Elon Musk Says Tesla Is Bringing Back Its Dojo3 AI Supercomputer, Now Space-Based

Elon Musk said Tesla is restarting development of its Dojo3 supercomputer to support AI training, after previously pausing the project to focus on in-vehicle AI chips.

By Maria Konash Published: Updated:

Tesla will restart work on Dojo3, the third generation of its in-house AI supercomputer, after shelving the project last year,  Elon Musk said in a post on X. Musk said the company is returning to Dojo now that the design of its next-generation AI5 chip is “in good shape.”

The Dojo project is intended to process large volumes of video and sensor data collected from Tesla vehicles to train the neural networks behind its Full Self-Driving software. Last year, Tesla disbanded the Dojo team as it concentrated resources on AI chips designed to run inference directly inside vehicles, rather than on large-scale training systems.

Those in-vehicle chips, including the upcoming AI6, are optimized for real-time decision-making rather than training. Tesla has said the AI6 chip will be manufactured by Samsung at its Texas facility under a $16 billion agreement.

Musk added a new and speculative dimension to the revived effort, describing Dojo3 as “space-based AI compute.” The concept, which envisions orbital data centers powered by solar energy and cooled by space temperatures, has attracted interest in tech circles but remains unproven and faces significant technical and economic challenges.

The renewed focus on AI infrastructure also comes as Tesla faces heightened regulatory scrutiny of its driver-assistance technology, with U.S. auto safety regulators examining whether Full Self-Driving is linked to thousands of potential traffic violations.

AI & Machine Learning, Cloud & Infrastructure, Mobility & Transportation Tech, News