Tesla is refocusing its artificial intelligence efforts by stepping back from the development of its Dojo supercomputer, CEO Elon Musk confirmed, shifting attention to the company’s internally designed AI5 and AI6 chips. The move marks a strategic pivot away from building a custom supercomputer specifically for AI training towards optimizing chips for product use, according to Musk.
The news was initially reported by Bloomberg News, which cited sources familiar with Tesla’s plans. Bloomberg also reported that Peter Bannon, who led Project Dojo, would be leaving the company, and that the Dojo team has seen around 20 members depart recently. Remaining staff working on the initiative are expected to be reassigned to other Tesla data center and compute projects.
Dojo was originally conceived as a high-performance platform to accelerate Tesla’s Autopilot and Full Self-Driving (FSD) system training. However, Tesla will now rely more on external technology partners such as Nvidia, AMD, and Samsung for AI training hardware, according to the Bloomberg report.
On social media platform X, Musk explained the rationale behind the decision, emphasizing the inefficiency of dividing resources between two distinct chip designs. “It doesn’t make sense for Tesla to divide its resources and scale two quite different AI chip designs,” Musk wrote. He noted that the AI5 and AI6 chips, which will be used in products like Tesla’s Optimus robot and Cybercab autonomous vehicle, will excel at AI inference tasks and perform well for training as well.
Musk further commented on the potential for future supercomputer clusters composed of AI5 and AI6 chips to simplify infrastructure and cut costs, calling it a possible evolution of the Dojo concept. “One could call that Dojo 3, I suppose,” he said, adding that the performance improvement from Tesla’s AI4 chip to AI5 was “far more than any chip version I’ve ever heard of by a lot.”