Azure availability zones are physically separate datacenters in a region with independent power, cooling, and networking. Paired regions are two regions hundreds of miles apart that Microsoft uses for sequenced platform updates and some native replication (geo-redundant storage). Putting three VMs in one region without zones is not HA. Putting everything in three zones and assuming a region-wide flood is covered is also not HA.
Zone-redundant versus zonal
A zonal resource pins to zone 1. If that building goes dark, the resource is gone. A zone-redundant resource (ZRS storage, zone-redundant VMSS, zone-redundant public IP, some database SKUs) replicates across zones. You pay latency and cost; you buy a datacenter failure. Check each SKU: "available in zones" ≠ "redundant across zones." A VM in zone 2 with a disk that is zonal in zone 2 is one fate.
Region West Europe
Zone 1 Zone 2 Zone 3 ← metro-scale isolation
Paired: West Europe ↔ North Europe ← hundreds of miles, GRS/GRS-styleLoad balancers need a zonal or zone-redundant frontend. A standard LB with zonal mistakes can black-hole after a zone loss. Kubernetes on AKS has zonal node pools; etcd and control plane SKUs have their own redundancy story — read the AKS SLA page, not a blog.
Paired regions are DR, not a third zone
Region pairs help with Microsoft-initiated recovery sequencing and with GRS storage replication to a specific mate. They do not replicate your AKS cluster by magic. You still design: async DB replica, DNS failover, stateless compute in both regions, RPO/RTO that match async replication. Some services offer native geo (Cosmos DB, SQL geo-replication); some do not.
Paired-region platform updates are staggered. That is not a reason to skip zones inside the primary. Most "Azure was down" customer incidents are still a single zone or a single-stamp misconfiguration.
How to choose
User latency: stay in one region with three zones if the SLO is "survive a building." Regulatory and disaster: second region, with an RPO you can say aloud. Cost: ZRS and dual-run compute are the line items. Test failover; unpaired "we have a pair" is a wiki.
Read Microsoft's availability zone and cross-region replication pages for the SKUs you actually buy. Then draw one diagram with a zone fire and one with a region loss. If both diagrams end on the same box, you did not design two budgets. You designed a slogan.