Reliability hasn't changed significantly for server-grade kit for decades.
All Intel-compatible servers have been essentially perfectly compatible with each other for decades.
All major brands will keep selling server kit. It's not like you have to throw away your IBM brand Intel-based servers and replace them with ARM chips from some no-name vendor!
My point is that in next year and the year after, and the year after that IBM will sell you faster and faster servers. Also... bigger and bigger. At roughly the same price point. With largely (if not entirely) compatible management systems, drivers, etc...
Speaking of drivers: VMware and similar hypervisors have almost completely eliminated driver compatibility as a concern. Clusters can contain wildly different gear without issues.
There's a lack of understanding in the industry by older managers that gained their experience in the "before times" when even the firmware had to be consistent for a cluster of servers.
For example, the local budget airline Jetstar would buy used servers from Ebay.
Why?
Why not!
Why would you need "support" if a server dies? It is stateless (diskless!) and just 1% of your capacity! Just throw it out.
Why would you care about driver updates if all of your VMs are using emulated hypervisor drivers only?
Why would you care if a server is "used" if chips don't "wear out"?
It's not arbitrage because of the labor spent looking through ebay listings and dealing with surprises (broken hardware, incorrect listings etc.)
Also the machines have to go somewhere and that costs a lot because if someone can do it cheaper they'll just sell VPSs for cheaper than you can host them. Doing it cheaper means more compute per rack/watt which means newer hardware.
