A multi-region architecture runs an application’s compute and data across two or more geographic regions, so users hit a nearby region and the service survives a whole-region outage. It is the highest tier of availability and of complexity.
Worked example: an active-active setup serves traffic from US and EU regions at once, each with a local database replica, with global routing (latency-based DNS or anycast) sending users to the closest healthy region. Gotcha: the hard part is data — active-active means concurrent writes in different regions that must be reconciled (conflict resolution, or region-partitioned data), so most teams start with active-passive (one region serves, another stands by) because true multi-master is a distributed-systems minefield.