Earlier quoted context omitted.
Colo is certainly a cheap way to do things for a steady load, and you get great performance for the money, but who wants to get paged at 4am? I know HN and StackOverflow do things this way but I don't even want to manage the OS these days. We used to host our servers on-site and it was a massive pain. At least with colo you don't need to personally worry about power or internet going down, but you still need to repla…
Curious as to why you think aws means you won't get paged at 4-am? If you have a redundant setup, you won't get paged as everything can just fail-over.
Yes, you could do this yourself but that would cost more and reduce the advantage. How many colo racks in how many data centres / countries do want to have to visit? Even with redundant hot spare fail-over you will need to fix the hardware at some point and will be running at risk until you do.
A hybrid approach can work, with colo for the base load and cloud for scaling spikes / redundancy.