I wonder if it ever collapses. It's probably unlikely, the analogy of layers that strictly build upon each other might not be the best, but on every iteration where we hide complexity, some knowledge from a few layers down is lost, and we might do one thing more efficient now, but another thing in a more convoluted way, without even realizing. So it's hard to say whether it was worth the tradeoff. I guess everyone working in tech for long enough has some pet peeve about where something was way easier 20 years ago, or stories about junior devs frequently not knowing X and doing Y in a really dumb way, and are super surprised when they learn X and then get told it had been around for decades. It's obviously a net gain after all, but if you were able to look at it from a distance with total knowledge, it's also hilariously stupid. But then that's exactly how evolution works, and why there's a nerve going from your brain to your throat taking a detour around the aorta.
When I think about the "tower of abstractions" collapsing I usually end up coming back to the thought that it won't be a total collapse. It'll be a collapse to or around a specific layer of abstraction that we need to rework from (eg: something significant changed about a previously safe assumption). Like the 1906 earthquake in SF, it knocked down most buildings that had architecture that could not support the stress of an earth quake.
