Elon Musk doesn’t want Tesla to be just an automaker. Tesla wants to be an AI company that has understood how to make cars drive themselves.
The crucial mission for this mission is Dojo, the usual Tesla supercomputer designed to train full driving neural networks (FSD). FSD is not actually fully self-guiding. It may perform some automated driving tasks, but it still requires a careful person behind the steering wheel. But Tesla thinks with more data, more computational power and more training, it can cross the threshold from almost self-guidance to full self-leading.
And that’s where Dojo comes.
Musk teases the dojo for some time, but the executive accelerates discussions about the supercomputer all over 2024. Now that we are in 2025, another supercomputer called Cortex has entered the conversation, but the importance of dojo for Tesla may It is still existential – with EV combating sales, investors want assurances that Tesla can achieve autonomy. Below is a timetable of Dojo also reports the promises.
2019
First reports of Dojo
April 22 – On Tesla’s autonomy day, the automotive industry had the AI team on stage to talk about an automatic pilot and full self-leader, and the AI that feeds both. The company shares information about Tesla’s customized chips designed specifically for neural networks and self-driving.
During the event, Musk teases Dojo, revealing that he is a supercomputer for AI training. He also notes that all Tesla cars produced at that time will have all the material necessary for full self-guiding and only need a software update.
2020
Musk starts Dojo Roadshow
February 2 – Musk she says Tesla will soon have more than a million connected vehicles worldwide to the sensors and the calculation required for full self-guidance-and the potential of Dojo.
“Dojo, the supercomputer of the training, will be able to process huge quantities of video training data and execute efficient outbreaks with a huge number of parameters, plenty of memory range and extremely high -range zone between nuclei. More for this later.”
August 14 – Musk repeats Tesla’s plan to develop a neural networks training computer called Dojo “to really process huge amounts of video data”, called “a beast”. It also says that the first version of Dojo is ‘About a year away,’ which would set the start date somewhere around August 2021.
December 31 – Says Elon Dojo does not needBut it will make self-guidance better. “It is not enough to be safer than human drivers, Autopilot must ultimately be more than 10 times safer than human drivers.”
2021
Tesla makes Dojo Official
August 19 – The automaker officially announces Dojo on Tesla’s first AI day, an event intended to attract engineers to Tesla’s AI team. Tesla also introduces the D1 chip, which the automotive industry says it will use – along with NVIDIA GPU – to supply the DOJO supercomputer. Tesla notes that the AI complex will host 3,000 D1 chips.
October 12 – Tesla releases a WhitePaper of Dojo Technology“A guide to the tesla’s floating point forms and arithmetic.” White paper describes a technical standard for a new type of binary numerical points used in deep learning networks and can be implemented “completely in software, entirely in material or any combination of software and hardware”.
2022
Tesla reveals Dojo’s progress
August 12 – Musk says Tesla will “Phase in Dojo. You will not need to buy so many GPU growth in next year. ”
September 30 – On Tesla’s second AI day, the company reveals that it has installed the first Council of DOJO, testing 2.2 megawatts cargo tests. Tesla says he constructs a tile per day (consisting of 25 d1 chips). The Tesla Demos Dojo on stage runs a fixed diffusion model to create an image created by AI of a “Cybertruck on Mars”.
It is important that the company sets a target date of a complete exapod clump that will be completed with Q1 2023 and says it plans to build a total of seven exapods in Palo Alto.
2023
A long shot bet”
April 19 – Musk tells investors during the earnings of Tesla’s first quarter that Dojo “has the opportunity to make a range of improvement in training costs” and also “has the ability to become a service we will offer to other companies with same way in the same way that Amazon Web Services offers web services.
Musk also notes that “he will look at Dojo as a kind of long shot”, but a “bet worth making”.
June 21 – The Tesla AI x account seats that the company’s neural networks are already in customer vehicles. The thread includes a chart with a timetable of Tesla’s current and intended computational power, which places the start of DOJO production in July 2023, although it is not clear whether it refers to chips D1 or the supercomputer itself. Says Musk On the same day that Dojo was already online and is running work at Tesla Centers.
The company also indicates that Tesla’s calculation will be the top five around the world until February 2024 (there is no evidence that this was successful) and that Tesla will reach 100 exaflops by October 2024.
July 19 – Tesla notes In reporting the second quarter profits has begun the production of Dojo. Musk also says Tesla plans to spend more than $ 1 billion on Dojo by 2024.
September 6 – Moscow to x This Tesla is limited by the calculation of AI training, but that Nvidia and Dojo will correct it. He says data management from the approximately 160 billion frames of the video tesla gets from his cars a day is extremely difficult.
2024
Plans for scale
January 24 – During the fourth quarter and full year of Tesla, Musk again recognizes that Dojo is a high -risk, high reward project. He also says that Tesla is seeking “the double path of Nvidia and Dojo”, that “dojo works” and “makes training jobs”. He notes that Tesla escalates it and has “plans for dojo 1.5, dojo 2, dojo 3 and whatnot”.
January 26 – Tesla announced plans to spend $ 500 million to build a DOJO supercomputer in Buffalo. Musk somewhat reduces investment, Post in x that while $ 500 million is a large amount, it is “only equivalent to a 10K H100 system by NVIDIA. Tesla will pass more of it at the Nvidia Hardware this year. Table soles to be competitive in AI are at least quite billions of dollars per year at this point. “
April 30 – At TSMC’s North America Symposium, the company says that the DOJO-D2 next-generation training tile, which puts the entire dojo tile on a single silicon disk, instead of connecting 25 chips to make a tile-already tile is already in production, according to IEEE spectrum.
May 20 – Musk notes That the back of the expansion of the Giga Texas plant will include the construction of “a super dense, clusters of supercomputing water”.
June 4 – A CNBC report It reveals that Musk diverts thousands of Nvidia chips intended for Tesla in X and XAI. After initially saying that the report was false, Musk positions in x That Tesla had no location to send Nvidia chips to activate them, due to the continued construction to the southern expansion of Giga Texas, “they would have just sat in a warehouse.” He noted that the extension would “host 50K H100s for FSD training”.
He too seats:
“Out of about $ 10b to AI -related costs that I said that Tesla would do this year. About half is internal. DOJO.
July 1 – Musk reveals in x That today’s Tesla vehicles may not have the right material for the next generation AI model. He says that the increase of about 5x in the number of parameters with the next AI “is very difficult to achieve without upgrading the vehicle’s conclusions.”
NVIDIA offer challenges
July 23 – During Tesla’s second trimester call, Musk says that the demand for Nvidia material is “so high that it is often difficult to get GPU”.
“I think this requires that we make much more effort on Dojo to ensure that we have the training capacity we need,” says Musk. “And we see a path to be competitive with Nvidia with Dojo.”
A graph on Tesla’s investor deck predicts that the Tesla AI training capacity will sew about 90,000 H100 GPUs by the end of 2024, from about 40,000 in June. Later that day in X, Moscow This dojo 1 will have “about 8K H100-Israelotes of Training Online by the end of the year”. It also publishes photos of the supercomputer, who seems to use the same exterior of a refrigerator -like stainless steel like Tesla Cybertrucks.
By dojo to the bark
July 30 – AI5 is ~ 18 months from large volume production, Musk says in a answer In a post by someone who claims to start a club “Tesla HW4/AI4 owners angry to stay behind when AI5 comes out”.
August 3 – Musk positions in x That he made a shot of “The Tesla Supercompute Cluster in Giga Texas (also known as Cortex)”. It notes that there will be about 100,000 H100/H200 NVIDIA GPU with “Mass Save for FSD & Optimus video training”.
August 26 – Musk positions in x A video of the crust, referred to as “the giant new AI supercluster built on Tesla HQ in Austin to resolve the real world AI”.
2025
Without updates in Dojo in 2025
January 29 – Tesla’s Q4 and 2024 profits of 2024 did not include any reference to Dojo. The bark, Tesla’s new AI training at Austin Gigafactory, made an appearance. Tesla noted in his own shareholder deck That it completed the growth of the bark, which consists of about 50,000 H100 NVIDIA GPU.
“The bark has helped make the FSD V13 (supervised), which can boast of significant improvements in safety and comfort thanks to increased data 4.2x, higher resolution video inputs … including improvements.” with the letter.
During the call, CFO Vaibhav Taneja noted that Tesla accelerated the construction of the cortex to accelerate the development of the FSD V13. He said that the accumulated capital costs associated with the infrastructure, including infrastructure, “have been about $ 5 billion so far”. In 2025, Taneja said he expects Capex to be flat as it is related to AI.
This story first published on August 10, 2024 and we will inform it as new information is being developed.
