On the other hand, perhaps the large cloud providers bring a level of complexity that outweighs their skills at keeping everything up. What I mean is, a basic redundancy and failover setup with two data centers is kind of straightforward. Sure you need a person on call 24/7 to oversee it, but it's conceptually not that complicated. And if you're running bare metal, you get a surprising level of performance per dollar and rack unit. On the other hand, the big clouds are immensely complex with multiple layers of software defined networking, millions of tenants, thousands of employees, acres of floor space, org charts, etc. If you're running your own infra as one competent sysadmin, you know nobody else in another department will push a networking code change that will break your shit in the middle of the night. Maybe it's not right for everyone, but it's not unreasonable to go on prem in 2021 despite the popular opinions otherwise. Source: my company runs on prem and routinely has 100% uptime years. Most unplanned downtime occurs early on a Sunday morning following a planned action during a maintenance window.
I was and continue to be surprised how reliable even old servers are. I run a small homelab (Debian VMs on Proxmox; a Docker host, a jumpbox, a NAS running ZFS, etc.) on seven year old hardware, and all of my problems are self-imposed. If I leave everything alone, it just works.
As a counterpoint, though, my last place had a large Java app, split between colo'd metal and AWS. Seemed like the colo'd stuff failed more (bad RAM mostly, a few CPUs, and an occasional PSU). Entirely anecdotal.
