However, Monday’s 10% surge added $70 billion, conjuring a value increase approximately the size of the BMW Group out of thin air. That kind of investor euphoria is typically restricted to the realm of artificial intelligence—this instance proves no different. 

Earlier, Morgan Stanley had published an über-bullish call on Tesla based almost solely on an analysis of the growth outlook for the EV maker selling its Dojo supercomputer as a service.

“The more we looked at Dojo, the more we realized the potential for underappreciated value in the stock,” Morgan Stanley wrote. The report upgraded Tesla to what is effectively a “buy” rating, naming it a top pick and predicting shares would return to their historic high of around $400 from November 2021 in 12 months’ time.

Dojo represents Tesla’s attempt to solve one of the biggest hardware problems facing AI— the bottlenecks in memory storage and bandwidth that inhibit effective scaling of the technology. This is particularly true when it comes to training on visual data.

Entering production in July, Dojo is supposed to rapidly expand to the equivalent of 300,000 Nvidia A100 GPUs, for a total compute of 100 exaflops (a measurement of supercomputer operational speed) by October of next year. 

Start of production for Dojo is JULY 2023. Within 6 weeks! https://t.co/6WEJr6znxs

— Bradford Ferguson (@bradsferguson) June 21, 2023

That would make it potentially the world’s most advanced machine in terms of processing power, and may help Musk realize his long-held dream of full self-driving (FSD). 

“Third-party Dojo services can offer investors the next leg of Tesla’s growth story,” the investment bank wrote.

At its heart is a Tesla proprietary “D1” chip, which features 50 billion transistors that—in theory—allow for a massive increase in the volume and speed at which data can move without a corresponding surge in energy costs.

Its streamlined design would allow Musk to feed vast amounts of visual information gathered by the millions of Tesla cars on the road to train his neural net. 

It could also aid AI-adjacent industries like robotics, health care, and security, according to Morgan Stanley. 

Not enough Nvidia chips to go around

First announced at Tesla’s inaugural AI Day two years ago, alongside its humanoid robot Optimus, Musk has since sought to temper expectations. In October he warned, “the jury is still out” on whether Dojo is worth the effort.

But he’s willing to risk a $1 billion investment anyway, since he cannot procure enough of Nvidia’s advanced A100 chips.

If it works, Tesla would likely be the first to market with self-driving software that can be flashed onto all the vehicles in its fleet via a remote update over the air. Musk would finally fulfill his promise that a million cars could overnight turn into robotaxis first predicted for the end of 2020.

Musk himself has sought to quell concerns from frustrated customers like Apple cofounder Steve Wozniak over self-driving claims. Last month he livestreamed a drive in the alpha build of FSD version 12 that supposedly features another root-and-branch overhaul of the software. 

This time, Tesla developers wanted to remove virtually all of the 300,000-plus lines of code in v11 and replace it with AI that can continuously learn and improve with each mile a Tesla car drives.

Musk has implied the release of v12—whenever that comes—would mean FSD no longer needs human supervision after nearly three years in beta testing. 

The fireworks from Monday’s $70 billion explosion in market cap did elicit controversy among veteran Tesla investors, however. 

Bulls criticized Morgan Stanley for taking so long to grasp the potential of Dojo, while bears pointed to a passage in the report that said analysts were never actually able to prove it does what Musk claims. 

“It is difficult to explicitly validate the many claims Tesla has made about Dojo’s cost and performance,” the report admitted.