if you think of something like server failure, one of my clients just had one.
a shiny proliant server with everything possible highgrade parts.
6 month old and it had 2 HD failures.

HP dude came and changed the drives with new ones and of he goes.
then, at the time of my arrival there is this support dude trying to get the system back from backups and he leaves just before me taking the server with him saying, there was something wrong with the tapes

I'm glad I wasn't in his shoes

anyway, the lesson is, hardware and it's support and it's costs are nothing compared to other costs of disaster failures.
this ofcourse is only true with smaller companies.
with thousand servers you gotta have some clustering going on, so taking one server does not hurt.
and if it does, you gotta redesign the clustering model.
imho.