Tesla Shuts Down Project Dojo, Accelerates AI Goals with Industry Giants
Tesla has officially shut down Project Dojo, marking the end of its in-house chip development for autonomous driving. Instead, the company will rely on Nvidia and AMD for computing power, while Samsung will handle chip manufacturing. CEO Elon Musk explained that maintaining two separate AI chip designs would drain resources and slow progress. Dojo, built for training Tesla’s self-driving system, was never intended for consumer use.
Dojo was Tesla’s custom supercomputer project, designed to train neural networks powering Full Self-Driving (FSD) and future Robotaxi services. Neural networks simulate human decision-making, enabling cars to detect and respond to road conditions. Although the current FSD system still requires a driver’s attention, it benefits from massive amounts of visual data collected by Tesla’s global fleet. Musk had described Dojo as the foundation of Tesla’s AI ambitions because it could “process truly vast amounts of video data.” However, the strategic shift means that reliance on partners is now seen as the faster route to scaling AI capabilities.
Why Supercomputers Matter for Tesla’s Vision
Tesla’s approach to autonomy focuses solely on cameras, rejecting additional sensors like lidar or radar. For FSD to work effectively, its neural networks must process large volumes of driving data and make instant decisions. Achieving this requires powerful supercomputers capable of running millions of simulations. In these systems, CPUs manage basic tasks while GPUs handle complex parallel processing essential for AI training.
Supercomputers allow Tesla to replicate human depth and motion perception digitally. This ability is crucial for recognising objects, predicting movements, and ensuring safe navigation. While Dojo aimed to boost performance and reduce reliance on suppliers, Tesla had still been using Nvidia hardware to power the system. By transitioning fully to external chipmakers, Tesla hopes to increase bandwidth, lower latency, and accelerate development.
The Road Ahead After Dojo
In August 2024, Musk introduced Cortex, a new AI training supercluster at Tesla’s Austin headquarters. This facility is intended to address real-world AI challenges, shifting focus away from Dojo’s original mission. Reports indicate Tesla will expand its partnerships, with a $16.5 billion deal signed in July 2025 for Samsung to produce AI6 inference processors. These chips will support AI training, FSD, and the Optimus humanoid robot.
With the change, Dojo’s lead engineer, Peter Bannon, will leave Tesla, and remaining team members will move to other compute and data centre projects. While Dojo’s closure ends one chapter, Tesla’s pursuit of full self-driving continues only now, the journey will be powered by some of the most advanced chips on the market from industry leaders.
