>Realize the savings to be had moving "fixed infra" back on-premises or into a colo versus sticking with a public cloud provider

As other people have pointed out: what happens when the PSU or mobo shits itself, what happens when the new version of vmware or docker (or whatever) shits itself, etc