Tesla’s supercomputer project, Dojo, is a powerful and bespoke hardware platform designed by Tesla’s AI division for training computer vision and full self-driving networks. Dojo uses a system on a chip called the Dojo chip, which integrates 25 chips to create one unified system. Tesla is also using wafer scale processing to create large, single-chip supercomputers, allowing for maximum yield and efficiency.
Although Tesla continues to purchase Nvidia chips for their self-driving AI, the investment in Dojo is a long-term bet with potential rewards of hundreds of billions of dollars. The demand for supercomputers has risen significantly due to the importance of artificial intelligence, leading to an arms race among tech companies.