Elon Musk doesn’t want Tesla to be just an automaker. He hopes Tesla will become an AI company that figures out how to get cars to drive.
Crucial to that task is DojoThis is a custom supercomputer designed by Tesla to train its complete autonomous driving (FSD) neural network. FSD is not actually completely autonomous; it can perform some autonomous driving tasks, but it still requires a careful person. But Tesla uses more data, more computing power and more training to think about it, and it can go from almost autonomous driving to fully autonomous driving thresholds.
This is where Dojo should enter.
Musk laughed at Dojo for years and discussed the supercomputer in 2024. But Dojo is now out, and another supercomputer called Cortex has entered the chat.
Here is a timeline for Dojo’s mentions and commitments. View this interpreter More information about Dojo, about what it is, why it matters and what happens next.
2019
Dojo’s first mention
April 22 – exist Tesla’s Autonomous Daythe automaker’s AI team is on the stage talking about autonomous driving and all autonomous driving, and AI powers them. The company shares information about Tesla’s custom chip, designed specifically for neural networks and autonomous vehicles.
During the event, Musk teased the dojo, showing that it was a supercomputer that trains AI. He also pointed out that all Tesla cars produced at the time would have all the hardware needed for full autonomous driving and would only require software updates.
TechCrunch Events
San Francisco
|
October 27-29, 2025
2020
Musk starts the dojo road show
February 2 – Musk explain Tesla will soon have more than a million connected vehicles around the world, with sensors and complete computing required for autonomous driving – and the capabilities of Dojo.
“Our training supercomputer Dojo will be able to process large amounts of video training data and effectively run hyperspace arrays with a large number of parameters, ultra-high bandwidth between large amounts of memory and cores. Do this information later.”
August 14 – Musk reiterate Tesla plans to develop a neural network training computer called “Dojo” to “process real video data” and call it the “beast.” He also said that the first version of Dojo is “About a year” This will put its release date somewhere around August 2021.
December 31 – Elon says No Dojo requiredbut this will make autonomous driving better. “This isn’t enough to be safer than human drivers, and autopilots ultimately need to be more than 10 times safer than human drivers.”
2021
Tesla makes Dojo official
August 19 – Automaker officially announces Dojo in Tesla’s first AI dayThe campaign aims to attract engineers to Tesla’s AI team. Tesla also introduced its D1 chip, which the automaker said will be used with Nvidia’s GPU to power the Dojo SuperComputer. Tesla notes that its AI cluster will house 3,000 D1 chips.
October 12 – Tesla releases one Dojo Technology White Paper“Tesla’s configurable floating point formats and arithmetic guide.” The white paper outlines a new technical standard for binary floating point arithmetic arithmetic used in deep learning neural networks that can be “completely in software, entirely implemented in hardware or in combinations of software and hardware.”
2022
Tesla reveals progress in dojo
August 12 – Musk says Tesla willThe stage is in Dojo. You won’t need to buy that many incremental GPUs next year. ”
September 30 – In Tesla The second AI dayThe company revealed that it has installed its first Dojo cabinet and tested a 2.2 MW load test. Tesla says it builds one ceramic tile every day (composed of 25 chips). Tesla Demos Dojo runs a stable diffusion model on the stage to create AI-generated images of “Seberak on Mars”.
Importantly, the company sets a target date for a full Exapod cluster that will be completed in Q1 2023 and says it plans to build seven Exapods in Palo Alto.
2023
A’thong to bet‘
April 19 – Musk tells investors Tesla’s first-quarter revenue Dojo “has the potential to improve the cost of training” and “has the potential to be a sellable service that we will provide to other companies in the same way that Amazon Web Services provides web services.”
Musk also noted that he “sees Dojo as a long-term bet” but “a bet worth it.”
June 21 – Tesla AI X Account Posts The company’s neural network is already in the customer’s vehicle. The thread contains a chart with Tesla’s current and projected computing power schedule, which will begin the start of Dojo production in July 2023, although it is not clear whether this refers to the D1 chip or the supercomputer itself. Musk said Dojo is already online and running the task on the same day in the Tesla data center.
The company also predicts that Tesla’s calculations will be the top five in the world by around February 2024 (no sign of success) and Tesla will not reach 100 Exaflops from October 2024 to October 2024.
July 19 – Tesla notes In its second quarter earnings report, it has started producing Dojo. Musk also said Tesla plans to spend more than $1 billion on Dojo by 2024.
September 6 – Musk column On X Tesla is limited by AI-trained computing, but Nvidia and Dojo will solve this problem. He said it was very difficult to manage data managed by Tesla about 160 billion frames of data Tesla got from cars.
2024
Plan to expand
January 24 – During Tesla’s fourth-quarter and full-year earnings call, Musk reiterated that Dowjo is a high-risk, high-reward project. He also said Tesla is pursuing the “dual path of Nvidia and Dojo”, i.e. “Dojo is working” and is “engaging in training.” He noted that Tesla is expanding and has developed “plans for Dojo 1.5, Dojo 2, Dojo 3 and Whatnot”.
January 26 – Tesla announces plans to spend $500 million to build one Dojo supercomputer in Buffalo. Then Musk downplayed the investment to some extent, Posted on x While the $500 million is a good amount, “just the equivalent of the 10K H100 system from NVIDIA. Tesla will spend more this year than Nvidia Hardware. Currently, the competitive table shares in AI are at least billions at this point.”
April 30 – The company said at TSMC’s North American Technology Symposium that Dojo’s next-generation training tiles – D2, putting the entire Dojo tiles on a silicon wafer instead of connecting 25 chips to make a tiles – already in production – IEEE spectrum.
May 20 – Musk notes The rear of the Texas Gee Processing Plant expansion will include the construction of a “super dense, water-cooled supercomputer cluster.”
June 4 – one CNBC Report Revealing Musk has transferred thousands of Nvidia chips for Tesla to X and Xai. After initially saying that the report was wrong Musk column on X Because of the southern extension of Giga Dexas, Texas, “Tesla didn’t send Nvidia chips to open it,” so they just sat in the warehouse. “He noted that the extension will “accommodate 50k H100 for FSD training.”
So does he Posts:
“Of all the approximately $10b spend I said Tesla will make this year, about half is internal, mainly Tesla-designed AI inference computers and Tesla-type AI inference computers and sensors that exist in all of our cars, plus Dojo. For building AI training super public coke, NVIDIA hardware is the best cost for $NVIA $3 in year.”
July 1 – Musk reveals on x Current Tesla cars may not have the right hardware for the company’s next-generation AI model. He said that with the number of parameters for the next generation of AI increased by about five times, “it would be difficult to achieve without upgrading the vehicle inference computer.”
NVIDIA Supply Challenge
July 23 – During Tesla’s second-quarter earnings call, Musk said the demand for NVIDIA hardware was “too high that it’s often difficult to get a GPU.”
“So, I think it requires us to put in more effort on Dojo to make sure we have the training capabilities we need,” Musk said. “And, we do see the path to compete with Dojo with Nvidia.”
A graph on Tesla’s investor deck predicts that Tesla’s AI training capability will rise to about 90,000 H100 equivalent GPUs by the end of 2024, up from 40,000 in June. On X later that day Musk column Dojo 1 will “train about 8K H100 equivalent online by the end of the year”. He also released photo The supercomputer appears to use the same refrigerator-like stainless steel appearance as Tesla’s Cybertrucks.
From Dojo to Cortex
July 30 – Musk is about 18 months away from mass production reply Post by someone who claimed to have founded “Tesla HW4/AI4 owners were angry and were left behind when AI5 came out.”
August 3 – Musk column on X He performed a drill in the Tesla SuperCompute cluster in Giga, Texas (aka Cortex). “He noted that it will be made from approximately 100,000 H100/H200 NVIDIA GPUs, “a large-scale storage for video training for FSD & Optimus.”
August 26 – Musk column on X His video tape, which he calls “a giant new AI training supercluster built at Tesla’s headquarters in Austin to solve real-world AI.”
2025
Dojo closes, its team disbands
January 29 – Tesla’s fourth quarter and full-year 2024 earnings call Dojo is not included. Tesla is Austin Gigafactory’s new AI training supergroup Cortex has indeed made a appearance. Tesla in its Shareholder deck It completes the deployment of the cortex consisting of approximately 50,000 H100 NVIDIA GPUs.
“Cortex helped enable FSD (supervised) V13, which adds 4.2x higher resolution video input to the data…and other enhancements for significant security and comfort,” the letter said.
During the call, CFO Vaibhav Taneja pointed out that Tesla has accelerated the construction of Cortex to speed up the launch of FSD V13. He said the accumulated AI-related capital expenditure, including infrastructure, “has about $5 billion so far.” Taneja said in 2025 that he hopes Capex is related to AI.
July 23 – Musk on Tesla’s second quarter 2025 revenue call explain The Dojo 2 is expected to be “operated on a large scale” sometime in 2026, with a ratio of about 100k H100 equivalents. The same breath he hints at possible layoffs.
“Thinking about the Dojo 3 and AI6 reasoning chips, it seems intuitively that we want to try to find a convergence there, basically the same chip,” Musk said.
July 28 – Tesla signed a $16.5 billion deal Get the next-generation AI6 chip from Samsung. The AI6 chip is Tesla’s bet on chip design, ranging from powering the Optimus humanoid biobots for FSD and Tesla to high-performance AI training in data centers.
August 6 – Bloomberg reports that the remaining nearly 20 Dojo workers have founded their own company, which has built AI chips, software and hardware called denseai.
August 7 – Bloomberg reported that Tesla had disbanded the Dojo team and closed the project. Dojo’s leader Peter Bannon also left the company.
Musk answer “It doesn’t make sense for Tesla to split its resources and expand two completely different AI chip designs. Tesla AI5, AI6 and subsequent chips are very useful for reasoning and at least very good for training. All efforts are focused on this.”
August 10 – “Once it became clear that all the paths were integrated into AI6, I had to turn off Dojo and make some tough personnel choices, and Dojo 2 is now a dead end to evolution,” he said. Musk posted on Xa social media platform he owns. “Dojo 3 can be said to live in the form of a lot of AI6 [systems-on-a-chip] On a board. ”
September 1 – Tesla Share Master Plan Part IV On social media platform X. There is no mention of Dojo or Cortex, although AI, more specifically, “physical AI” plays a central role.
This story was originally published on August 10, 2024. The final update of Tesla’s dojo schedule was released on September 2, 2025.
