Elon Musk doesn\u2019t want Tesla to be just an automaker. He wants Tesla to be an AI company, one that\u2019s figured out how to make cars drive themselves.\u00a0<\/p>\n
Crucial to that mission was Dojo<\/a>, a custom-built supercomputer designed by Tesla to train its Full Self-Driving (FSD) neural networks. FSD isn\u2019t actually fully self-driving; it can perform some automated driving tasks, but still requires an attentive human behind the wheel. But Tesla thinks with more data, more compute power and more training, it can cross the threshold from almost self-driving to full self-driving.\u00a0<\/p>\n And that\u2019s where Dojo was supposed to come in. <\/p>\n 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 chat. <\/p>\n Below is a timeline of Dojo mentions and promises.\u00a0Check out this explainer<\/a> on what Dojo for even more information on what it was, why it mattered, and what comes next.<\/p>\n 2019<\/strong><\/p>\n First mentions of Dojo<\/strong><\/p>\n April 22 \u2013 <\/strong>At Tesla\u2019s Autonomy Day<\/a>, the automaker had its AI team onstage to talk about Autopilot and Full Self-Driving, and the AI powering them both. The company shares information about Tesla\u2019s custom-built chips that are designed specifically for neural networks and self-driving cars.\u00a0<\/p>\n During the event, Musk teases Dojo, revealing that it\u2019s a supercomputer for training AI.\u00a0He also notes that all Tesla cars being produced at the time would have all hardware necessary for full self-driving and only needed a software update.<\/p>\n Techcrunch event<\/p>\n \n\t\t\t\t\t\t\t\t\tSan Francisco 2020\u00a0<\/strong><\/p>\n Musk begins the Dojo roadshow<\/strong><\/p>\n Feb 2 \u2013 <\/strong>Musk says<\/a> Tesla will soon have more than a million connected vehicles worldwide with sensors and compute needed for full self-driving \u2014 and touts Dojo\u2019s capabilities.\u00a0<\/p>\n \u201cDojo, our training supercomputer, will be able to process vast amounts of video training data & efficiently run hyperspace arrays with a vast number of parameters, plenty of memory & ultra-high bandwidth between cores. More on this later.\u201d<\/p>\n August 14 <\/strong>\u2013<\/strong> Musk reiterates<\/a> Tesla\u2019s plan to develop a neural network training computer called Dojo \u201cto process truly vast amounts of video data,\u201d calling it \u201ca beast.\u201d He also says the first version of Dojo is \u201cabout a year away,\u201d<\/a> which would put its launch date somewhere around August 2021.<\/p>\n December<\/strong> 31<\/strong> \u2013<\/strong> Elon says Dojo isn\u2019t needed<\/a>, but it will make self-driving better. \u201cIt isn\u2019t enough to be safer than human drivers, Autopilot ultimately needs to be more than 10 times safer than human drivers.\u201d<\/p>\n 2021<\/strong><\/p>\n Tesla makes Dojo official<\/strong><\/p>\n August 19 \u2013 <\/strong>The automaker officially announces Dojo at Tesla\u2019s first AI Day<\/a>, an event meant to attract engineers to Tesla\u2019s AI team. Tesla also introduces its D1 chip, which the automaker says it will use \u2014 alongside Nvidia\u2019s GPU \u2014 to power the Dojo supercomputer. Tesla notes its AI cluster will house 3,000 D1 chips.\u00a0<\/p>\n October 12 \u2013 <\/strong>Tesla releases <\/strong>a <\/strong>Dojo Technology whitepaper<\/a>, \u201ca guide to Tesla\u2019s configurable floating point formats & arithmetic.\u201d The whitepaper outlines a technical standard for a new type of binary floating-point arithmetic that\u2019s used in deep learning neural networks and can be implemented \u201centirely in software, entirely in hardware, or in any combination of software and hardware.\u201d<\/p>\n 2022<\/strong><\/p>\n Tesla reveals Dojo progress<\/strong><\/p>\n August 12 \u2013 <\/strong>Musk says Tesla will \u201cphase in Dojo.<\/a> Won\u2019t need to buy as many incremental GPUs next year.\u201d<\/p>\n September 30 \u2013 <\/strong>At Tesla\u2019s second AI Day<\/a>, the company reveals that it has installed the first Dojo cabinet, testing 2.2 megawatts of load testing. Tesla says it was building one tile per day (which is made up of 25 D1 chips). Tesla demos Dojo onstage running a Stable Diffusion model to create an AI-generated image of a \u201cCybertruck on Mars.\u201d<\/p>\n Importantly, the company sets a target date of a full Exapod cluster to be completed by Q1 2023, and says it plans to build a total of seven Exapods in Palo Alto.\u00a0<\/p>\n 2023<\/strong><\/p>\n A \u2018long-shot bet<\/strong>\u2018<\/p>\n April 19 \u2013 <\/strong>Musk tells investors during Tesla\u2019s first-quarter earnings<\/a> that Dojo \u201chas the potential for an order of magnitude improvement in the cost of training,\u201d and also \u201chas the potential to become a sellable service that we would offer to other companies in the same way that Amazon Web Services offers web services.\u201d<\/p>\n Musk also notes that he\u2019d \u201clook at Dojo as kind of a long-shot bet,\u201d but a \u201cbet worth making.\u201d<\/p>\n June 21 <\/strong>\u2013<\/strong> The Tesla AI X account posts<\/a> that the company\u2019s neural networks are already in customer vehicles. The thread includes a graph with a timeline of Tesla\u2019s current and projected compute power, which places the start of Dojo production at July 2023, although it\u2019s not clear if this refers to the D1 chips or the supercomputer itself. Musk says<\/a> that same day that Dojo was already online and running tasks at Tesla data centers.\u00a0<\/p>\n The company also projects that Tesla\u2019s compute will be the top five in the entire world by around February 2024 (there are no indications this was successful) and that Tesla would reach 100 exaflops by October 2024.<\/p>\n July 19 \u2013 <\/strong>Tesla notes<\/a> in its second-quarter earnings report it has started production of Dojo. Musk also says Tesla plans to spend more than $1 billion on Dojo through 2024.\u00a0\u00a0<\/p>\n September 6 \u2013 <\/strong>Musk posts on X<\/a> that Tesla is limited by AI training compute, but that Nvidia and Dojo will fix that. He says managing the data from the roughly 160 billion frames of video Tesla gets from its cars per day is extremely difficult.\u00a0<\/p>\n 2024<\/strong><\/p>\n Plans to scale<\/strong><\/p>\n January 24 \u2013 <\/strong>During Tesla\u2019s fourth-quarter and full-year earnings call, Musk acknowledges again that Dojo is a high-risk, high-reward project. He also says that Tesla was pursuing \u201cthe dual path of Nvidia and Dojo,\u201d that \u201cDojo is working\u201d and is \u201cdoing training jobs.\u201d He notes Tesla is scaling it up and has \u201cplans for Dojo 1.5, Dojo 2, Dojo 3 and whatnot.\u201d<\/p>\n January 26 \u2013 <\/strong>Tesla announced plans to spend $500 million to build a Dojo supercomputer in Buffalo<\/a>. Musk then downplays the investment somewhat, posting on X<\/a> that while $500 million is a large sum, it\u2019s \u201conly equivalent to a 10k H100 system from Nvidia. Tesla will spend more than that on Nvidia hardware this year. The table stakes for being competitive in AI are at least several billion dollars per year at this point.\u201d<\/p>\n April 30 \u2013 <\/strong>At TSMC\u2019s North American Technology Symposium, the company says Dojo\u2019s next-generation training tile \u2014 the D2, which puts the entire Dojo tile onto a single silicon wafer, rather than connecting 25 chips to make one tile \u2014 is already in production, according to IEEE Spectrum<\/a>.\u00a0<\/p>\n May 20 \u2013 <\/strong>Musk notes<\/a> that the rear portion of the Giga Texas factory extension will include the construction of \u201ca super dense, water-cooled supercomputer cluster.\u201d<\/p>\n June 4 \u2013 <\/strong>A CNBC report<\/a> reveals Musk diverted thousands of Nvidia chips reserved for Tesla to X and xAI. After initially saying the report was false, Musk posts on X<\/a> that Tesla didn\u2019t have a location to send the Nvidia chips to turn them on, due to the continued construction on the south extension of Giga Texas, \u201cso they would have just sat in a warehouse.\u201d He noted the extension will \u201chouse 50k H100s for FSD training.\u201d\u00a0\u00a0\u00a0<\/p>\n He also posts<\/a>:\u00a0<\/p>\n \u201cOf the roughly $10B in AI-related expenditures I said Tesla would make this year, about half is internal, primarily the Tesla-designed AI inference computer and sensors present in all of our cars, plus Dojo. For building the AI training superclusters, NVidia hardware is about 2\/3 of the cost. My current best guess for Nvidia purchases by Tesla are $3B to $4B this year.\u201d<\/p>\n July 1 \u2013 <\/strong>Musk reveals on X<\/a> that current Tesla vehicles may not have the right hardware for the company\u2019s next-gen AI model. He says that the roughly 5x increase in parameter count with the next-gen AI \u201cis very difficult to achieve without upgrading the vehicle inference computer.\u201d<\/p>\n Nvidia supply challenges<\/p>\n July 23 \u2013 <\/strong>During Tesla\u2019s second-quarter earnings call, Musk says demand for Nvidia hardware is \u201cso high that it\u2019s often difficult to get the GPUs.\u201d\u00a0<\/p>\n \u201cI think this therefore requires that we put a lot more effort on Dojo in order to ensure that we\u2019ve got the training capability that we need,\u201d Musk says. \u201cAnd we do see a path to being competitive with Nvidia with Dojo.\u201d<\/p>\n A graph in Tesla\u2019s investor deck predicts that Tesla AI training capacity will ramp to roughly 90,000 H100 equivalent GPUs by the end of 2024, up from around 40,000 in June. Later that day on X, Musk posts<\/a> that Dojo 1 will have \u201croughly 8k H100-equivalent of training online by end of year.\u201d He also posts photos<\/a> of the supercomputer, which appears to use the same fridge-like stainless steel exterior as Tesla\u2019s Cybertrucks.\u00a0<\/p>\n From Dojo to Cortex<\/p>\n July 30 \u2013\u00a0 <\/strong>AI5 is ~18 months away from high-volume production, Musk says in a reply<\/a> to a post from someone claiming to start a club of \u201cTesla HW4\/AI4 owners angry about getting left behind when AI5 comes out.\u201d\u00a0<\/p>\n August 3 \u2013 <\/strong>Musk posts on X<\/a> that he did a walkthrough of \u201cthe Tesla supercompute cluster at Giga Texas (aka Cortex).\u201d He notes that it would be made roughly of 100,000 H100\/H200 Nvidia GPUs with \u201cmassive storage for video training of FSD & Optimus.\u201d<\/p>\n August 26 \u2013 <\/strong>Musk posts on X<\/a> a video of Cortex, which he refers to as \u201cthe giant new AI training supercluster being built at Tesla HQ in Austin to solve real-world AI.\u201d\u00a0<\/p>\n 202<\/strong>5<\/p>\n Dojo shutdown, its team disbanded<\/strong><\/p>\n January 29 <\/strong>\u2013 Tesla\u2019s Q4 and full-year 2024 earnings call<\/a> included no mention of Dojo. Cortex, Tesla\u2019s new AI training supercluster at the Austin gigafactory, did make an appearance, however. Tesla noted in its shareholder deck<\/a> that it completed the deployment of Cortex, which is made up of roughly 50,000 H100 Nvidia GPUs.\u00a0<\/p>\n \u201cCortex helped enable V13 of FSD (Supervised), which boasts major improvements in safety and comfort thanks to 4.2x increase in data, higher resolution video inputs \u2026 among other enhancements,\u201d according to the letter.\u00a0<\/p>\n During the call, CFO Vaibhav Taneja noted that Tesla accelerated the buildout of Cortex to speed up the rollout of FSD V13. He said that accumulated AI-related capital expenditures, including infrastructure, \u201cso far has been approximately $5 billion.\u201d In 2025, Taneja said he expects capex to be flat as it relates to AI.<\/p>\n July 23 \u2013 <\/strong>During Tesla\u2019s Q2 2025 earnings call, Musk said<\/a> Dojo 2 was expected to be \u201coperating at scale\u201d sometime in 2026, with \u201cscale being somewhere around 100k H100 equivalent.\u201d In the same breath, he hinted at possible redundancies.\u00a0<\/p>\n \u201cThinking about Dojo 3 and the AI6 inference chip, it seems like intuitively, we want to try to find convergence there, where it\u2019s basically the same chip,\u201d Musk said.\u00a0<\/p>\n July 28 \u2013 <\/strong>Tesla signed a $16.5 billion deal<\/a> to get its next-generation AI6 chips from Samsung. The AI6 chip is Tesla\u2019s bet on a chip design that can scale from powering FSD and Tesla\u2019s Optimus humanoid robots to high-performance AI training in data centers.\u00a0<\/p>\n August 6 \u2013 <\/strong>Bloomberg reports that close to 20 Dojo workers left to start their own company that builds AI chips, software, and hardware called DensityAI.\u00a0<\/a><\/p>\n August 7 \u2013 <\/strong>Bloomberg reports that Tesla has disbanded its Dojo team and shut down the project. Peter Bannon, Dojo\u2019s lead, left the company, as well.\u00a0<\/p>\n Musk responded<\/a> to the reports on X, saying: \u201cIt doesn\u2019t make sense for Tesla to divide its resources and scale two quite different AI chip designs. The Tesla AI5, AI6 and subsequent chips will be excellent for inference and at least pretty good for training. All effort is focused on that.\u201d<\/p>\n August 10 \u2013 <\/strong>\u201cOnce it became clear that all paths converged to AI6, I had to shut down Dojo and make some tough personnel choices, as Dojo 2 was now an evolutionary dead end,\u201d Musk posted on X<\/a>, the social media platform he owns. \u201cDojo 3 arguably lives on in the form of a large number of AI6 [systems-on-a-chip] on a single board.\u201d<\/p>\n September 1<\/strong> \u2013 Tesla shares its Master Plan Part IV<\/a> on social media platform X. There is no mention of Dojo or Cortex, although AI, and more specifically, \u201cphysical AI\u201d takes a central role. <\/p>\n This story originally published August 10, 2024. The final update of the Tesla Dojo timeline published September 2, 2025.<\/p>\n
\n\t\t\t\t\t\t\t\t\t\t\t\t\t|
\n\t\t\t\t\t\t\t\t\t\t\t\t\tOctober 27-29, 2025\n\t\t\t\t\t\t\t<\/p>\n