We have one: we run thousands of VMs at a time, all accessing the same "database". Since we already have a good amount of horizontally-scaled compute, having to maintain a separate database cluster, or large vertically-scaled database instance, to match our peak load requirements is problematic in terms of one or more of cost, complexity, and performance. In particular, horizontally-scaled distributed databases tend not to scale up and down efficiently, because of the complexity and time involved in joining the cluster, so the cost benefits of horizontal scaling of compute are lost.
An approach like this can fit well in cases like these.