Tesla Dojo AI Supercomputer – Rise, Shutdown and Revival

Tesla Dojo: The Rise and Fall of an in House AI bet:

Tesla Dojo was built as a custom AI supercomputer to train neural networks for Full Self Driving. The system was designed to process massive volumes of real world driving video collected from Tesla vehicles.

First deployed in 2023, Dojo became a core part of Tesla’s in house AI strategy. Engineers aimed to reduce dependence on external GPUs while accelerating FSD development.

Why Tesla Shut Dojo Down:

Tesla later explored a successor platform based on its own D2 chips. Elon Musk ultimately called that direction an evolutionary dead end.

In August 2025, Tesla disbanded the Dojo team and scrapped the project. The move triggered key departures, including project lead Peter Bannon, while roughly 20 engineers went on to launch a new AI infrastructure startup.

Key Reasons Behind the Shutdown:

• Limited long-term scalability of D2 based designs
• Faster progress available through external GPU platforms
• Rising costs of fully custom AI infrastructure

Shift to External GPUs:

After ending Dojo, Tesla leaned more heavily on hybrid AI training. That included significant reliance on Nvidia hardware rather than replacing it outright.

Tesla’s evolving AI strategy also mirrors changes in how it monetizes Full Self Driving. As AI training becomes more compute-intensive, Tesla has shifted away from a one time FSD purchase toward a subscription based FSD model, aligning recurring revenue with the long term cost of AI development and infrastructure.

Dojo Returns with AI5:

Tesla later confirmed the project is back. The revived effort, often referred to as Dojo3, will center on the in house AI5 chip.