Explainer: Why Tesla Shut Down Its Dojo AI Supercomputer Project

What was Dojo, and why did it matter?
Dojo was Tesla’s custom-built AI supercomputer project, launched in 2021, designed to process massive video data to support its Full Self-Driving (FSD) technology. It combined Tesla’s in-house D1 chips with Nvidia GPUs and aimed to make Tesla less reliant on external compute providers.

Why is Tesla shutting it down now?
Tesla disbanded the Dojo team and reassigned its members, according to Bloomberg. The project’s end follows the departure of its top engineers, including project head Peter Bannon and about 20 others who formed their own AI startup, DensityAI.

What does this mean for Tesla’s AI efforts?
Although Tesla is ending its in-house supercomputing push, it’s not abandoning AI. The company is shifting toward using external partners like Nvidia, AMD, and Samsung to power its AI systems, including its next-gen AI6 chips.

What’s happening with the engineers who left?
The group that left Tesla is now behind DensityAI, a stealth startup developing chips, hardware, and software for AI data centers. The company was founded by former Tesla engineers Ganesh Venkataramanan, Bill Chang, and Ben Floering.

Was Dojo successful before it ended?
While Musk previously said Dojo would be key to Tesla’s FSD goals, progress appeared to stall by mid-2024. In 2023, Morgan Stanley predicted Dojo could add $500 billion to Tesla’s value, but internal momentum shifted toward a new project called Cortex.

What is Cortex?
Cortex is Tesla’s new AI training supercluster being built at its Austin headquarters. Musk introduced it in 2024, signaling a pivot away from Dojo while still chasing AI dominance.

What about Tesla’s robotaxi plans?
Tesla ran a limited robotaxi launch in Austin this June using human-monitored Model Y vehicles. The launch drew criticism due to several reports of problematic driving behavior.

How is Elon Musk responding to this shift?
Musk has pushed to reframe Tesla as an AI and robotics company and is currently being offered a $29 billion pay package to remain focused on Tesla, instead of spreading attention across ventures like xAI.

What’s next for Tesla’s AI chips?
Tesla recently signed a $16.5 billion deal with Samsung to manufacture its AI6 inference chips. These chips are intended to power Tesla’s Optimus robot, FSD systems, and AI data centers, potentially replacing the role Dojo was supposed to fill.

Tesla’s decision to shutter Dojo reflects a strategic shift toward working with tech giants over building in-house supercomputing solutions, while still chasing leadership in AI and robotics.