Their own experiments don't support their own conclusions.

Once you have AKS/GKE/EKS all configured with cert-manager, external-dns and the same Ingress or Gateway provider then you -do- have full portability. I.e you could go back and forth between them without needing to rewrite manifests (except for FQDNs or whatever that you are having external-dns pick up).

The point is the "cloud provider native" container systems just aren't standard, never will be standard and have no incentive to be standard. Their entire purpose is lock-in first, maybe differentiated features second (though I am not aware of any differentiated feature of a "native" system that I would care about).

Everyone focuses far too much on "Kubernetes is more complexity than I need" angle. The effort invested to stand it up is shown to be really low here (2 days is nothing in the grand scheme of any application that will ever make money) and the costs to maintain it will be similarly low while providing a solid basis to expand, migrate, etc. The concepts that need to be understood are roughly analogous except again, the knowledge of the specifics is portable.

Personally speaking I am never going to understand why there is still so much anti-Kubernetes crap out there. It's a small investment with massive pay offs in flexibility, consistency, portability and most importantly of all better state mangement.

Which is something completely ignored here and pretty much everywhere, managing state on "native" solutions sucks. Terraform and Pulumi aren't just bad, they are -horrible-. Just because they are the least-worst tools to manage those systems doesn't make them good and anything you can do to avoid having to interact with them is worth your time.

The problem with multi-cloud is that the dragons lie not in the functionality--that's well defined by the APIs--but in the behavior.