You’d think they wouldn’t have started with GCP. There are plenty of datacenters where you can buy racks and racks of servers, and talk to a human when something goes wrong, and even walk in and access your servers. That’s what I’d be using if I were to build a Rackspace today.
Colo is worse than cloud when you're getting started. Sure, you can talk to a person but everything else is much lower quality. People are obsessed with having someone to yell at but yelling does not fix outages.
They started on GCP and have been migrating to their own "Metal" DC doing exactly what you're describing. But GCP is still their overflow given how rapidly they are growing and holds some amount of networking that routes to their DC.