Blades in huge datacenters are not serviced one by one. Typically you'd wait until say 20% of the blades fail and only then fix and upgrade the whole rack at once.