I have experience as a reliability engineer in infrastructure. I would say it's very context dependent, but for the most part it's cheaper to maintain up to a point. A lot of the contextual distinction comes with the criticality of the component. Critical components, by the definition of being critical, have a high downside of failure (whether that's cost of operational). I will say that there are some occasions where it may be cheaper (both in financial cost and availability) to run something to failure.

The real problem IMO is that cognitive biases can prevent people from making those objective decisions. When dealing with things like probability of failure where there's uncertainty, it's easy to be swayed by a competing priority that feels like it has more immediacy and less uncertainty.