Tesla to restart Dojo 3 chip development as AI5 design goes stable

In a significant boost to its artificial intelligence roadmap, Tesla has announced the restart of the development of Dojo 3, the company’s next-generation supercomputer designed for training advanced neural networks.

This decision comes as the design of the AI5 inference chip reaches a stable stage, providing much-needed time for engineers to focus on what Tesla’s leadership described as an existential priority for the future in autonomous driving and robotics.

In mid-2025, Tesla consolidated its silicon efforts around a single unified architecture. Maintaining separate paths for inference chips used in vehicles and robots, alongside a distinct training-focused Dojo lineage, proved inefficient.

Resources were redirected entirely toward perfecting the AI5 and forthcoming chips, while pausing the dedicated Dojo program. This decision reflected a recognition that diverging designs diluted talent and momentum, rendering earlier Dojo iterations less viable in the long term.

Now, with AI5 emerging as a powerhouse, Tesla finds itself with the bandwidth to restart work on this specialized training hardware. Dojo 3 represents an evolution, potentially building on clustered configurations of advanced chips like AI6 and beyond. This approach builds on Tesla’s strength in producing chips at scale, driven by plans to integrate them into millions of vehicles and humanoid robots.

At the heart of Dojo’s importance lies Tesla’s driving data advantage. The company’s growing fleet captures petabytes of real-world driving and environmental footage daily, providing the raw material for training self-driving models.

Custom training supercomputers like Dojo allow Tesla to process this data at exascale levels efficiently, accelerating breakthroughs in Full Self-Driving capabilities and Optimus robot intelligence.

This vertical integration, from edge inference in every car and bot to massive centralized training clusters, positions Tesla to outpace rivals reliant on third-party hardware. By controlling the entire stack, the EV maker aims for superior performance per dollar and per watt, turning compute into a core competitive moat.

The resumption of Dojo 3 underscores Tesla’s massive push toward AI. With fast iterations planned across future chip generations, the company is recruiting new talent to build what could become the world’s most prolific AI silicon.