Elon Musk does not want Tesla to be just an automaker. He wants Tesla to be an AI company, one that has figured out how to make cars drive themselves. Crucial to that mission was Dojo, a custom-built supercomputer designed by Tesla to train its Full Self-Driving neural networks. Full Self-Driving is not actually fully self-driving; it can perform some automated driving tasks but still requires an attentive human behind the wheel. Tesla believes that with more data, more compute power, and more training, it can cross the threshold from almost self-driving to full self-driving. That is where Dojo was supposed to come in.
Musk teased Dojo for years and ramped up discussions about the supercomputer throughout 2024. But Dojo is now out, and another supercomputer called Cortex has entered the picture.
The timeline of Dojo began in 2019. At Tesla’s Autonomy Day on April 22, the company had its AI team onstage to talk about Autopilot and Full Self-Driving. The company shared information about its custom-built chips designed for neural networks and self-driving cars. During the event, Musk teased Dojo, revealing it was a supercomputer for training AI. He also noted that all Tesla cars being produced at the time would have all the hardware necessary for full self-driving and would only need a software update.
In 2020, Musk began the Dojo roadshow. On February 2, he stated Tesla would soon have over a million connected vehicles worldwide with the sensors and compute needed for full self-driving and touted Dojo’s capabilities. On August 14, he reiterated the plan to develop Dojo to process vast amounts of video data, calling it a beast and saying the first version was about a year away. On December 31, he said Dojo was not needed but would make self-driving better, stating Autopilot ultimately needs to be more than ten times safer than human drivers.
Tesla made Dojo official in 2021. On August 19, the automaker officially announced Dojo at its first AI Day, an event meant to attract engineers. Tesla also introduced its D1 chip, which it said would power Dojo alongside Nvidia GPUs. The AI cluster was to house 3,000 D1 chips. On October 12, Tesla released a Dojo Technology whitepaper outlining a technical standard for a new type of binary floating-point arithmetic used in deep learning.
In 2022, Tesla revealed Dojo progress. On August 12, Musk said Tesla would phase in Dojo and would not need to buy as many incremental GPUs the next year. On September 30, at Tesla’s second AI Day, the company revealed it had installed the first Dojo cabinet and was building one tile per day. The company set a target for a full Exapod cluster to be completed by Q1 2023 and planned to build seven Exapods in Palo Alto.
2023 was the year of the long-shot bet. On April 19, Musk told investors during Tesla’s first-quarter earnings that Dojo had the potential for an order of magnitude improvement in training cost and could become a sellable service. He called it a long-shot bet but one worth making. On June 21, the Tesla AI account posted that the company’s neural networks were in customer vehicles, with a timeline placing the start of Dojo production in July. Musk said that same day that Dojo was already online. The company projected its compute would be top five in the world by February 2024 and reach 100 exaflops by October 2024. On July 19, Tesla noted in its second-quarter earnings report it had started production of Dojo, and Musk said Tesla planned to spend over $1 billion on it through 2024. On September 6, Musk posted that Tesla was limited by AI training compute but that Nvidia and Dojo would fix that.
In 2024, plans to scale were announced. On January 24, during Tesla’s fourth-quarter earnings call, Musk acknowledged Dojo was high-risk, high-reward. He said Tesla was pursuing a dual path of Nvidia and Dojo, that Dojo was working and doing training jobs, and that Tesla was scaling it up with plans for Dojo 1.5, 2, and 3. On January 26, Tesla announced plans to spend $500 million to build a Dojo supercomputer in Buffalo. Musk later downplayed the investment, noting it was equivalent to a 10k H100 system from Nvidia and that Tesla would spend more on Nvidia hardware that year. On April 30, it was reported that Dojo’s next-generation training tile, the D2, was already in production. On May 20, Musk noted that the rear portion of the Giga Texas factory extension would include a super dense, water-cooled supercomputer cluster. On June 4, a CNBC report revealed Musk had diverted thousands of Nvidia chips reserved for Tesla to X and xAI. Musk later stated Tesla did not have a location for the chips due to construction and that the Giga Texas extension would house 50k H100s for FSD training. He also estimated Tesla would spend $3B to $4B on Nvidia purchases that year. On July 1, Musk revealed that current Tesla vehicles may not have the right hardware for the company’s next-gen AI model.
Nvidia supply challenges emerged. On July 23, during Tesla’s second-quarter earnings call, Musk said demand for Nvidia hardware was so high it was often difficult to get GPUs, requiring more effort on Dojo to ensure training capability. A graph predicted Tesla AI training capacity would ramp to roughly 90,000 H100 equivalent GPUs by the end of 2024. Later that day, Musk posted that Dojo 1 would have roughly 8k H100-equivalent training online by year’s end and shared photos of the supercomputer.
The shift from Dojo to Cortex began. On July 30, Musk said AI5 was about 18 months away from high-volume production. On August 3, Musk posted that he did a walkthrough of the Tesla supercompute cluster at Giga Texas, which he called Cortex. He noted it would be made of roughly 100,000 H100/H200 Nvidia GPUs for training FSD and Optimus. On August 26, Musk posted a video of Cortex, referring to it as the giant new AI training supercluster being built at Tesla HQ.
In 2025, the Dojo shutdown occurred. On January 29, Tesla’s Q4 and full-year 2024 earnings call included no mention of Dojo. Cortex, however, was noted in the shareholder deck as being completed and made up of roughly 50,000 H100 Nvidia GPUs. The CFO stated Cortex helped enable FSD V13 and that accumulated AI-related capital expenditures were approximately $5 billion. On July 23, during Tesla’s Q2 2025 earnings call, Musk said Dojo 2 was expected to be operating at scale in 2026 and hinted at possible redundancies, suggesting a convergence with the AI6 inference chip. On July 28, Tesla signed a $16.5 billion deal with Samsung to make its next-generation AI6 chips. On August 6, Bloomberg reported that close to 20 Dojo workers left to start their own company called DensityAI. On August 7, Bloomberg reported that Tesla had disbanded its Dojo team and shut down the project, with lead Peter Bannon leaving the company. Musk responded on X, saying it did not make sense for Tesla to divide its resources on two different AI chip designs and that all effort is focused on the AI5, AI6, and subsequent chips. On August 10, Musk posted that once it became clear all paths converged to AI6, he had to shut down Dojo and make tough personnel choices, as Dojo 2 was an evolutionary dead end. He stated that Dojo 3 arguably lives on in the form of AI6 systems-on-a-chip on a single board. On September 1, Tesla shared its Master Plan Part IV on social media with no mention of Dojo or Cortex, although AI and physical AI took a central role.