Tesla has ceased development of its Dojo supercomputer, originally conceived to enhance the company's Full Self-Driving (FSD) capabilities. The project, initiated in 2019, aimed to process vast amounts of video data from Tesla vehicles to train advanced AI models. Dojo's architecture involved custom-designed D1 chips, with plans for massive scaling using ExaPODs. Each D1 chip contained 354 computing cores and was manufactured by TSMC using a 7nm process.
The decision to discontinue Dojo follows the departure of key personnel, including team leader Peter Bannon, and a shift towards leveraging external partnerships for AI chip development. Tesla is now focusing on its AI5 and AI6 chips, with Samsung securing a $16.5 billion deal to supply AI chips. CEO Elon Musk stated that concentrating resources on the AI6 chip rendered continued Dojo development redundant. The company will reassign the Dojo team to other computing and data centre projects.
Tesla's future AI computing infrastructure will rely on external technology suppliers like Nvidia and AMD. Samsung's Texas plant is slated to produce Tesla's AI6 chip, with AI5 production expected in late 2026. This strategic pivot reflects a move from in-house hardware development to a collaborative approach in advancing Tesla's AI ambitions.
