Alternative take: I was at Google when they killed Reader. At the time they pulled the plug, Reader had already been destaffed for years and was kept going in one guy's 20% time. The infrastructure effectively ran itself - the underlying Google datacenter teams replaced machines when they died, everything else was automated.

My experience there was pretty consistent with that: if you leave things alone, they won't break. Virtually all breakage was caused by changing things. On a well automated system repairing the physical components, rolling OS updates from time to time and so on, is not a huge amount of work.