The Veto Economy

For most of the last decade, data centers moved through local permitting the way office parks do: staff review, no public hearing, an administrative “yes”. Entitlement, the legal right to build a specific use on a specific site, was a line item, not a variable, and that assumption held long enough to become an expensive habit.

A modern hyperscale campus is not what it was ten years ago. It is a cluster of windowless concrete structures on the scale of a small industrial district, a dedicated substation, transmission corridors across the surrounding landscape, backup generators audible at the property line, evaporative cooling systems consuming millions of gallons of water annually, and two to three years of heavy construction traffic before the first server goes live. Average facility capacity has more than quadrupled over the past decade. The asset communities are being asked to host today bears little physical resemblance to the one they waved through without a public hearing in 2015.

The consequences show up in the approval process itself. A campus reviewed as technology gets benchmarked against office park standards. The same campus, understood as heavy industrial infrastructure, triggers public hearings, environmental review, and a different set of decision-makers with a different set of concerns. The asset is the same. What changed is the category it sits in, and in land use politics that distinction decides who gets to object, on what grounds, and through which institutional channels.

Developers tend to treat local resistance as a single, manageable phenomenon, which is precisely why it keeps catching them off guard. What appears as community opposition is a coalition of distinct constituencies, each carrying a different grievance, each holding a different institutional lever. Addressing one layer rarely dissolves the others. A project that satisfies the planning commission can still fail at the rate hearing, the ballot initiative, or the legislative session that follows.

Water is the most prevalent concern, appearing in more than 40 percent of contested projects, ahead of electricity costs, noise, and land use. Its prevalence reflects something specific: a water objection draws on environmental groups, agricultural interests, and residential ratepayers simultaneously, constituencies that would not ordinarily find common cause. The potency of opposition in most contested markets comes not from the intensity of any single complaint but from the structural alignment of complaints that ordinarily belong to different political camps.

Communities bear the industrial footprint, the water draw, the grid load, the construction years, while the economic benefits flow primarily to users and investors elsewhere. Treating that asymmetry as a communications problem rather than a structural one produces worse outcomes analytically and operationally. Each grievance connects to a constituency with a specific institutional lever, and when those levers are pulled in combination, opposition can outlast the developer's timeline.

States and municipalities have been extending incentives calibrated to employment-intensive industries to an asset class that generates almost no permanent employment by any comparable standard.

A 2017 U.S. Chamber of Commerce study of 244 facilities found an average of 1,688 workers during construction and 157 once operating. Nine in ten construction jobs disappear when the building opens. At the most automated campuses, permanent staffing runs about 25 to 40 operators per 100 megawatts. A one-gigawatt build costing eleven billion dollars generates fewer permanent jobs per dollar of investment than any large-scale energy project competing for the same grid capacity.

A Brookings study spanning roughly 770 counties found that for hyperscale facilities, incentives amount to approximately 2 percent of total construction investment. Siting at this scale is governed by power availability, fiber, and land cost. The tax abatement does not move the decision. Communities extending incentives to hyperscale projects are largely paying firms to locate where they were going to locate anyway. When the per-job subsidy figures above surface in a budget hearing or a local newspaper, the politics of the next incentive package become considerably harder.

Incentive repeal, sunset clauses, and per-job caps are active legislative tools in multiple states. Projects underwritten on the assumption that incentive terms are stable carry an exposure most pro formas do not model.

In March 2025, Loudoun County, Virginia, which contains the world’s densest concentration of data center capacity, ended by-right approval for the use. Within months, jurisdictions with no pending data center applications were drafting ordinances anyway, getting the authority on the books before a project could arrive and force the question.

At the state level, the Gardner Policy Institute counted eight active moratorium and ban efforts in May 2025. A year later the figure was seventy-eight. More than 100 local jurisdictions have enacted construction pauses. Over 300 state-level data center bills were filed in the first six weeks of 2026 alone. Virginia, Georgia, and Oklahoma, states that had competed aggressively on incentive packages, are actively reconsidering those programs. In December 2025, a sitting U.S. senator called for a national moratorium, the first federal official to do so. The center of gravity in data center permitting has moved from the developer’s legal team to the ballot box.

Into that environment, AI has arrived not just as a demand multiplier but as a political accelerant. The IEA projects global data center electricity consumption will more than double by 2030, reaching roughly 945 terawatt-hours, with AI-optimized capacity quadrupling within that total. U.S. data centers are on course to account for nearly half of all electricity-demand growth this decade. In PJM, serving 65 million Americans, capacity costs rose from $2.2 billion to $14.7 billion in a single auction cycle, and the increase is already showing up in household electricity bills.

A planning commission that once weighed a single building now faces a different question from its constituents: if this campus is approved, what follows? More campuses, additional substations, new transmission corridors, a permanent reorientation of the local grid toward compute rather than households. Public comment has shifted from compliance to trajectory, from whether a facility meets standards to what kind of place the jurisdiction is agreeing to become. Technical mitigation, closed-loop cooling, setback agreements, noise buffers, rarely answers that question.

Site-selection models price power, fiber, land, latency, and incentive packages with increasing precision. The variable they are structurally blind to is political feasibility: the probability that a site which can be built will actually be permitted, on a timeline the capital structure can sustain.

The two layers of feasibility are not correlated in the ways the models assume. The most technically attractive sites tend to produce specific political risk configurations. Cheap rural land is often adjacent to communities with the time and the cause to sustain a prolonged fight. Available substation capacity in an area that has not yet absorbed significant load growth means the rate impact of new demand lands immediately and visibly on existing customers. A welcoming statehouse is not the same as a welcoming county commission. The models optimize for the first layer and leave the second to judgment, which is why projects keep failing on the dimension that never appeared in the underwriting.

Industry trackers attribute roughly $46 billion in delays to opposition since mid-2024. Sightline Climate estimates a third of data centers slated to open may slip or never break ground. The IEA warns up to a fifth of planned projects globally are at risk absent transmission build-out that will itself face opposition in many of the same jurisdictions. A political defect discovered after land control, utility study, and rezoning can represent an eight-figure write-off and two lost years of position in the utility interconnection queue, the waiting list that determines when a facility can actually draw power from the grid. The same defect found at screening costs a desk review.

The standard development sequence commits capital before assessing political risk. In a market where a single contentious hearing can seed a moratorium and that moratorium can generate copycat ordinances across a region within months, that sequencing puts the most expensive discovery at the end of the process.

Political and social risk gets assessed before land control, mitigation gets designed in before opposition has a specific project to organize against, and community engagement starts early enough to be genuine rather than defensive. Local benefit commitments get quantified and disclosed before incentive programs come under political challenge, and sentiment gets tracked continuously rather than encountered for the first time at the podium.

The practical difference is when you find the problem. A political defect at screening costs an analyst’s time. The same defect after interconnection deposits, option payments, environmental studies, and rezoning filings costs eight figures and two years of queue position.

For thirty years, the binding constraints on digital infrastructure were technical and financial: chips, fiber, capital, and eventually megawatts. Each yielded, at some cost, to money and engineering. The constraint now binding in market after market is political, and it does not yield to a larger incentive package or a better legal team.

The sites that perform over the next decade will share one characteristic that current underwriting models do not score: permission that was acquired early, designed for systematically, and durable enough to survive the full development cycle. A site with abundant power and hostile neighbors is not a good site anymore. The firms that assess political feasibility with the same rigor they bring to grid capacity will pay less for better ground and avoid the losses that don't show up in the model until it's too late to exit. The firms that do not will keep learning the cost of permission the expensive way, one planning commission at a time. What that cost looks like by geography, and whether the regulatory landscape developers are betting on is as stable as they assume, is a question the data is beginning to answer.