AI image
The bull case for TSLA stock among many is solving full-self driving. The dream is cars that run on autopilot, like a taxi but without a driver. If it’s cheap enough, it would eliminate the need for many people to even have cars.
It’s certainly been a dream of Elon Musk for a decade. Unfortunately, he overpromised so many times that it’s become something of a meme.
“We will be showcasing our purpose-built robotaxi or Cybercab in August,” Elon Musk said in the latest earnings call. My suspicion is that this will just be a concept car with comfortable seats and no steering wheel.
But what if he actually delivers?
If so I’m sure TSLA stock would soar following the unveiling event, which is scheduled for August 8.
TSLA stock daily
I also think it would be a fleeting victory. The bull case for TSLA has been that mountains of data collection from real-world cars would help them to built a self-driving system via some kind of brute force method. I believe we reached the limits of that approach.
Instead, Tesla has pivoted to AI techniques.
Clues are in the latest earnings call, where Musk revealed this:
We also continue to expand our AI training capacity in Q1, more than doubling our training compute sequentially … We’ve installed and commissioned, meaning they’re actually working, 35,000 H100 computers or GPUs. GPU is a wrong word, they need a new word. I always feel like a wank when I say GPU because it’s not. GPU stands — G stands for graphics. Roughly 35,000 H100S are active, and we expect that to be probably 85,000 or thereabouts by the end of this year in training, just for training. We are making sure that we’re being as efficient as possible in our training. It’s not just about the number of H100s, but how efficiently they’re used.
This is a dead giveaway that Tesla is employing a new technique to solve self driving. I’m hopeful that it works (though doubtful about the long regulatory approval process).
The problem is that if Tesla
solves FSD with AI, they won’t be the last ones to do it. Everyone else will be right
behind. This technique is something that Tesla has just pivoted to, so other automakers would only need to acquire 85,000 H100 chips and use the same techniques, which are hardly a secret.
There’s no moat in being six months ahead in automotive technology.
Yes, there would be an explosion in $99/month FSD sales and prices might even
be ratcheted back up, but will you stay a loyal customer when Hyundai is
offering equal software at $19.99/month? If that’s the case, it won’t maintain
a more-durable long-term advantage than anti-lock breaks or all-wheel drive.
Yes, they have more data but with reinforcement learning, the size of the data set isn’t that important, it’s more about the quality. And other automakers are also collecting data. In an industry where the refresh cycle is +5 years, being first isn’t that valuable, certainly not valuable enough to justify a 10x higher multiple on TSLA stock.
So while I certainly wouldn’t fade any kind of breakthrough announcement, I doubt the enthusiasm lasts a year. Moreover, as Morgan Stanley writes today “China has already won the battle for global EV supremacy.”
However there is an area where I think Tesla could build a durable advantage that could drive the stock much higher. I’ll write more on that in a future post.