I think this underestimates how users will naturally gravitate towards more centralized instances, or they’ll give up because the bigger instances are closed.
(This is purely my personal opinion, of course!) In the scenario in which a few large instances dominate, the idea of the fediverse failed. One may estimate the likelyhood of success or failure given how they expect humans to behave, but in the end experiment beats theory. I think that for the fediverse to work a significant cultural shift has to occur, but I don't think that it is an impossible shift. I would like the fediverse to succeed, and so I choose to take part in the experiment.
This also ignores that the system isn’t horizontally scalable at all, so scaling up gets even more expensive
Yes, that might cause some serious issues. The project is still in an early-development phase, and I don't understand the technical aspects well enough yet to be able to identify whether there is obviously a fundamentally invincible barrier when it comes to scalability. My optimistic hope is that the developers are able to optimize horizontal scalability fast enough to meet the demand for scale. If it turns out to be impossible to scale, then only rich enough parties would be able to have viable instances, and that could be a reason for failure.