Trial focuses on Tesla’s supercomputer

Tesla is suing a former employee for stealing trade secrets. It draws attention to what the electric car manufacturer calls Project Dojo. What does Tesla want with a supercomputer?

Alexander Yatskov did not have a very good week last week. The engineer found a summons on the doormat. His former employer, electric car maker Tesla, is suing Yatskov for stealing classified business information.

The essence

  • Tesla has sued a former employee for allegedly stealing trade secrets from Project Dojo.
  • The Dojo is the supercomputer with unprecedented computing power that Tesla is puzzling with.
  • Tesla wants to use the Dojo to train its artificial intelligence at a faster pace. It should shorten the road to the self-driving car.

As an engineer, Yatskov was involved in what Tesla has dubbed Project Dojo, a project involving the development of a supercomputer. According to Tesla, Yatskov downloaded a lot of data about the project to a personal computer, against the rules. In addition, after being confronted and asked to hand over his laptop, he must have returned a “dummy” – an exact copy of his work laptop, but without a trace of any mission-critical information on it.

The lawsuit is not only interesting for lawyers. She also focuses on the project that Tesla has been working on since 2019. Then CEO Elon Musk announced via Twitter that his company is working on a supercomputer. “It’s a beast,” it sounded lyrical at the time, but it was a long wait for a glimpse of what the Dojo can really do for Tesla.

In August last year, the company lifted a corner of the veil during its ‘AI Day’, a press day where everything related to artificial intelligence at Tesla was presented. In a nutshell, Dojo’s goal is to quickly process and analyze huge amounts of data from thousands of Teslas on the field.


Dojo is what technicians describe as a ‘neural network’, a computer system that, fed with data, is capable of ‘learning’ through algorithms. The vast amounts of data – including videos of the road from driving Teslas – are being analyzed to train the artificial intelligence Tesla is working on for its self-propelled software. This software is constantly being improved and then shipped to Tesla drivers.

When Tesla was not happy with the range of supercomputers on the market, Tesla decided to build one of its own. And Tesla sets the bar particularly high. The company says it wants to build a computer with a computing power of more than 1 exaFLOP, a milestone that no company, research institution or government has yet managed to complete.

FLOPS are the units in which the computing power of computers is expressed – the acronym stands for ‘floating point operations per second’. The most powerful supercomputer can be found in Japan today at the RIKEN Center for Computational Science. The Fugaku computer currently has a measured computing power of 442,010 teraFLOPS. That’s less than half of what Tesla aims to do with the Dojo.

For the number nerds: 1 exaFLOP is good for 1 million teraFLOPS or a quintillion (1018) FLOPS. To match what a computer with 1 exaFLOP can do in a second, perform a calculation per second. second for 31,688,765,000 years.



To match what a computer with 1 exaFLOP can do in a second, perform a calculation per second. second for 31,688,765,000 years.

Tesla would like to get there in steps. In August, the company introduced its proprietary D1 chip, which would individually accommodate 362 teraFLOPS capacity. The intention is to collect these chips in what the company calls ‘Dojo tiles’, which together would have 9,000 teraFLOPS computing power. In a first phase, these tiles could then be reassembled into ‘cabinets’. And the ultimate goal is to be able to assemble 10 cabinets to get to 1.1 exaFLOP.

second league

Opinions are divided among experts about Dojo’s plans, and whether it would play all the way in the same league as existing supercomputers. According to some experts, it is unfair to compare a project like Dojo, built with a very specific task in mind, with computers that have to solve a much wider range of often more complex problems. Moreover, even in the more ‘simple’ goal, the computing power that Tesla envisions still has a long way to go.

But that Musk and co. pill by a fine piece of technology, everyone agrees. Dojo can even become extremely important to Tesla, according to some Tesla observers. Not only would it shorten the route to a fully self-driving car, it could also shape other plans Musk hatches. Just think of the humanoid robot that the Tesla boss will soon send after your purchases.

Leave a Comment