Elon Musk’s vision for Tesla goes beyond simply being an automaker. He wants Tesla to be an AI company, with the ability to develop fully self-driving cars. To achieve this, Tesla has been working on a custom-built supercomputer called Dojo. Dojo is designed to train Tesla’s Full Self-Driving (FSD) neural networks, which are responsible for autonomous driving capabilities.
The journey towards developing Dojo started in 2019 when Elon Musk first mentioned the supercomputer during Tesla’s Autonomy Day. At the event, Musk discussed the AI team’s work on Autopilot and Full Self-Driving, showcasing Tesla’s custom-built chips designed for neural networks and self-driving cars. Musk hinted at Dojo being a supercomputer for training AI.
Throughout 2020, Musk continued to tease Dojo’s capabilities, emphasizing its potential to process vast amounts of video training data. He described Dojo as a “beast” and projected its launch date to be around August 2021.
Finally, in August 2021, Tesla officially announced Dojo at its first AI Day. The event aimed to attract engineers to Tesla’s AI team and showcased the D1 chip, which, alongside Nvidia’s GPU, powers the Dojo supercomputer. Tesla revealed plans to house 3,000 D1 chips in its AI cluster.
By September 2022, Tesla had made progress with Dojo, installing the first Dojo cabinet for load testing. The company aimed to complete a full Exapod cluster by Q1 2023, with a total of seven Exapods planned for Palo Alto.
In April 2023, Musk highlighted the potential of Dojo, stating that it had the potential for an order of magnitude improvement in the cost of training. He even considered offering Dojo as a sellable service to other companies.
Tesla’s progress with Dojo continued in 2024, with plans for scaling. Musk acknowledged the high-risk nature of the project but expressed confidence in its success. Tesla announced a $500 million investment to build a Dojo supercomputer in Buffalo. Musk downplayed the investment, emphasizing that being competitive in AI required several billion dollars per year.
In April 2024, it was revealed that the next-generation training tile for Dojo, the D2, was already in production. Tesla also planned to construct a super-dense, water-cooled supercomputer cluster at its Giga Texas factory extension.
However, Tesla faced challenges in the supply of Nvidia hardware, which affected their AI training capabilities. Musk highlighted the demand for Nvidia GPUs and stated that Tesla would focus more on Dojo to ensure sufficient training capability.
Despite the obstacles, Tesla remained committed to its AI ambitions. The company projected significant growth in AI training capacity, with Dojo 1 expected to have approximately 8,000 H100-equivalent of training online by the end of 2024.
In conclusion, Tesla’s Dojo supercomputer represents a crucial component in the company’s pursuit of fully self-driving cars. While the journey has been challenging, Tesla has made significant progress in developing and scaling Dojo, demonstrating its commitment to pushing the boundaries of AI technology in the automotive industry.
