We're missing a part of the case though: why do you need to be a car-maker to be the vanguard for self-driving taxis?
The best case scenario for a self-driving company would be to target software and sensor solution packages that they can sell or license to other manufacturers. Such a vendor can focus on the self-driving problem and not have to bother with things like "we found a surprisingly big market niche for a 11-passenger minibus, but no platform for it" or "to sell it in the EU we need the headlights to be 5cm lower". I'd expect the margins are also a hell of a lot higher if they don't have to include two tonnes of steel with each auto-driver license they sell.
Maybe they build a small number of test mules, or just chop-shop a few off-the-shelf cars as a R&D fleet, but they hardly need to be a seven-figures-per-year manufacturer to be supplying those needs.
That's even assuming they come out green in the competition to deliver robotaxis. Right now the leading player in the US market is a company who is neither Tesla nor a legacy vehicle manufacturer. It's an adtech who started gluing the contents of a Radio Shack onto the worst cars you could possibly think of (Chrysler Pacificas and Jaguar i-Paces? Really?) and turned it into something that's an everyday thing in several major cities.
Tesla FSD story reminds me of the fracas that was early OS/2. IBM sold people 286 hardware on the promise of it running OS/2, so they had to waste a lot of effort building a 286-capable OS/2 that was clunky and almost immediately obsolete. No matter how talented Tesla's R&D team are, they're walled in by design choices made on existing vehicles (i. e. relying on cameras instead of lidar). I wonder if they'd be better off being ran as an arm's length startup to address the problem more generically, and then they can sell it to other firms if it turns out that the best solution won't work on existing Tesla hardware.