When Kevin O’Leary — “Mr. Wonderful” of Shark Tank fame — sat down with Tucker Carlson last week to defend his proposed Stratos data center in Utah’s Box Elder County, something remarkable happened. Carlson, one of the most-watched conservative voices in America, methodically dismantled the project piece by piece. Why, Carlson asked, should Utah taxpayers subsidize a private facility whose tenants include some of the richest corporations in the world? How is it capitalism when three county commissioners — not the people of Utah — approve a project twice the size of Manhattan? O’Leary’s answer — “Welcome to America, buddy” — landed with a thud. Oh, and he is Canadian.
The same week, at the University of Central Florida, a commencement speaker told a gymnasium full of arts and humanities graduates that artificial intelligence is “the next Industrial Revolution.” They booed her off the stage. At another ceremony, former Google CEO Eric Schmidt told graduates “you will help shape artificial intelligence” — and was met with the same chorus of jeers.
The air is coming out of the tires. The AI data center boom — celebrated in boardrooms and congressional testimony alike — is generating a mounting bipartisan backlash from Tucker Carlson’s audience to graduating students to Utah farm owners who don’t want a 40,000-acre industrial campus consuming twice their state’s entire electricity supply on their doorstep. The question worth asking is not whether to build the infrastructure America’s digital future requires. It is whether developers have the wisdom to build it in the right places.
Most don’t. And the consequences of that failure are beginning to compound.
The Wrong Places
The data center industry has a fundamentals problem disguised as a land problem. Developers have chased cheap acreage, generous tax subsidies, and permissive rural zoning — and in doing so have systematically ignored the two resources that actually determine whether a data center succeeds over a decade or two: power and water.
The results are visible in the data. According to a comprehensive analysis using the World Resources Institute’s Aqueduct Water Risk Atlas, 65 percent of America’s 4,629 tracked data center facilities sit in areas classified as medium-high water stress or above. One hundred and forty-three operate in zones where water demand already exceeds renewable supply. Nevada leads the stress rankings with an average score of 4.8 out of 5. Arizona scores 4.1. Texas — home to 554 data centers — scores 3.1, with 537 of those facilities in water-stressed areas.
The Ogallala Aquifer, which underlies much of the Great Plains from the Texas Panhandle to Nebraska, is being drawn down at rates that make the word “abundant” a reckless fiction. Water drawn from deep aquifers in parts of Iowa and the Texas Panhandle will not be replenished on any human timescale. O’Leary’s proposed Utah campus would consume up to 9 gigawatts of electricity — more than double Utah’s current total electricity usage — in a state that depends on an already over-allocated Colorado River and a rapidly depleting groundwater table.
This is not a business plan. It is a liability waiting to be recognized.
The Right Places
There is a better model hiding in plain sight, and it is already being built — not in the Utah desert, but in Lancaster, Pennsylvania, on the footprint of two abandoned printing plants that had sat idle for years.
In February 2025, developers Chirisa Technology Parks and Machine Investment Group announced plans to convert 144 acres of former industrial land — the old R.R. Donnelley and LSC Communications printing facilities at 216 Greenfield Road and 1375 Harrisburg Pike — into a major AI data center campus leased to CoreWeave. By August, Blue Owl Capital had closed a $4 billion joint venture to fund the project, with CoreWeave committing up to $6 billion total. The campus will deliver an initial 100 megawatts of capacity, scalable to 300 megawatts, with $200 million in grid infrastructure improvements funded by the developers themselves.
No farmland was touched. No open space was converted. The city’s zoning officer confirmed the sites’ existing Suburban Manufacturing designation was fully compatible with data center use. Lancaster’s farmland preservation advocates — including the active watchdog group Respect Farmland — didn’t lift a finger to oppose it. There was nothing to oppose. The land was already industrial, already served by heavy electrical infrastructure, already connected to municipal water and sewer. The community got construction jobs, permanent employment, and a tax base replacement for buildings that had generated nothing for years.
What makes Central Pennsylvania exceptional is not just the availability of brownfield land. It is the convergence of fundamentals that sophisticated data center capital should be chasing everywhere. The region sits within PJM Interconnection — one of the nation’s largest deregulated electricity markets, where large customers can negotiate directly for power rather than depending on a regulated utility’s rate-setting process. The Susquehanna River watershed provides abundant, rechargeable freshwater that stands in stark contrast to the fossil water being mined beneath the Great Plains. And within forty miles of Lancaster, two nuclear facilities provide the firm, dispatchable baseload power that GPU clusters running 24 hours a day, 365 days a year actually require. The Crane Clean Energy Center — formerly Three Mile Island Unit 1, already powering Microsoft data centers — is back online. Unit 2 is on track for restart by late 2027, backed by a $1 billion federal loan. The Susquehanna Steam Electric Station in Berwick anchors northeastern Pennsylvania’s power profile.
A Distributed Model That Scales
Here is the argument that has not yet been made in the national data center conversation: you do not need 40,000 acres in a desert to build gigawatt-scale AI infrastructure. You need the discipline to think differently about scale.
A single 1-gigawatt hyperscale campus requires 500 to 800 acres — much of it land-banked for future phases that may never be built. But twelve 100-megawatt data centers, distributed across brownfield industrial sites throughout Central Pennsylvania’s Lancaster, Lebanon, Dauphin, York, Cumberland, and Berks counties, would together deliver 1.2 gigawatts — on perhaps 1,000 to 1,200 total acres of already-industrial land, spread across a grid that can absorb the load without strain, fed by nuclear baseload, cooled by watershed water that renews itself with every rain.
The distributed model delivers additional advantages that the megacampus cannot. Smaller facilities in existing industrial zones move through permitting faster and with less community friction. Distributed grid draws are less disruptive to regional electricity markets than a single massive interconnection request. For the inference workloads that increasingly dominate AI computing demand, distributed regional facilities actually perform better — lower latency, closer to population centers, more resilient to single-point failures.
Pennsylvania Department of Environmental Protection’s land recycling program has cleaned up more than 6,000 sites since 1995. EPA brownfield assessments have identified thousands of acres of former industrial land across the Commonwealth awaiting productive reuse. The American Farmland Trust’s Cost of Community Services research has consistently found that agricultural land generates more in tax revenue than it consumes in public services — while commercial and industrial development on existing infrastructure imposes zero marginal road, utility, or stormwater costs on local taxpayers. Every brownfield data center is a double victory: a derelict site returned to productive use, and a farm saved from a warehouse or a data center that had no business being there.
A Message to Developers
The communities that will welcome you already exist. The infrastructure is already there. The water is already flowing from a watershed that recharges with every storm. The power is already deregulated, already backstopped by nuclear generation that runs when the sun doesn’t shine and the wind doesn’t blow. The industrial land is already zoned, already assessed, already waiting.
You don’t need 40,000 acres in a Utah desert. You don’t need to fight farmland preservation groups, outrun water utility capacity, or ask three county commissioners to override a community’s objections before the cameras arrive. You need the wisdom to recognize what the developers in Lancaster, Pennsylvania already figured out.
It may seem improbable that the national model for responsible AI infrastructure would emerge from a county where horse-drawn buggies share the road with semitrucks. But perhaps it shouldn’t surprise us at all. The people who live among the Amish never forgot that land is not a commodity to be liquidated for the next speculative boom — that water is not an accounting abstraction but a finite gift, that a community’s consent is not an obstacle to be routed around but a privilege to be earned. They preserved their farmland precisely because they understood that prudence is not the enemy of progress. It is its prerequisite.
The fundamentals always win. In Lancaster County, the fundamentals are already here. The rest of the country would do well to take notes.
Visit us on Social Media:
USTW
IfSPP
