Imo It seems like FSD is severely underestimated by the market right now. From the latest reviews and tweets it seems that 11.3.x was massive improvement and it sounds like 11.4.x also will be a massive improvement to the point that many people are reporting majority zero intervention drives and not only in California. Mainly it proves the point that Tesla has a method of improving the system rapidly that is not hitting any walls yet. So 11.5 etc will likely keep bringing improvements to the point that it's getting hard to laugh at FSD anymore and the talk will quickly go from "Robotaxi will never work" to "Robotaxi easy to do, competition is coming to eat Tesla's lunch" lol.
Tesla has said that they intend to 10x their NN compute in 2023 and another 10x in 2024 And it's not like they were starting from a small capacity, they had one of the largest Nvidia clusters in the world in 2022. In 2024 they will have a the equivalent of 100 of them. How can this be ignored by the market?
It sounds like Europe might get FSD around Q1 2024 and they taking steps at expanding it to China. And Europe will not start from US2021 FSD, no it will pretty much be as good as US 2024 FSD after a few months. Even though they don't drive with FSD in Europe, you can bet they are running it in shadow mode collecting would been failures. The first quarter it's enable in Europe will be a rather large pure profit post as decent percent of their European fleet will enable it at once.
Still market seems to give FSD a value of ~0. Imo there are different scenarios with different estimated likelihoods and values. Some of the higher value ones such as robotaxi we can disagree on, but just the expensive addon "autopilot on city streets" is starting to look very likely and should be plenty valuable in itself. Let's say this can add another 5% margins(15k option, 15% take rate, 50k ASP ->~5% margins), imo that's pretty huge. 30% take rate and it's 10% margins. If not $15k then monthly and monthly is even higher margins but requires Wall Street to think so it will take longer to show in the valuation but the value is still there. This seems like a decent floor for the value of FSD, not to mention insurance gains etc. Robotaxi may or may not be far away, but even ignoring robotaxi FSD should be included in the calculation.
Then there is the question of the value of Dojo as a Service. If Dojo is indeed 6x cheaper than A100, Tesla can charge 50% of the price of A100 and still have 50%+ margins. And the market for this is clearly very large. Twitter and X.ai seems pretty locked in to be customers and Tesla can divert some of their nVidia expenses to Dojo expenses which in the end is the same as being a customer. And the market is starved for compute right now, like Elon said startups have to buy the chips from back alley dealers. Dojo as a service is not completely unlike AWS that's carrying Amazon right now and not completely unlike nVidia which has a higher value that Tesla. And AI seems to take over more and more so clearly this market is not negligible as Wall Street seems to think.