I suspect it would be maintained at the same quantization that it's delivered: pods. Pod has some unresolvable issue (things would have to go real bad), then eject it.

What maintenance do you have in mind, specifically, in a modular system like this would necessarily be?

If the argument is that off the shelf systems, designed within a context where humans are a Jira ticket away, won't work well in space, then I can agree with that. But, I don't think anyone is attempting that. Datacenter in space doesn't necessarily mean "standard rack computer in space". Keep in mind that redundant and self-correcting systems already exist in modern GPUs, with active rerouting around dead cores, detected with a BIST, at boot! These kinds of things are standard in financial systems.