Which Edge Does a Data Center Move?

The map we built for Illinois with our colleagues divides the state into 30-meter cells, each about a fifth of an acre. Each cell knows what sits on it—crop, forest, wetland—how much carbon it stores or releases in a year, which wildlife depends on it, and what the land does for people, in dollars. We built it to help Illinois plan its natural and working lands as the climate changes. Lately people have been asking it something else: where should data centers go?

This year, county boards across Illinois have been taking up data center proposals. McLean County barred data centers from its agricultural zones. Bloomington and Normal paused every proposal for six months; Normal has since stretched its pause to May 2027. The state suspended its tax incentive while the legislature argued over who should be supplying the power. The land in question is usually farmland at the edge of a metro area, and the question before the board is whether to let the town grow one notch outward. Boards have real reasons to say yes—a data center brings investment and tax revenue.

That question leaves too much out. A data center is a decision as much as a building, and it moves several edges at once—most of them far from the fence line.

The ring we inherited

A century ago, sociologist Ernest Burgess proposed a model of Chicago as a series of concentric circles, with commercial development in the center and rural areas represented by anything beyond the outermost circle. Land-change models—including the one the Land-use Evolution and impact Assessment Modeling (LEAM) Laboratory has run at the University of Illinois since the late 1990s—worked from the same picture: a city that keeps spreading outward, one field at a time. In that view, a data center on a cornfield is just growth at the fringe.

A ring gives you one edge to argue about. The land keeps several, and they almost never line up.

What the layers show

To build our map of Illinois, we scored the land in each layer, marked the top 10 percent as that layer’s core, and put the cores on one map. They barely touch. Of the cells in either core, 7 percent belong to both the developed core and the strongest carbon sinks, mostly urban trees; 1 percent overlap with the biodiversity core. The strongest sinks share more ground: 17 percent with the biodiversity core and 44 percent with the economic value core, mainly in the Shawnee Hills and along the Mississippi and Illinois rivers. A ring drawn around Chicago would catch one of these cores—economic value—and only the edges of the others.

The layers don’t weigh the same, either. Added up, Illinois’ natural and working lands emit more greenhouse gas than they absorb—about 18.5 million metric tons of carbon dioxide equivalent a year: 38.8 million emitted, 20.3 million taken in. Cropland alone gives off 22.9 million more than it stores; forests do the opposite, soaking up a net 5.7 million.

So the cornfield at the edge of town already has a lot going on. It’s a carbon source; it may sit beside a wetland remnant or miles from one; and it supports the family that farms it. Build on it, and all of that changes at once.

Which edge does a data center move?

Land. A campus takes land, though not much of it by a county’s standards. The recent Illinois proposals run from 280 acres in Sangamon County to 795 acres in Joliet, and one projection puts all new data centers through 2035 on about 4,700 acres—0.02 percent of the state’s farmland. The decision reaches further than the fence, though: to the substation, the transmission lines, whatever gets built to feed the place power.

Carbon. Here the ring view gets the sign wrong. Paving an Illinois cornfield doesn’t destroy a carbon sink—the state’s cropland, taken as a whole, gives off more than it takes in. That doesn’t make paving harmless; construction has its own emissions, and any single field can differ from the statewide average. Still, most of a data center’s carbon comes from the electricity it runs on. The International Energy Agency put data center use at about 415 terawatt-hours worldwide in 2024 and expects roughly 945 by 2030. Whether that power comes from a coal plant kept on life support or a solar farm on the next field matters far more than the footprint does.

Water. Some operators cool their servers with a sealed loop of water used over and over—easy on the aquifer, hard on the power bill, and so, in the end, harder on carbon. Others use evaporative cooling, which sips less electricity but constantly draws on the aquifers and rivers that farms and towns depend on. Our map has no water layer, so water is the edge we see least; the newest version of the model begins to close that gap.

Habitat. Our habitat layer stands on about 827,000 records of where species have actually been seen. In the Metro-East work described below, how far a field sat from existing habitat helped pick which ones to restore; the same question should mark the places not to pour concrete.

People. In our survey, 71 percent of people from conservation groups put biodiversity first, against 23 percent of farm groups and 32 percent of government representatives. At a hearing, that split turns into different questions—one person about the creek, another about the lease payment, a third about the power bill. When the hearing is set up as a simple yes-or-no vote on growth, nobody ever weighs those questions against each other.

What the Metro-East taught us

On the opposite side of the Mississippi River from St. Louis, in Madison and St. Clair counties, Illinois, we scored about a million 30-meter cells of cropland on carbon storage, wildlife benefit, cost to farmers, and distance to existing habitat. In the model, we restored the best fields first. The top 109,779 acres close about 70 percent of the two counties’ yearly carbon gap of 351,500 metric tons of CO₂. Then the effort stops—not because the city is pressing in, but because a federal rule does: the Conservation Reserve Program, which pays farmers to take land out of crops, caps enrollment at 25 percent of each county’s cropland.

We expect data centers to behave the same way. What decides where a campus lands is rarely the land itself; it’s the queue to plug into the grid, the tax deal, the zoning code, the water permit. Find the rule first. The rule is usually what draws the edge.

The building that runs the map

There is one more edge, and it’s an uncomfortable one for us. The LEAM Laboratory has started building a system in which AI agents rebuild its workflow for a new place from local data, with a person signing off at each step. But every model we run draws its power from a data center somewhere. The chapter behind this post ends on a question we couldn’t crack: how should the energy, water, and land that AI uses show up in the accounts of the plans it helps make? A planning model ought to be able to see the building that runs it.

A pause is not a plan

A moratorium buys time; it settles nothing. The real question is what you do with the time. The LEAM Laboratory has spent more than twenty years simulating how land changes and what those changes do downstream. The newest version of its model adds water and soil moisture, and it’s built to travel: same framework, refit to local data wherever you take it. Before the fence goes up, it can trace how a data center would play out across the layers—where its power comes from, what its cooling drinks, which habitat it leans against, whose income moves.

We’re not neutral about how places should grow. We’d rather see growth that moves several edges the right way than growth that moves one edge fast. But a stance still needs evidence under it. A model like this doesn’t pick sites; it lays the trade-offs out where boards and residents can see them before they vote. Nobody has to choose between banning data centers and waving them through. We can decide with the layers in view.

Three habits before the fence goes up

We’d ask three things before any site plan.

First, say whose edge you’re drawing—carbon (through the power source, not the footprint), water, habitat, farm income, public trust.

Second, test the decision against the other layers. A campus on an emitting cornfield can move four edges the right way—pair it with renewables, cool it in a closed loop, put money into restoring the nearest creek. One that flattens a bottomland forest to sit next to a substation moves one and shrugs at the rest.

Third, find the rule. Rules move edges, and design can only push back once you know which rule is doing the moving.

Reading land as layers is what landscape architects train for, but anyone can ask the question. Next time a data center comes up for a vote, ask which edges it moves—and for whom. The cloud is landing in the countryside. It’s a good moment to stop drawing rings.

Further reading

Jiajia Wang, Siqi Zhu, Yexuan Gu, and Brian Deal, “Layers, Not Rings: Geospatial Modeling for Peri-Urban Transitions in a Polycrisis Era,” Routledge, forthcoming (source of the Illinois figures).

Interactive Illinois NWL+ maps (iESPSS).

International Energy Agency, Energy and AI (2025).

NPR Illinois, “Pritzker says state won’t trade farmland for data centers” (August 27, 2026).

KSL, “Closed-loop cooling systems save water but can be a drain on electricity” (May 12, 2026).