The data center industry spent the last decade believing that power availability was the binding constraint on where capacity would land. That mental model is now obsolete. The binding constraint in 2026 is community license. The campuses that close in the next 36 months will be filtered less by megawatts available and more by whether the host jurisdiction is still willing to permit the project at the moment the shovel needs to enter the ground.
This is a structural shift, not a sentiment shift. It is measurable in county minutes, ballot results, rate filings, and water permits. It will determine where the next 100 gigawatts of AI capacity actually gets built.
What The Map Looked Like 24 Months Ago
In early 2024, the prevailing assumption inside hyperscale real estate teams was that any jurisdiction with a 230 kV line, a willing landowner, and a friendly assessor would compete for the project. Site selection ran on a checklist: substation proximity, fiber, water for evaporative cooling, county tax abatement posture, fast-track permitting. Community engagement was a downstream workstream handled by a public affairs vendor after the LOI was signed.
The map reflected that assumption. Northern Virginia carried roughly 70 percent of US cloud capacity. Phoenix, Dallas, Atlanta, Columbus, and Reno absorbed the overflow. New build pipelines stacked into rural and exurban counties on the edges of these metros, on the assumption that ag-zoned parcels with three-phase power were the path of least resistance. The model was: drop a 300 to 500 megawatt campus into a county of 80,000 residents, employ 40 people, pay property tax, and let the utility socialize the grid upgrades across the residential base.
That model worked for a window. It is now politically dead in most of the country.
The Immune Response Is Now Measurable
Loudoun County, which generates roughly 35 percent of its tax revenue from data centers, has tightened its zoning ordinance and pushed new entitlements into a longer, more contested process. Prince William County rejected the PW Digital Gateway expansion in its original form after sustained community opposition. Fairfax revised its data center zoning text to limit by-right development. Across Virginia, the conversation has moved from siting incentives to siting restrictions inside 18 months.
Memphis became a national flashpoint over the xAI Colossus deployment and its diesel generation footprint, with the Shelby County Health Department under public pressure to revisit air permits and the NAACP filing intent to sue over particulate exposure in South Memphis. The conflict reframed the entire industry's posture on backup generation in environmental justice communities.
Tucson rejected Project Blue, an Amazon data center proposal, after a council vote that cited water consumption in a basin already under shortage declaration. Chandler imposed water restrictions that constrain new hyperscale interconnects. Hillsboro, Oregon faces sustained pressure over Tualatin Valley water allocations to the western data center cluster.
New Jersey utility regulators are processing a wave of residential rate complaints tied to PJM capacity auction results, which cleared at roughly nine times prior levels and translated directly into double-digit percentage increases on residential bills. Ohio HB 15 fights are now centered on whether data centers should sit in their own rate class with cost-causation pricing. Georgia regulators have approved interim rate riders specifically designed to shift large-load infrastructure costs back to the loads that caused them.
By a conservative count tracked across county and municipal filings, more than 200 US localities have enacted moratoria, restrictive zoning amendments, or de facto pauses on data center development between January 2024 and mid-2025. The velocity of new restrictions is accelerating, not decelerating. Every new headline about a residential rate increase or a water curtailment puts another county on the path.
This is not isolated NIMBYism. It is a coordinated immune response, and it is now the dominant variable in site selection.
Why The Old Model Triggers It
The community math is brutal and consistent across jurisdictions.
A hyperscale data center produces roughly 0.1 to 0.15 permanent jobs per megawatt of IT load. A 300 megawatt campus employs 30 to 45 full-time staff at steady state. Compare that to other industrial uses competing for the same parcels and the same grid capacity. Advanced manufacturing produces three to eight jobs per megawatt. Food processing produces five to fifteen. Even logistics and warehousing produce two to four. The data center is the lowest job density industrial tenant in the modern economy.
The rate-shift mechanics make the math worse. When a hyperscaler interconnects a 500 megawatt load, the utility files for cost recovery on the associated transmission upgrades, substation builds, and generation capacity. Under traditional cost-allocation methodology, those costs are spread across the rate base. The hyperscaler pays an industrial tariff. The retiree on a fixed income three counties away pays a residential tariff. Both tariffs rise. The residential tariff rises faster as a percentage of household income.
PJM's 2024 to 2025 capacity auction cleared at approximately 800 percent of the prior year, with auction results translating into 10 to 20 percent residential bill increases across the footprint. The capacity scarcity that drove the auction outcome is overwhelmingly attributable to large-load interconnection queue growth, of which data centers are the dominant component. Every residential customer in PJM territory now has a direct, line-item financial reason to oppose the next interconnection.
Water competition layers on top. Evaporative cooling at hyperscale consumes one to three million gallons per day at typical 100 megawatt blocks. In western and southwestern basins under sustained shortage, that volume is no longer politically allocatable. Tucson, Chandler, parts of the Front Range, the Tualatin basin, and the Salt Lake corridor have either imposed explicit data center water restrictions or are processing them through utility commissions and city councils.
The composite result is a tenant that consumes the most stressed local resource (grid capacity and water), generates the lowest job count per unit of consumption, exports operating margin to a corporate headquarters in another state, and asks the host community to accept higher utility bills as the cost of admission. There is no civic counter-narrative that survives a public hearing.
What Survives The Immune Response
A different operating system is emerging in the projects that are still closing in 2026. It has four structural components.
Industrial-only zoning with no housing adjacency. The campuses that permit fastest are siting on heavy industrial parcels with no residential overlay within meaningful distance. This eliminates the noise, visual, and adjacent property value arguments that drove the Prince William and Loudoun reversals. It also concentrates the project in a tax district where the local government has structural reasons to defend the use.
On-site or behind-the-meter generation that does not pressure residential rates. The single most durable response to the ratepayer problem is to disconnect the campus load from the residential rate base. Behind-the-meter natural gas with carbon capture optionality, co-located solar plus storage, small modular reactor pre-positioning, and direct PPA structures that route around the regulated utility all accomplish this. They are more expensive per megawatt. They are also the only configurations that survive a state PUC under residential rate pressure.
Local tax structure that captures meaningful revenue per megawatt. Property tax alone is no longer sufficient. The projects clearing community thresholds are negotiating per-megawatt annual payments, gross revenue royalties, or PILOT structures that scale with campus expansion. The dollar figure that matters is per-megawatt fiscal contribution to the host jurisdiction, and the floor for political durability is rising rapidly. Five to ten thousand dollars per megawatt per year, captured at the local level, changes the political math at the county board.
Workforce programs that produce trades careers, not press releases. Funded apprenticeship pipelines with the local community college, contractual hiring floors enforced through liquidated damages, and operations staffing models that route through local labor markets rather than imported contractors. The campuses building this are creating a constituency of skilled trades households inside the host jurisdiction. That constituency votes in the local elections that determine entitlements.
These are not community benefit agreements layered on top of a conventional project. They are the project. The economics get harder. The closing probability gets higher.
What That Implies For The Next 100 GW
Three forward projections follow directly.
First, the geography of new hyperscale capacity will shift further toward jurisdictions that have already absorbed industrial land use as part of their civic identity. Heavy industrial corridors in Louisiana, parts of Texas outside the major metros, the Ohio River industrial belt, and select Midwest manufacturing counties will absorb a disproportionate share of new build. Exurban Virginia, ag-edge Arizona, and bedroom-community Georgia will absorb less than current pipelines suggest. Announced gigawatts in the wrong jurisdictions will quietly disappear from the pipeline through 2027.
Second, the cost stack for entitled, buildable hyperscale capacity will rise materially. Behind-the-meter generation, enforceable local hiring commitments, per-megawatt fiscal contributions, and water solutions that do not draw from contested municipal supply add real dollars to total project cost. The marginal megawatt of AI capacity in 2028 will cost more than the marginal megawatt in 2024, even before accounting for chip and construction inflation. Capital allocators who modeled returns on 2023 assumptions will reprice.
Third, state-level intervention will accelerate. The interim rate rider precedent in Georgia, the proposed data center rate class in Ohio, and the New Jersey ratepayer pressure all point in the same direction: large-load customers will increasingly bear the cost-causation share of grid upgrades they trigger. This is not anti-data center policy. It is the political pricing of an externality that the industry has been allowed to socialize for a decade. The states that move first will lose some announcements and keep more residential goodwill. The states that move last will face ballot measures.
The composite picture is a market filtering on community license. The campuses that close will be the ones built on the new operating system. The campuses that do not will sit in the pipeline as press releases until they are quietly cancelled or relocated.
Hyperscale dropping into ag and residential edges, importing labor, exporting cash, and asking the host community to pay higher electric bills to subsidize OpEx is the model that triggered the immune response. That model will not build the next 100 gigawatts.
Autonomous Industries. AI Industrial Infrastructure.