Tesla has shut down its Dojo supercomputer project and disbanded the team behind it, ending a six-year effort to build custom AI hardware for self-driving and robotics.
CEO Elon Musk said in August 2025 that Dojo 2, Tesla’s planned upgrade built on in-house D2 chips, was “an evolutionary dead end.” The move came weeks after Tesla signed a $16.5B deal with Samsung to supply next-generation AI6 chips, designed to power Full Self-Driving (FSD), Optimus humanoid robots, and high-performance AI training.
Dojo had been hyped as key to Tesla’s AI ambitions, with Musk once suggesting it could rival Nvidia and unlock new revenue streams. But progress lagged, and by late 2024 Tesla shifted focus to “Cortex,” a massive GPU-powered training cluster in Texas.
The shutdown led to the departure of Dojo lead Peter Bannon and 20 staffers who formed a startup called DensityAI. Analysts say the collapse underscores Tesla’s reliance on partners like Nvidia and Samsung, despite years of trying to develop its own chips.
Related: Federal regulators grant Zoox exemption for its custom-designed robotaxis
While some view Dojo’s demise as a failure, others see it as a strategic pivot away from risky in-house hardware to proven external solutions, a shift that could reshape Tesla’s AI strategy going forward.
