Physical Address
304 North Cardinal St.
Dorchester Center, MA 02124
Elon Musk does not want Tesla to be a car manufacturer. He wants Tesla AI company to figure out how to drive themselves to cars themselves.
The decisive importance of this mission is DOJO, Tesla’s custom -made supercomputer, which aims to train full self -propelled (FSD) neural networks. FSD is not really self -driving; You can do some automated driving tasks, but still require an attentive person behind the government. However, Tesla thinks with more data, more power and more workouts to cross the threshold from almost independent driving to full self -driving.
And here Dojo comes in.
Musk has been teasing Dojo for a while, but in 2024 he had discussions about the supercomputer. Now that we are in 2025, another supercomputer called Cortex has entered the chat, but Dojo’s importance for Tesla remains existential -EV -EV -EV -EV. The decline in sales and investors want guarantees to make Tesla reach autonomy. Below is a schedule for mentioning and promising Dojo.
2019
The first mentions dojo
April 22 – On the day of Tesla Autonomy, the AI team of the car manufacturer was on stage to talk about autopilot and entire self-piercing, and AI feeds both. The company shares information on Tesla’s custom -made chips, designed specifically for neural networks and self -propelled cars.
During the event, Musk teases Dojo, finding that this is a supercomputer for AI training. He also notes that all Tesla cars produced at the time would have all the hardware needed for complete self-driving and only need a software update.
2020
The musk starts the dojo roadhow
February 2 – Musk tell Tesla will soon have more than one million connected vehicles with sensors and the calculation of full self-propelled and dojo capabilities.
“Dojo, our training supercomputer, can process a huge amount of video training data and effectively run hyperspace arrays with a large number of parameters, lots of memory and ultra-high bandwidth between the seeds. More about this later. ”
On August 14th – – Musk Tesla’s plan to develop a neural network training computer called Dojo is to “truly process a huge amount of video data” called “monster”. He also says that the first version of Dojo is the one “In about a year” which is located in August 2021.
December 31 – – Elon says Dojo is not requiredBut that makes self -driving better. “Not enough to be safer than human drivers, Autopilot eventually must be more than 10 times safer than human drivers.”
2021
Tesla makes Dojo official
August 19 – The car manufacturer officially announces Dojo -ta Tesla’s first AI day, an event that attracts Tesla AI. Tesla also introduces the D1 chip, which the car manufacturer claims to use the Dojo supercomputer, along with NVIDIA GPU. Tesla notes that the AI cluster will contain 3000 D1 chipset.
October 12 – Tesla is released THE Dojo Technology Whitapaper“Guide to Tesla configurable floating -point formats and arithmetic.” Whitapaper outlines the technical standard for a new type of binary floating point arithmetic used in deep learning neural networks and can be “fully in software, completely hardware, or in any combination of software and hardware”.
2022
Tesla reveals dojo’s progress
August 12 – Musk says Tesla will “Phase in Dojo. You don’t have to buy so much incremental GPU next year. ”
September 30 – On the second day of Tesla, the company finds out that it has installed the first dojo cabinet, testing a 2.2 megawatt load test. According to Tesla, one tile is built daily (consisting of 25 D1 chips). Tesla Demos Dojo Oncest Stage runs a stable diffusion model to create an AI -generated image of a “Cybertruck on Mars”.
An important aspect is that the company sets the target of a complete exapod cluster that is completed in Q1 2023 and claims to be planning a total of seven Exapod in Palo Altó.
2023
The ‘long -term stake“
April 19 – Musk tells investors during Tesla’s first quarter income that Dojo “has the opportunity to improve the cost of training” and “have the opportunity to become a sale that would be offered to other companies in the same way. that Amazon Web Services offers web services. ”
Musk also notes that “he looked at Dojo as a long-term stake”, but a “stake is worth doing.”
On June 21st – – The Tesla Ai X Account post That the company’s neural network is already in customer vehicles. The fiber contains a graph with a timeline of Tesla’s current and forecast calculation performance, which sets the dojo production start in July 2023, although it is not clear that it refers to the D1 chips or the supercomputer itself. Musk says On the same day Dojo was online and run tasks in Tesla Data Centers.
The company also plans to calculate Tesla around February 2024 around February 2024 (there is no sign that it was successful) and that Tesla will reach 100 Exaflop by October 2024.
July 19 – Tesla note In his second quarter revenue report, he began production of Dojo. Musk also says Tesla spends more than $ 1 billion on Dojo until 2024.
September 6 – Musk comments x that Tesla is limited by the AI training calculation, but NVIDIA and Dojo fix it. He says it is extremely difficult for approx.
2024
Scale
January 24 – During Tesla’s fourth quarter and full annual revenue call, Musk once again recognizes that Dojo is a high -risk, high -rewarded project. He also says that Tesla continued the “double path of NVIDIA and Dojo” to “Dojo work” and do “training work”. He notes that Tesla sizes it and “has plans for Dojo 1.5, Dojo 2, Dojo 3 and Whatnot” plans.
January 26 – Tesla has announced that it will spend $ 500 million on a dojo supercomputer in Buffalo. Musk then somewhat underestimates the investment, Dispatch x While $ 500 million is a large amount, “only the NVIDIA 10K H100 system is equivalent. Tesla will spend more on NVIDIA hardware this year. The desktop stake in AI’s competitiveness at this point is at least billions of dollars a year. ”
April 30 – At the TSMC’s North American technology symposium, the company says the dojo is a next-generation training tile D2, which puts the entire dojo tile on a single silicon wafer, instead of connecting 25 chips to make a tile-according to manufacture, , according to IEEE spectrumOr
May 20 – Musk note that the back part of the Giga Texas factory extension includes the structure of the “super dense, water -cooled supercomputer cluster”.
June 4 – THE CNBC report Musk explores that it has diverted thousands of NVIDIA chips to Tesla to X and Xai. After initially they said the meaning is false, Musk comments on x That Tesla did not have a place where you can send the NVIDIA chipset to turn on them, due to the constant construction of the giga Texas southern extension: “So they would have been in one warehouse.” He noted that the extension will “50k H100 house for FSD training”.
He too post:
“For AI-related expenses, about $ 10 billion, I said Tesla will take place this year, about half of the internal, primarily Tesla-designed AI inference computer and sensors that appear in all our cars, plus dojo. NVIDIA hardware is approx. 2/3. Tesla is the best invention of NVIDIA purchases this year is $ 3B to $ 4B. ”
July 1 – – – The musk turns out on x Current Tesla vehicles may not have the right hardware for the company’s next -generation AI model. He says that the parameter number is approximately 5x increase with the next generation AI “It is very difficult to reach without updating the vehicle’s injustment computer”.
Nvidia supply challenges
July 23 – During Tesla’s second quarter revenue call, Musk said demand for NVIDIA hardware is “so high that it is often difficult to acquire GPUs.”
“I think this requires a lot more effort to make Dojo to ensure that we get the necessary training skills,” says Musk. “And we see a way to be competitive with NVIDIA with Dojo.”
The graph on Tesla’s investor -deck of Tesla predicts that Tesla AI will increase to about 90,000 H100 equivalent GPU by the end of 2024, compared to about 40,000 in June. Later on that day on x, Musk comments that Dojo 1 will be “about 8K H100 equivalent by the end of the year”. She also comments photo From the supercomputer that seems to use the same refrigerator-like stainless steel exterior as Tesla Cybertrucks.
From dojo to bark
July 30 – Ai5 ~ 18 months away from large volume production-says Musk a response For a comment that claims to launch the “Tesla HW4/AI4 owners” club to miss when the AI5 appears. ”
August 3rd – Musk comments on x that he made an overview of the “Tesla Super Comput Cluster in Giga Texas (also known as Cortex)”. He notes that this is made from approximately 100,000 H100/H200 NVIDIA GPU, “Massive storage for FSD and Optimus video.”
August 26th – Musk comments on x Cortex’s video, called “a huge new AI training super clast, built in Austini Tesla HQ to solve the real world.”
2025
There is no update from Dojo in 2025
January 29th – In the fourth quarter of Tesla and the total annual revenue call of 2024 did not contain a mention of Dojo. However, Cortex, Tesla’s new AI coach’s super -cluster, appeared at Austin Gigafactory. Tesla noted shareholder deck that you have finished installing Cortex, which consists of about 50,000 H100 NVIDIA GPUs.
“Cortex has contributed to the FSD (Supervision) V13, which boasts significant improvement in security and comfort due to a 4.2x increase in data, including higher -resolution video inputs …,” the letter said.
During the call, Vaibhav Taneja CFO noted that Tesla had accelerated the formation of the bark to accelerate the introduction of the FSD V13. He said that the capital costs associated with AI, including infrastructure, were “about $ 5 billion so far”. In 2025, Tanea said she expected Capex to be flat as it was related to AI.
This story was originally published on August 10, 2024 and will be updated by developing new information.
