Agreed. Scaling application servers is "easy" since they can be scaled horizontally; in other words, you can just throw more hardware at the problem. Scaling databases is "hard" because of CAP/PACELC-type issues in maintaining consistency, and because some DBs (like Postgres) don't have good support for horizontal scaling.
I’ve found DDD’s insistence that cross-aggregate aggregation be done through IDs to be helpful here. It makes transitioning to other persistence implementations easier.
