Tesla has officially confirmed the shutdown of its ambitious Dojo supercomputer project, with Elon Musk labeling the initiative an “evolutionary dead end.”
This decision marks a significant strategic pivot for the company, which is now redirecting its resources towards the more versatile and scalable AI5 and AI6 chips. Initially, Dojo was conceived as a revolutionary, in-house hardware solution to handle the immense data training required for Tesla’s Full Self-Driving (FSD) software and its Optimus humanoid robots. However, Musk explained that continuing to develop Dojo alongside the more efficient AI5 and AI6 chips was impractical and created a costly split in engineering efforts.
The new architecture, which relies on a unified approach for both in-vehicle inference and large-scale data center training, has proven to be more effective and simplifies the complex networking and cooling infrastructure that Dojo required. This change in direction has also led to key personnel changes, including the departure of Peter Bannon, who had been leading the Dojo project. By focusing on the AI5 and AI6 chips, which are being manufactured by partners like TSMC and Samsung, Tesla is moving away from its vision of a fully proprietary hardware solution in favor of a more pragmatic and widely supported ecosystem.