You need to buy the hardware. However, you don't really need a dude on-hand to swap stuff on a daily basis, unless you're trying to host backblaze. The approach we take (with our data center 1000 miles away) is to provision excess machines. So if we need 6 machines we'll provision, say 8. Failure modes are always assumed to be "the whole machine" -- so a machine is either in service or not. Over time (years) one or two machines might fail in one way or another. Every couple of years we mount a rescue mission to repair/replace the bad machines, do some upgrades etc. We have redundant switches and routers and would make a special trip to replace one of those if there were a failure. The entire deployment has a "scaled to zero" cloud hot standby in place for the eventuality that the whole setup gets nuked somehow.