Microsoft Azure

The second hyperscaler, and until now absent from a spoke whose whole thesis is a price/complexity spectrum anchored at aws. Written from Microsoft’s own reliability documentation (T3 — a vendor describing its own infrastructure, which is what coverage edge 5 asked for).

Over 70 regions, each a set of datacentres and networking inside one metropolitan area, and each contained in exactly one geography — the data-residency boundary (United States, Europe, …). Every geography has at least one region with availability zones. Some regions belong to sovereign cloud geographies, which “work the same way as other regions, but they often have limited services and features.”

Two resiliency constructs, and they are not equivalent:

• Availability zones — “independent sets of datacenters that contain isolated power, cooling, and network connections,” close enough for low latency, far enough to survive a storm or a local power failure.
• Region pairs — Microsoft pairs some regions for geo-replication and disaster recovery, and “you can’t choose your own pairs.” Notably, “many newer regions aren’t paired, and instead use availability zones as their primary means of redundancy.”

Some services are nonregional: deployed across two or more regions by Microsoft, so a regional failure leaves a healthy instance serving requests.

What this adds to the spoke’s argument

The comparison this wiki keeps making — hyperscaler versus Hetzner and the fixed-price VPS tier — has been running on aws alone. Azure’s documentation shows the same structure and therefore supports the same reading: what a hyperscaler sells above raw compute is failure-domain geography. Zones with independent power, curated cross-region pairings, a residency boundary you can name in a contract. That is the part a €5 VPS does not have and mostly does not claim to.

It also shows the model drifting: pairs were the original answer to regional failure and newer regions do not use them, which is Microsoft moving redundancy down a level, from region-to-region to zone-to-zone.

What it does not say: the price. This page is Microsoft’s reliability documentation, and the spoke ran on it for weeks without a single Azure figure, because the pricing pages render their tables in JavaScript. The numbers arrived on 2026-08-10 from the Retail Prices API instead: a 2-vCPU D2s v5 in eastus is $0.096/hour on demand and 60% less on a three-year reservation.

Not read here: Azure pricing, which is the other half of the comparison, and the service catalogue.