Kubernetes is a tax on operators, and most teams are paying it without ever seeing the return. The control plane adds complexity, and every component you run is another attack surface that needs patching and locking down. If you're not at serious scale with real multi-team needs, a plain VM or a managed container service gets you 80% of the value for 20% of the pain. The mainstream push to "just use K8s" ignores that the failure mode is often the ops team burning out, not the app crashing. Run it when you must, not because it's trendy.