I have read that Nvidia’s next chip, Blackwell, will reduce the time for learning Autonomous driving to about 6 months and they can use simulation of driving to feed the system vs. all the Billions of miles Tesla has with FSD. Meaning that the moat Tesla has vs. Any other auto company on autonomous driving is no longer 5-6 years.
And, Tesla even mentioned that if they have a client that wants to license their autonomous driving technology, it would take 3+ years to bring the car to market.
So, if we are all thinking that TSLA stock will jump because they are close or at autonomous Driving later this year, that, any auto company that buys the new Blackwell chip will be close behind and thus, TSLA stock will not have the moat we all thought?
Your thoughts?
Having the compute to learn a model is arguably the most trivial piece of all of the stuff that needs to come together to learn a functional and useful model (such as autonomous driving). Existence or non-existence of the Blackwell chip only meaningfully changes the size of the moat relative to any competitor that has all of the elements in place for learning a model to perform autonomous driving.
Using simulation to feed the system is an important contributor to the set of data being used to train a system. It is not a replacement for real data in an open ended system such as driving. Pure simulation works for games such as Go or Chess in order to learn valid moves that improves one's odds of winning the game. Simulation does not work for predicting somebody sweeping the board clear in a fit of rage.
The moat that a Tesla, Waymo, MobileEye, Cruise, etc.. is not just a function of how many Nvidia chips they own. Its also a function of the data they've gathered, the ongoing data they are gathering (data workflow), their testing and regression testing regimen, the ability to train and re-train the model (this is where the chips come in, enabling faster training times), efficiency of the software, engagement with regulators and first responders, and all of the myriad pile of stuff that needs to come together to have an autonomous driving solution.
Heck - software developers can easily develop inefficient software that will overwhelm the fastest hardware and make it seem slow. Its really the software that is slow. On the other end of the spectrum really clever software developers can make bad hardware seem fast though it takes a lot more work.
Two companies with equally robust and expensive systems for development and training of their autonomous driving models, then yes - the company with the newer chips will make faster progress than the company with slower chips because they'll be able to do more iterations training the model, and thereby improve it faster. And that's it.