Half of the "robust" and highly complex solutions I see employing modern practices go down way more often than the simple version that the GP described.
Probably I don't know enough to argue, but those are the repeated arguments and which I agree with. In my current place, I know we can reduce the infra bill significantly if we move to renting physical servers or VPSs, but nobody would like to gamble on the "big words" (availability,..). We already have a system to deal with downtime in one region but it will take much more efforts that would lead to a minicloud if we want to go with the dedicated hosting.