What I'm saying is that Meta will create a platform on which Fediverse instances can be hosted. They'll add features to that platform that make it easier and easier to host such an instance. They'll offer APIs or whatever that'll support instances in ways that other hosting environments won't. And then when the code base has changed to depend on their particular hosting environment, they'll use the power that gives them over us.

Glad you brought up AWS. Amazon and other tech companies have created kubernetes platforms (Amazon's is EKS, which runs on top of AWS and its services like EC2) that make it really easy to spin up clusters, auto-scale them, use custom objects that are specific to their platforms, control external access to them, monitor them, etc. While "bare metal" kubernetes implementations exist, they are a royal pain in the ass to setup and run, and they support a fraction of those bells and whistles. And as time goes on, the difference between one of these "native cloud environments" and anything anyone would (try to) setup themselves gets greater and greater. And systems that are developed for kubernetes rely more and more on those bells and whistles (e.g. despite kubernetes being allegedly agnostic to what it is running on underneath, companies choose to support "just EKS" or "EKS and GKS" and no other environments). Perhaps a particular software suite depends on PersistentVolumes that can be moved between nodes, or mounted on multiple nodes simultaneously, or whatever. EKS might support this when other environments don't. Or a custom AWS annotation on a LoadBalancer Service might provide some kind of control that the software depends on to function properly and be externally accessible in a way that the software depends on. So this is a nice corollary to how things might go with the Fediverse.