The Power of Dojo: How Tesla’s Supercomputer Could Boost Its Market Value by $500 Billion
Image from https://sourceability.com/
On September 11, 2023, the stock market received a powerful boost, like King Kong drinking high doses of caffeine, when Morgan Stanley analysts said Tesla’s Dojo supercomputer could fuel a $500 billion jump in the electric vehicle maker’s market value by helping to accelerate its foray into robo-taxis and software services. The news flash skyrocketed Tesla’s share value, which rose by more than 9% in a few days.
Deadlines and planned dates for Tesla’s Dojo supercomputer
Tesla, which is already the most valuable automaker in the world, began production of the supercomputer in July to train artificial intelligence (AI) models for autonomous cars and plans to spend more than $1 billion on Dojo until it is launched in the first quarter of next year. The company expects that the upcoming version of its Full Self Driving system, FSD, will be available by the end of that year.
For more details about how FSD works, visit the link: https://www.tesla.com/support/autopilot
The computing power of Tesla’s Dojo super-computer
D1 chip of Tesla's Dojo Super-computer
Dojo is a supercomputer that uses machine learning models to train AI models applied to autonomous driving. It is capable of processing up to an exaflop (10^18) of floating point operations per second or 1000 petaflops, which would make it the fastest AI training computer ever developed while maintaining power efficiency and a small format compared to most other supercomputers.
The Dojo D1 chip uses 7-nanometer technology and delivers breakthrough computing performance and bandwidth. The D1 chip is the second chip designed by Tesla itself and follows the FSD chip that is present in the FSD 3 computer hardware in Tesla vehicles. According to Ganesh Venkataramanan, Senior Director at Tesla and Dojo Project Leader, “The D1 chip was completely designed by the Tesla team in-house. From architecture to package. This chip is like a GPU-level computer with flexibility at the CPU level and double the I/O bandwidth at the network chip level.” The grouping of these chips into training tiles in turn makes up the computing clusters. According to the company, 2 x 3 tiles can be combined in one tray and two trays in a computer cabinet, which would result in more than 100 petaflops per cabinet. Because of the massive bandwidth, Tesla says it can link all of these together to create the HexaPod, which will break the exaflop barrier of computers in a 10-cabinet system. This has been one of the main challenges in the development of powerful supercomputers.
Tesla’s Dojo Supercomputer: A Public Project but also a Secret Weapon
Dojo is the key to allowing cars made by Tesla to drive themselves safely. According to Morgan Stanley’s analysis, we have already seen this story in the past. How? Well, with Amazon.
The example of Amazon is not trivial, since the e-commerce giant began to diversify its offer by becoming, beyond a conglomerate focused on entertainment with movies or series, a technology company that offers services. That’s where AWS (Amazon Web Service) comes in, the online and cloud support system on which millions of websites around the world depend. Financial companies like Morgan Stanley point out that this is precisely the key to success. “If Tesla is successful, it can open up new markets to target,” they comment. Cars of tomorrow trained with proprietary AI without having to be manufactured by Tesla? The business model is beginning to reveal itself, no matter that it smacks of premeditated monopoly.
Dojo supercomputer will mean a quantum leap in the AI race for full self-driving
Dojo is designed to handle massive amounts of data in driving training systems, and that’s a huge advantage over other manufacturers. “It will give an asymmetric advantage in a market with a potential value of 10 trillion dollars,” explained Adam Jonas, Morgan Stanley’s analyst, "what could make software and services the biggest driver of value for Tesla going forward".
This supercomputing architecture collects a huge amount of data from the manufacturer’s entire existing car fleet and develops the best vehicle responses. Of course, we are talking about machine learning, in which AI knows what to do based on a lot of inputs and recorded data and begins to detect patterns, responses, and hundreds of millions of possible scenarios for each driving situation. Even though Tesla’s Autopilot system and their Full Self-driving capability have currently well-known naysayers coming from the tech field, this time things seem to be different, and maybe soon we will see how Tesla cars drive us to our destinations with the same success with which SpaceX spaceships join to the ISS.
