When a major cloud provider has a regional outage, the aftermath follows a familiar pattern: a wave of postmortems from companies that assumed they were protected, and weren't.
Most of them had done real work. They'd deployed across multiple Availability Zones, which is genuinely good practice — it protects against the failure of a single data center inside a region. What it doesn't protect against is the region itself having a bad day.
That distinction rarely matters until it does. Multi-AZ is about resilience within a region: if one data center goes down, traffic shifts to another one nearby, usually in milliseconds. Multi-region is about what happens when the entire region — every AZ in it — becomes unavailable at once. It's rarer, but when it happens, only a multi-region architecture keeps you running.
Building for multi-region isn't automatic, and it isn't free. It means replicating data across regions with a defined RPO, deciding how failover actually gets triggered, and testing that failover under real conditions — not just diagramming it. A disaster recovery plan that has never been executed is a hypothesis, not a plan.
The question worth asking isn't whether your architecture uses multiple Availability Zones. Most modern deployments do, by default. The question is what happens to your business if the entire region goes down — and whether you've actually tested the answer, or just assumed it.
More on this: Disaster recovery