Why does the cheapest cloud region on your provider's dropdown almost always sit in the same handful of places? That dropdown isn't a menu of geographies. It's a map of where the electricity, the fiber, and the tax breaks landed first, and almost everything else about your bill, your latency, and your compliance paperwork flows from that.
Hold one idea in your head while you read the rest: the United States hosts a disproportionate share of the world's data centers, and that share climbs further once you weight by capacity instead of counting sites. That single fact shapes the architecture decisions on your desk this quarter.
The Dropdown Is a Power Map, Not a Globe
A recent episode of the DEV.co podcast argues that the U.S. has more data centers than the next 14 countries combined, and the imbalance only widens when you measure by hyperscale capacity rather than facility count. The reason is unromantic. Builders follow cheap, reliable power to flat, buildable land near dense fiber routes, and they pick jurisdictions that will permit the thing in under a decade.
That concentration isn't spread evenly across the country either. An NYU Tandon analysis found five US metros host nearly a third of the national footprint, with the Washington–Arlington–Alexandria corridor alone holding 610 sites. When your provider offers you us-east-1, you are, in practice, buying a slice of that corridor.
That Concentration Sets Your Latency Floor
Latency is a physics problem bounded by a real-estate problem. Light in fiber covers roughly 200 kilometers per millisecond, so the one-way distance from your user to the nearest region sets a floor no amount of caching erases. If the dense regions sit in Northern Virginia, central Ohio, Oregon, Dublin, and Frankfurt, your users near those hubs get single-digit millisecond round trips and everyone else pays a tax measured in tens of milliseconds.
The concentration also shapes what "multi-region" buys you. Two regions in the same metro share fate during a grid event. Two regions on different continents introduce replication lag your application may not tolerate. The honest planning question isn't "which regions are available" but "which pair of regions is far enough apart to be independent and close enough to your users to be usable."
The Same Concentration Shows Up on Your Invoice
Price follows the map too. Published list pricing for general-purpose VMs typically runs several percentage points higher in European regions than in the densest US hubs, and the markup tends to hold steady across instance sizes. That isn't a rounding error at scale. It's a durable tax on running workloads outside the densest hubs, and it compounds over a three-year reservation.
Egress is the second half of the same story. Moving data between regions is cheap along the well-worn paths between the big US hubs and expensive along the sparse ones. Pulling data out to the public internet costs more again. If your architecture assumes data lives in one region and serves users from everywhere without saying so out loud, the invoice will eventually say it for you.
Residency Is Where the Map Gets Political
The concentration creates a second problem that no amount of region selection fully solves. A European region operated by a US-headquartered provider still sits inside US legal reach under the CLOUD Act, which can compel disclosure of data the provider controls regardless of where the bits are stored. Picking Frankfurt over Virginia helps with latency and local regulators. Sovereignty is a separate question.
Microsoft's own region selection guidance walks through the same tradeoff: map each data type to the laws that apply, confirm the certifications the region actually holds, and check latency against where your users really sit. Residency is three overlapping questions — where the data lives, who can compel its production, and what certifications the region carries — and the answers rarely line up on a single pin.
The Region Picker Is an Architecture Decision
The dropdown looks like a convenience. It's the first architecture decision of the project, and it constrains the next ten without ever announcing itself. Give it the same care you'd give a database choice, read the map behind it, and the rest of the stack gets easier to defend when the bill, the regulator, or the user on the other side of an ocean finally asks.