A single metro area has 2,882 megawatts of data center capacity under construction right now, a figure that surged 52.3% year over year in the first half of 2026 and pushed it past Northern Virginia as the biggest construction pipeline in North America. That one metro's pipeline, on its own, dwarfs the total installed footprint of most national data center markets outside the top handful. The United States holds a disproportionate share of the world's data centers by facility count, and even that share understates things, because the measurement changes when you count the market in watts instead of buildings.
That's the thread worth pulling on. Facility counts flatter the rest of the world. Capacity, power commitments, and preleasing tell a sharper story about where the compute your software depends on actually runs.
Counting buildings makes the global map look lopsided but survivable. A country with a few dozen colocation sites and a couple of regional champions can point at its footprint and feel like it's on the board. The problem is that a colocation building in a secondary market and a hyperscale campus in a primary one are not the same unit.
One might draw ten megawatts. The other draws several hundred, and the newest ones are sized in gigawatts. A recent DEV.co podcast episode on why America's share of the world's data centers keeps growing walks through that distinction using the global rankings as its starting point, and it's a useful listen before the next architecture review that touches region selection.
Hyperscale campuses have gotten bigger faster than they've gotten more numerous. That compounding is the reason the U.S. share expands when you switch from counting sites to counting watts. A country can add buildings and still lose ground, because the buildings going up elsewhere are larger, denser, and attached to firm multi-hundred-megawatt power commitments.
Preleasing is the tell. North American vacancy has sat near record lows for years, and newly built capacity is being spoken for before the roof is on. The pattern holds this year too, with preleasing rates running well above the historical norm.
Capacity is not pooling. It's being allocated the moment it exists.
The reasons the concentration deepens rather than diffuses are not mysterious, and they don't balance each other out:
The region dropdown in a cloud console is, in practice, a shortlist of the places where power, fiber, and tax policy aligned. When a team picks "us-east" without thinking, they're inheriting that alignment. When they pick a region in a smaller market to be closer to users or to satisfy a data-residency rule, they're also inheriting the opposite: fewer availability zones, fewer instance types, longer lead times on GPU capacity, and prices that reflect scarcity.
Three consequences are worth naming out loud:
The capacity your application, your model inference, and your customers' sessions depend on is not evenly distributed across the globe. It's piling up in a handful of metros that already had the power, the fiber, and the incentives. The ranking that counts buildings hides this. The one that counts watts is the one your architecture is actually built on.
This website uses cookies.