One/Why counties say no
Research

Why counties say no to data centers

A working explanation of the local approval problem for powered-land developers: what is actually being decided, why it got harder in 2025 and 2026, and the four things approved sites share. Written by One, a site-approvals practice in the Southeast.

Short answer

Counties reject data center rezonings for four fears that nobody answered in writing before residents heard about the project: the power bill, the water, the noise, and what happens when the campus is finished, sold or expanded. A fifth is identity: who are you, behind the code-name LLC.

The order of disclosure matters more than the size of the benefits package. In the cases we have studied, every denied project followed the same sequence: officials under NDA first, a small circle for months, then disclosure into a room that had already organized. Money, architecture and amenities offered after that point did not change an outcome.

What is actually being decided

Decompose a hyperscale campus into its inputs and ask which one is scarce, non-fungible, cannot be bought, is binary, and is controlled by a third party with a veto.

InputScarcePurchasableBinaryThird-party veto
LandNoYesNoNo
CapitalNoYesNoNo
TenantNon/aNoNo
Fiber, waterSometimesYesNoRarely
PowerYesYes, with timeNoThe utility and regulator, who want the load
Local consentYesNoYesYes

Power is the input everyone talks about, and it is real, but it fails the veto test: utilities are load-hungry, regulators are certifying generation, and a large-load tariff turns the rate question into a contract. Power is a when problem. Consent is a whether problem. It is decided by a handful of elected officials who bear the political cost of the project immediately and personally, while its benefits arrive years later and mostly to other people. It is also where the land's value steps from dollars per acre to dollars per megawatt, and nowhere else.

A gate decided by someone else cannot be crashed with money or run in parallel. You cannot pay a county commission to vote faster, and pursuing five counties at once to hedge the risk produces the pattern, “they are everywhere, they hide behind LLCs,” that raises the denial rate in all five.

Why it got harder in 2025 and 2026

Data Center Watch, which tracks local opposition, counted at least 75 projects worth roughly $130 billion blocked or delayed in the first quarter of 2026, about as much as all of 2025, and another 45 projects worth about $68 billion in the second quarter. The number of active opposition groups it tracks roughly doubled in a single quarter, to more than 800 across 49 states. Statewide moratorium bills were introduced in 14 states in the first three months of 2026; none became law, and local moratoria are now adopted pre-emptively, sometimes before an application exists.

Seven drivers explain the inflection. Electricity rates became the master grievance once PJM's capacity prices rose sharply and its market monitor named data centers the primary reason, so every county now has a resident who has read that data centers raised bills in Ohio. Loudoun and Prince William counties in Virginia supplied the visible precedent for what a cluster looks and sounds like. Memphis supplied the environmental-justice frame after gas turbines ran without permits. Water became a top-three objection in nearly every fight, regardless of whether the design uses any. The scale of 2025 and 2026 capex multiplied the number of communities receiving a proposal. Organizational learning turned isolated grievances into a networked movement with shared ordinances and messaging. And state rate cases gave the anxiety a legislative outlet.

The practical consequence: the burden of proof has flipped. In 2023 a developer had to show why a project was good. In 2026 a developer has to show, in writing, why each of the four fears will not come true, and has to show it before anyone asks.

The four fears, and the fifth

Each fear has an enforceable answer: a letter under the utility's large-load tariff, a water covenant, a property-line noise limit with public monitoring, a no-expansion clause and a funded decommissioning bond, and a named developer with a named local champion. The answers exist in every approval we have studied. What separates the approvals from the denials is whether they were on paper before residents heard the name.

What the record shows

Where it worked

Port Washington, Wisconsin went from disclosure to a rezoning in months and to a named hyperscale tenant within a year. The city had a fallen-through fab site and an existing industrial expectation, a mayor who championed it, a tax-increment district the city wanted, and state legislation that ring-fenced the generation cost.

Monroe County, Georgia rezoned a campus next to Plant Scherer, in a county that already hosts a coal plant and a gas plant, with regional water-authority service and a development authority able to structure a payment in lieu of taxes. It was framed as low impact, high revenue, and the frame was credible because the county had seen the revenue before.

Columbia County, Georgia approved a campus near Harlem in February 2026 by a four-to-one vote, with berms and buffers, a landowner family as the project's local face, and a county that already collects fees from an existing Google campus. The benefit was legible before the vote. The same night, the same commission denied a second campus four to one.

Where it failed

In Wrightstown, Wisconsin, the first email to the village administrator came in January 2026, residents learned of the project in the spring, and an advisory referendum in August returned nearly 86 percent “no.” The village then adopted a twelve-month moratorium. Manitowoc County, Wisconsin adopted a moratorium before anything was filed, because offers to farms leaked. In Dundee, Michigan, officials met in July, residents in October, and the pre-development agreement was retracted the same month. In Tonganoxie, Kansas a moratorium expired on a three-to-two vote and opponents then won two of three commission seats. In every case the sequence was the same: officials first under NDA, a small circle for months, then disclosure into a room that had already organized.

Where money and design were tried as substitutes

Saline Township, Michigan denied a rezoning four to one. The developer sued two days later; a township with a thin insurance policy signed a consent judgment in October 2025 that delivered about $14 million in direct benefits, split among fire services, a farmland preservation trust and a community fund, plus a noise limit and a no-expansion clause. The money did not buy the approval; the lawsuit did, and the gifts that followed read locally as bribes. In Phoenix, a campus with architect-designed buildings, hundreds of new trees and a required public park is being built because the land was zoned for the use decades ago and the approvals were administrative. Neighbors are still protesting. Design did not buy consent; grandfathered zoning did.

The pattern

Approval correlated with four things: the county had prior industrial or data center experience (a power plant, an existing campus, a fab site); a local champion or landowner-partner introduced the project rather than a code-name LLC; a legible, county-controlled benefit mechanism already existed (a development authority PILOT, a fee stream, a tax-increment district, a state statute); and the state's rules answered the rate question structurally (a large-load tariff, a ring-fencing act).

Failure correlated with the mirror image and, above all, with the length of time between the first official conversation and the first resident conversation. Amenities, architecture and cash offered after that gap had opened did not change an outcome in any case examined.

A denial is not a null result. It produces a model ordinance, a Facebook group with a playbook, and a news story the next county's commissioners read before their own vote. A walk-away costs an option; a denial costs the state.

The political screen

Site selection is down to a science on the physical layer. The political layer is not modeled at all. The screen below scores a county before the first option is signed, on the dimensions the record suggests matter. A county scoring poorly on three or more should not be optioned on its grid score, or should be optioned only with a plan to fix the specific gaps before disclosure. This is the single highest-leverage change available to a developer, because it converts the approval rate from a random variable into a selected one.

DimensionSignal of capacitySignal of risk
Decision structureProfessional county manager; five to seven members; prior large rezoningsA board of part-timers with no planning staff
Election calendarVote lands more than twelve months before an electionHearing inside an election window
PrecedentA power plant, an existing campus or a fab site within thirty milesFirst of its kind for the county
Moratorium exposureNo pending local pause; a state without moratorium momentumA neighbor county just adopted one
Opposition networkNone, or dormantAn existing group with a playbook
Comprehensive planIndustrial or utility designation already on the parcelAgricultural preservation with a purchased-rights program
Fiscal needA school district that needs money is an allyA county with a surplus and no appetite
Utility standingUtility trusted; large-load tariff in forceRecent rate hike; a co-op in a rate dispute
WaterPublic authority with capacity; closed-loop designPrivate wells; an aquifer story already running
Champion availabilityA landowner family, a mayor, a development authority director willing to front itNo local face; the developer is the only voice
Media environmentA functioning local paper that will print the FAQOnly the opposition group's page

A redesigned path

The usual path is quiet options, utility talks, officials under NDA, disclosure, vote, then benefits and tax. The redesigned path inserts two stages between officials-under-NDA and disclosure and moves the benefits work ahead of the vote.

StageExit conditionTypical time
0. ScreenPhysical and political scores both pass; gaps namedWeeks
1. QuietGrid answer in hand; the first five officials briefed under NDA, none surprised later3 to 6 months
2. ChampionA named local face; three to five validators; twenty conversations; go, fix or walk30 to 60 days
3. CommitmentsA term sheet or host-community agreement in draft before anyone outside the circle has heard the name30 to 60 days
4. DisclosureNamed developer, local champion in front, FAQ, listening sessions; opposition forms around details, not existence60 to 120 days
5. VoteApproval with conditions that match the commitments1 to 3 months
6. Keep the promises publiclyNoise and water monitoring published; annual report; the next county can call this oneOngoing

Stage 6 is not an afterthought. Columbia County said yes to a second campus because the first one paid what it said it would. The cheapest approval in a portfolio is the one bought by the last project's behavior.

If you have a stuck site

That is the engagement One runs: ninety days, one site, ending in a plain recommendation.

Sources