Tesla has officially restarted work on its Dojo3 supercomputer project following significant progress on its in-house AI5 chip design, Elon Musk announced on the X platform. The decision marks a reversal from the company’s 2025 decision to disband the Dojo team and shift focus to vehicle-based AI chips.
The Restart and AI5 Milestone
Musk stated that Tesla would resume the Dojo3 project “now that the AI5 chip design is in good shape.” The company is actively hiring engineers to work on what Musk describes as “the highest volume chips in the world.” According to Musk, the AI5 chip is nearly ready for deployment and represents Tesla’s most competitive offering yet, delivering performance comparable to Nvidia’s Hopper architecture on a single chip and Blackwell-class performance when two chips work together while consuming significantly less power.
From Abandoned to All In-House
Dojo3 will mark a significant departure from previous iterations. Unlike Dojo2, which relied on a mixture of Tesla’s in-house chips and Nvidia GPUs, Dojo3 will feature purely in-house hardware with no external dependencies.
The supercomputer’s primary purpose is to enhance the training of Tesla’s Full Self-Driving neural networks by processing video footage and data collected from Tesla vehicles. The AI chips powering Tesla vehicles are optimized for inference rather than training, making a dedicated training supercomputer essential for the company’s machine learning advancement.
A Bold Nine-Month Roadmap
Tesla’s restart of Dojo3 coincides with a new nine-month chip design cadence, where the company plans to release new chip architectures every nine months. AI5 is nearly ready for deployment, AI6 will target data centers and the Optimus robot platform, and AI7—integrated into Dojo3—will support what Musk describes as “space-based AI compute.”
The ambitious timeline aims to accelerate Tesla’s competitive advantage against established AI chip manufacturers, though success remains uncertain given the company’s previous struggles with earlier Dojo iterations.