data centers.