Comment by vidarh

The point is that past burn-in, the failure rates are low enough for years that they're a rounding error and you can plan for just letting the failed equipment sit there.

Allowing the failed equipment to sit there can in fact cut costs because it allows you to design the space without consideration of humans needing to be able to access and insert/remove servers.

The higher the cost of bringing someone in to do maintenance, the more likely it is you will just design for redundancy of the core systems (cooling, power, networking), and accept failures and just disable failed equipment.