Thought experiment: Let's say you are simulating ten thousand satellites on your computer, and the simulation runs until there is a crash. Now let's say the simulation runs for an hour normally. If you increase the speed of the simulation, you get to a crash in a shorter time. Satellites move about 30x the speed of airliners. Hence, if everything else was similar, one would expect 30x the amount of collisions.

Orbits are predictable, but they intersect and decay [at different rates] and occasionally get perturbed by space weather. This already needs periodic conjunction avoidance manoeuvres, and whilst orbits are fast satellite manoeuvres are slow, so the notice you need to avoid a conjunction is measured in hours rather than seconds. Can't imagine a scenario in which it would be sustainable for LEO to even approach the density of commercial aviation, except perhaps for a hypothetical where a single entity actually managed all the satellites.

The other underestimated dimension is that satellite manoeuvres use up a finite supply of expensively-launched propellant. That's tolerable when Starlink is doing 50k conjunction avoidance manoeuvres in six months across its constellation, but once it becomes each satellite moving at least weekly you either need bigger satellites carrying more propellant or have to accept significantly higher collision risk than they currently do.