Dojo, the cornerstone of Tesla's AI empire: born for video data training, paving the way to general artificial intelligence (AGI)

On September 11th, local time Tesla stocks surged more than 10%, increasing to 122% during the year, and market capitalization increased by USD 79.6 billion in a single day to USD 868.3 billion.

The lead that ignited the surge came from a Morgan Stanley report — Damo pointed out that the Dojo supercomputer would increase Tesla's market value by as much as USD500 billion, while raising Tesla's benchmark price target price to USD400. For reference, after a sharp rise on Monday, Tesla's stock price closed at USD273.58.

Analysts Adam Jonas and Daniela M Haigian's team pointed out in the report that autonomous driving systems are known as the “mother of artificial intelligence projects”. In the process of seeking solutions to autonomous driving problems, Tesla developed Dojo supercomputing, which can open up “new potential markets” for Tesla.

What is Dojo?

Dojo is a supercomputer developed by Tesla. It can use massive amounts of video data to complete “unsupervised” data labeling and training.

In the literal sense, Dojo means “dojo, martial arts hall,” which is in line with its meaning — a training ground built by Tesla for AI.

On AI Day 2021, Tesla already unveiled the Dojo supercomputer, but at the time it was “fleeting”; it still only had the first chip and training block, and the company was still pushing to build a complete Dojo Exapod.

Meanwhile, Tesla also said that, in theory, Dojo Exapod will be the fastest AI training supercomputer in the world.After that, Dojo Exapod was finally unveiled. Each Dojo Exapod integrates 120 training modules, has 3,000 built-in D1 chips, has more than 1 million training nodes, and has a computing power of 1.1 EFLOP* (1,000 trillion floating-point operations per second). In terms of microarchitecture, each Dojo node has a core and is a mature computer with dedicated CPU memory and I/O interfaces.

Currently, Dojo is being used for artificial intelligence, machine learning and computer vision training.Tesla has been producing the Dojo supercomputer since July of this year, an important step towards faster and cheaper neural network training.

The company plans to invest more than USD 1 billion in Dojo by the end of 2024. In the future, Tesla plans to use it at the same time NVIDIA and Dojo's computing power.

According to the computing power development plan released by Tesla in June, Dojo will become the top five computing power facilities in the world in the first quarter of next year, and will reach 100 eFlops computing power in October next year.

The cornerstone of Musk’s “AI empire”

Arguably, Dojo has become the cornerstone facility of Musk's “AI empire.”

Looking back at 2019, when Musk first introduced Dojo to the public, he said, “Tesla does have a major project, let's call it Dojo. It's a super powerful training computer that aims to input massive amounts of data and train on a video level... With the Dojo computer, we can train on a large scale without supervision on a large number of videos.”

Indeed, if the nourishment of artificial intelligence is data, then the nourishment for Tesla's autonomous driving is video data.

In order to achieve complete neural networks rather than code control, the FSD V12 gets about 160 billion frames of video from the Tesla team every day for training, yet less than 1% of the videos are the most useful.

Musk said that the neural network envisioned by Tesla must undergo at least 1 million video training before it can be formed. By the beginning of this year, FSD V12 had analyzed 10 million videos.

At the end of August, Musk had just “shown” the FSD V12 via live broadcast. In this live broadcast, vehicles are able to excuse pedestrians, avoid roadblocks, and turn at intersections on their own on roads that are not pre-set. Musk has emphasized many times that there is not a single line of code corresponding to FSD V12, and that the vehicle is artificially set to perform these actions --FSD 12 accomplishes these actions entirely as a result of extensive video training.

Through video training data, AI can learn to drive on its own and “do things like a human”.

Of course, mediocre and random data is not enough; the data supplied to the neural network needs to be carefully selected.

Musk also placed special emphasis on high quality data from great drivers, is the key to training Tesla for autonomous driving.

“A large amount of mediocre data doesn't improve driving, and data management is quite difficult. We have a lot of software that can control what data the system selects and what data to train.”

The path to general artificial intelligence

Today, for Musk, the value of Dojo is not limited to the autonomous driving business; in fact, Dojo has become the computing power infrastructure for the development of Tesla's AI business system as a whole.

About 70% of human information is obtained through visual perception, which is also the Tesla plans to prepare for automobiles and robots.

Tesla has revealed before, the head of the Tesla robot Tesla Bot “Optimus Prime” will be equipped with the same intelligent driving camera as its own car, and share an AI system with the car. In other words, Tesla humanoid robots continue the vision-oriented sensing technology route.

Musk revealed in June of this year that the Dojo has been online and in operation for a few months, and that it is not only suitable for Tesla's fully autonomous driving. In addition, he said, Dojo V1 is highly optimized for large-scale video training, not general purpose AI; however, Dojo V2 will break through this limitation.

This also means, the upgraded version of Dojo is more likely to target general artificial intelligence (AGI).

The Morgan Stanley report mentioned earlier also mentioned this — after comparing it with other technology companies' supercomputing, Damo found that Dojo's future looked brighter. Considering Tesla's upcoming autonomous robot taxis and internet services, the launch of Dojo seems more clear, and may drive a major upgrade in Tesla's ecosystem.