Running a service across multiple cloud regions for reliability and latency, the next tier above multi-AZ.
Multi-region is the architecture pattern of running a service across multiple cloud regions (us-east-1, eu-west-1, ap-southeast-2), as opposed to multi-AZ within one region. It guards against region-level outages (a tier above AZ failures), reduces latency for global users, and supports data residency requirements. The cost is significant: data replication, write coordination, deploy synchronization, and the meaningful jump in operational complexity from one region to many.
For most services, multi-AZ is enough. Multi-region is justified only by specific drivers: global low-latency UX, region-loss DR, regulatory data residency, or customer SLAs above 99.99%. Going multi-region without one of those drivers spends a quarter of engineering on coordination problems for very little reliability gain over a well-run multi-AZ deployment.
See the part of the platform that handles multi-region in production.