I totally agree with your general scepticism regarding robotaxi before 2030. I definitely think Elons dream of everyone giving up car ownership for a flawless robotaxi servcie instead is not going to happen without major improvements to both the sensor suite and FSD computer.
But I do disagree that there will be any fast-follow.
If we agree that the path to true robotaxi is with a machine-learning large-dataset model, then there simply is NOBODY even approaching the level of data collection in the real world that tesla have, and continue to accumulate. The big auto firms are paying lipservice to self-driving but not even equipping their vehicles with the sensors and connections required to collect the data (even assuming they had any staff trained to work with it).
It will be a decade, minimum, between a large rival equipping its fleet with a sensor suite and OTT collection back to a central location, and them actually achieving robotaxi, even assuming infinite budget and compute.
The elephant in the room is the possibility that robotaxi is achieved by some AI breakthrough not dependent on large-0data collection, but that seems to get less likely each year. And even then, I suspect elon & Tesla would realize this before anybody else.
Because I see zero risk of fast follow, I see FSD as nothing but upside, with only a slight risk of refund liability if FSD cannot be delivered soon.
ring-fenced LIDAR based mapping silliness like waymo/cruise is an unscalable joke that can be easily ignored imho.
