Self-managed Talos K8s, self-managed CloudNativePG, and the operational overhead of networking, DNS, etc. All of these were used to be fully managed by AWS for them; zero operational cost.

I'd guesstimate a 2× increase in their operational complexity. So, if they previously required 0.5 DevOps of a full-timer, they'll now need one more DevOps full-timer just to handle the added complexity.

Does that make sense to you?