Many applications don't need multiple physical machines worth of database.

Persistent storage has been a solved problem, especially when using a managed Kubernetes service. What's the problem with running stateful workloads on a cluster?