A 3 year depreciation cycle may be reasonable for some spaces but not others.

I'm not going to lose sleep over not getting firmware updates; when they've got all the bugs on my stuff ironed out I don't want more firmware updates.

I haven't done the spreadsheet work to identify the cross-over point for where old systems become uneconomical but when I've done it in the past for my heavily clustered workloads it's typically better to buy an off-lease server that's 3 years old and run it for another 8 years, than it is to forklift all my infrastructure every 3 years chasing the newest generation thing.

Certainly the cloud's got some advantages, but running equipment until it's really old is also pretty good too. I'd bet that AWS doesn't throw older systems out just because they're old.