Tesla Shuts Down Dojo AI Supercomputer, Marks Shift in Elon Musk’s AI Strategy

Tesla has officially shut down its long-hyped Dojo AI supercomputer and disbanded the team behind it as of August 2025, ending a six-year project that promised to revolutionize autonomous driving and AI development. Elon Musk previously hailed Dojo as a cornerstone of Tesla’s AI ambitions, even doubling down on the project in July 2024 ahead of Tesla’s robotaxi reveal in October.

The Dojo 2 supercluster, which was to leverage Tesla’s in-house D2 chips, was projected to reach scale by 2026. However, Musk reversed course, calling it “an evolutionary dead end” and pivoting toward the company’s new AI6 chips developed in partnership with Samsung.

Dojo was designed to train Tesla’s Full Self-Driving (FSD) neural networks, supporting advanced autonomous driving and humanoid robots. While FSD is currently active in hundreds of thousands of Tesla vehicles, requiring human oversight, the supercomputer was meant to accelerate Tesla’s robotaxi service and future AI innovations.

Elon Musk standing beside a Tesla Dojo supercomputer unit during a company event.

Despite its promise, Tesla shifted focus last year to Cortex, a separate AI supercluster in Austin, which now handles much of the company’s AI training workload. Tesla’s Q4 2024 shareholder update detailed Cortex progress but omitted updates on Dojo. Whether Dojo’s closure impacts Cortex remains unclear.

The shutdown saw Tesla lose several key AI engineers, including Dojo lead Peter Bannon. Some former team members have launched new AI ventures, like DensityAI, further reducing Tesla’s internal AI talent pool. Analysts suggest that talent departures can severely impact specialized tech projects.

The move coincides with Tesla’s $16.5 billion deal with Samsung for AI6 chips, which are designed to power FSD, Optimus humanoid robots, and large-scale AI training. Musk indicated that Dojo’s hardware strategy converged with AI6, making the original Dojo design redundant.

Tesla’s Dojo project centered on the company’s proprietary D1 chips, designed to optimize AI workloads and reduce reliance on expensive Nvidia GPUs. Although Tesla planned ambitious targets—like achieving one of the world’s top five supercomputers and reaching 100 exaflops in 2024—it appears these goals were never realized.

The Dojo closure highlights Tesla’s pivot from high-risk in-house hardware development to a more collaborative approach relying on external chip partners. While some critics view this as a failure, others see it as a strategic move to streamline AI efforts while maintaining ambitious goals for autonomous driving and robotics.

Tesla confirmed it will continue investing $500 million in a Buffalo-based supercomputer, but it will no longer carry the Dojo name. The shift marks a clear evolution in Musk’s AI strategy, emphasizing scalability, efficiency, and partnerships over wholly internal hardware projects.