Tesla is dissolving the team behind its Dojo supercomputer project, effectively ending its in-house chip development efforts aimed at powering autonomous driving systems, according to a Bloomberg report citing people familiar with the matter.
Peter Bannon, who led the Dojo initiative, is leaving the company, while remaining team members will be reassigned to other computing and data center projects within Tesla. The decision comes in the wake of the departure of around 20 Tesla employees who recently launched a new AI startup, DensityAI. The company, founded by former Dojo head Ganesh Venkataramanan along with ex-Tesla engineers Bill Chang and Ben Floering, is expected to unveil its plans soon and is building chips and software for AI data centers used in robotics, autonomous agents, and automotive systems.
The disbanding of the Dojo team marks a major shift in Tesla’s artificial intelligence strategy, especially after years of CEO Elon Musk promoting the project as a critical piece of Tesla’s pursuit of full self-driving technology. “Dojo is designed to process truly vast amounts of video data,” Musk said in previous public remarks. The project, launched in 2019 and officially unveiled in 2021, included the in-house development of the D1 chip, with a planned successor, the D2, to overcome performance bottlenecks.
Musk continued to promote Dojo as recently as Tesla’s second-quarter earnings call but has more recently turned attention to “Cortex,” a new AI supercluster being built at Tesla’s Austin headquarters. Industry speculation around Dojo’s future had been mounting since August 2024, when public discussion around the project largely ceased.
Tesla is now expected to increase its reliance on external chip and compute partners. The company recently signed a $16.5 billion deal with Samsung to manufacture its AI6 inference chips, a new architecture designed to scale from full self-driving applications to Tesla’s Optimus humanoid robots and high-performance AI training. Tesla is also expected to continue leveraging Nvidia and AMD technologies.
“Thinking about Dojo 3 and the AI6 inference chip, it seems like intuitively, we want to try to find convergence there,” Musk said during the July earnings call, suggesting possible overlap and redundancy between the projects.
The move to wind down Dojo comes at a time when Tesla is attempting to reinforce its AI credibility. Musk has repeatedly urged investors to view Tesla as a technology and robotics company, rather than just an automaker. The company’s limited robotaxi pilot in Austin earlier this year, using Model Y vehicles with safety drivers, faced multiple reports of erratic driving behavior.
Meanwhile, Tesla’s board has approved a $29 billion pay package aimed at keeping Musk focused on Tesla’s AI and robotics initiatives, amid concerns about his involvement with other ventures, including his separate AI company, xAI.