You could (and should) run the various instances of your Alertmanager
cluster in different clusters, if you run in multiple clusters.
A Prometheus server is usually set up to monitor local targets. So you
would run both members of an HA pair in the same cluster. And then run
at least one pair in each cluster.