The Ground
Beneath the Cloud

In Loudoun County, Virginia, the densest concentration of data centers on Earth sits on ground the national soil survey rates, politely, as a problem. Score every mapped soil unit in the county for what it takes to build a campus, and only 39 percent comes back as buildable. The largest single category is not good and not bad. It is marginal: 3,125 of the county's 8,024 soil map units, more than a third of the ground, sit in the middle band where a campus is possible but never cheap.

This is the part of the data-center boom that never makes the site-selection deck. Power gets a column. Water gets a column. Fiber, latency, and tax abatement get columns. The dirt, the thing that has to carry a quarter-million-square-foot slab and a yard full of buried copper, gets a geotechnical boring after the option is already signed. It does not have to work that way. The information has been public, and free, the whole time.

A national asset almost nobody reads

The national soil survey records, for nearly every acre in the country, the properties that decide whether ground is cheap or expensive to build on: how far down the seasonal water table sits, how much the clay swells and shrinks with the seasons, how fast buried steel and concrete corrode, how deep you can dig before rock, and whether the soil floods, ponds, or counts as a regulated wetland. Each is a line item on a construction budget. Together they tell you, before a single boring, whether a parcel will fight you.

The catch is that all of this lives behind a query language and a century of terms of art. So we built one score on top of it, the way the survey builds its own engineering interpretations, with fuzzy logic: every property is run through a membership curve that turns a raw number into a limitation between zero and one, the engineering ratings for buildings, roads, and excavations fold in, and the whole thing resolves to a single comprehensive index from 0 to 100. We call it GroundScore. For a data center it is the soil core of the GroundScore DCI, the Data Center Index that fuses it with grid, water, fiber, and land.

Where the buildable ground actually is

Run GroundScore across whole states and the ranking does not follow the headlines. The most forgiving ground sits in the Plains and the Southeast; the hardest sits in the wet, clay-rich glacial soils of the industrial Midwest and the valley floors of the Northwest. The map of where data centers actually clustered and the map of where the ground is easy are not the same map. They were drawn by power, fiber, and politics, and the soil was left for the contractors to sort out later.

Read the ranked list and the spread is stark. Toward the top, well over half the ground is ready to build. Toward the bottom, less than a third is, and the rest carries a wetness, shrink-swell, or flooding penalty that turns into mass grading, engineered fill, and dewatering before the first slab is poured.

The corrosion nobody budgets for

A hyperscale campus is, electrically, a city. It runs miles of buried conduit, a grounding grid of bare copper and steel, and concrete ductbanks carrying the feeders that keep the halls alive. Soil decides how fast all of that corrodes. The survey rates corrosion of steel and concrete on every component in the country, and the numbers are not comforting: a large share of U.S. ground is moderately to highly corrosive to steel, which is exactly the material doing the quiet, load-bearing, safety-critical work underground. On a high-corrosion site you pay for it twice, once in cathodic protection and coatings up front, and again in the shortened life of everything you buried.

Two sites, two verdicts

The model is at its most useful when it disagrees with the brochure. Here are two real soils, scored the same way, that tell opposite stories about the same building.

Dulles is not unusable. It is expensive. A 22 means a wet, shrink-swell clay with a shallow water table and a wetland signature, the kind of ground that demands mass grading, engineered fill, under-slab drainage, and a permitting conversation before anything moves. The Alley has been quietly paying that bill for fifteen years. It simply never named it. Cipriano, in the Arizona desert, is the opposite: deep, well drained, low shrink-swell, no water table to fight. Same building, a fraction of the ground risk.

Why AI makes the dirt matter more

The bigger and denser the campus, the more the ground costs. AI training halls are heavier, run hotter, and are more schedule-sensitive than the cloud build-outs that came before, and as the first-choice grids fill up they are landing in second- and third-choice geographies, exactly the places where soil risk is least understood and most likely to surprise a pro forma. The deals are getting larger and the margins for a ground surprise are getting thinner at the same time.

That is the case for reading the soil before the option closes instead of after the foundation cracks. GroundScore is the ground pillar; the GroundScore DCI sets it next to the grid, the water, the fiber, and the land, so a site that wins on power is not quietly losing on the one variable that cannot be value-engineered away: the earth it stands on.

GroundScore is computed from generalized soil survey map units (SSURGO) via Soil Data Access, attributed to the National Cooperative Soil Survey. Map unit boundaries are generalized; a site-specific geotechnical investigation is required before any commitment. State coverage expands as the national run completes.