The Infrastructure Pitch That Lost to a Nuclear Plant

If your growth plan depends on a community saying yes, the benefit case you have been rehearsing is probably not what they are deciding on.

By mid-2026, the physical footprint of artificial intelligence had become one of the few issues in American politics that unites voters who agree on almost nothing else. The fight over data centers, the facilities that house the servers behind AI, moved out of zoning board meetings and into governors’ races and Senate campaigns.

The polling is stark. A June 2026 Echelon Insights survey found that voters opposed building an AI data center in their community by 62 percent to 27 percent. Then came the part that matters most. Respondents were shown messaging about the economic and technological benefits, the industry’s best argument, and opposition still stood at 58 percent. In the same survey, more people supported a nuclear plant in their community (34 percent) than a data center (27 percent). Nuclear power has been the textbook example of a facility nobody wants next door for decades, and it now polls better than the infrastructure behind a chatbot.

The Annenberg Public Policy Center found the same direction. In its June and July survey, 61 percent of Americans opposed new data centers in their area, up from 49 percent in February and March. That 12-point move is the largest shift on any AI question the center has measured. It was not confined to one party: 69 percent of Democrats, 54 percent of Republicans, and 53 percent of independents were opposed.

The industry kept testing whether the benefits were big enough. The voters were answering a different question.

A poll can be an outlier. Behavior across institutions that disagree on everything else is harder to dismiss.

Residents have flooded zoning hearings, circulated petitions, and sued local governments over proposed projects, including in Coweta County, Georgia, and Sangamon County, Illinois. Campaigns on both sides have noticed. A Republican gubernatorial candidate in Wisconsin ran an ad branding his Democratic opponent with a data center nickname. A Democratic candidate in Texas ran one tying a household electric bill to data center demand and aimed it at a Republican governor. In Pennsylvania, a Republican challenger attacked Governor Josh Shapiro for putting data center developers ahead of communities.

Governors are moving too, across party lines. Texas Governor Greg Abbott ordered additional scrutiny of projects seeking grid connections, including requirements that data centers pay their own infrastructure costs. Shapiro signed an executive order requiring developers to secure local approval and meet standards on energy costs, environmental protection, transparency, and community engagement.

Most telling, the National Republican Senatorial Committee reportedly warned leading AI companies privately that public opposition was becoming a political liability, citing the Ohio Senate race. When opposing campaigns, governors from both parties, and a party’s own Senate committee reach the same conclusion independently, the problem is not a messaging gap. It is a behavioral pattern.

Start with the mismatch between where the industry argues and where residents decide.

Construal level theory: People think about things that are distant, future, or hypothetical in abstract terms, focusing on why it matters and what it achieves. They think about things that are near and imminent in concrete terms, focusing on how it will actually affect them. The same proposal can win support at one level and opposition at the other, because different features of it become salient.

The case for data centers lives at the abstract level: national competitiveness with China, economic development, jobs, and tax revenue. House Speaker Mike Johnson and President Trump both made versions of that argument this summer. The case against lives at the concrete level: the electric bill, the water supply, the noise from cooling equipment, the land next door. A resident at a zoning hearing is not evaluating American technological leadership. They are evaluating their street. Many officials back AI in theory, but their constituents’ skepticism sharpens as the facility gets closer to home.

The asymmetry between those two levels is made worse by a second mechanism.

Loss aversion: People weigh the pain of a loss more heavily than the pleasure of an equivalent gain. Things people already hold, such as stable bills, quiet, and a known water supply, are experienced as possessions to be protected, so threats to them are weighted more than promised gains of the same size.

Local costs land on identifiable neighbors and read as losses to things they already have. The benefits are diffuse, spread across a state, a tax base, or a company’s shareholders, and are uncertain from where the neighbor sits. Brookings senior fellow Darrell West described residents watching their electric rates rise while seeing little transparency in the negotiations. A loss that is concrete and local will beat a gain that is abstract and shared, almost every time.

That also helps explain why the benefits messaging barely moved the Echelon numbers, from 62 percent opposed to 58.

Persuasion knowledge model: People develop everyday theories about the tactics others use to influence them. When they detect a persuasion attempt, they tend to discount the message and become more skeptical of its source, especially when the persuader has something to gain.

A developer or an industry-aligned official arguing for the benefits of a project is, to a skeptical resident, an interested party making a pitch. The pitch is read as a pitch. A four-point movement is what you would expect from an argument that is being discounted as it is delivered. The survey did not test this directly, so treat it as consistent with the data, not proven by it.

Two of the instruments developers have used to move quickly and quietly turned out to be liabilities in themselves.

Shapiro’s order removes AI data center proposals from the state’s fast-track permitting program and bars the use of nondisclosure agreements tied to projects. Both tools exist for efficiency. Fast-track permitting shortens timelines. NDAs protect competitive information and prevent speculation from driving up land prices. But residents do not read them as efficiency. A fast-track path reads as a way around the community, and an NDA reads as proof that the developer expects the answer to be no if people knew the terms.

The signal is damaging precisely because it comes before the public hearing. By the time a developer arrives to make the benefit case, the process has already told residents how much the developer expected to need their consent.

The NRSC warning carries a second signal. The industry’s natural political allies are now privately treating its projects as something to be insulated from, not defended. That is information the industry itself should read before reading any poll.

The deeper issue is whether residents believe the process that produced the project was fair.

Procedural justice: The perception that a decision-making process was transparent, consistent, and gave affected parties a genuine voice, independent of whether the outcome itself was favorable. Research consistently finds that people accept unfavorable outcomes more readily when they judge the process fair, and resist favorable ones when they do not.

Residents cannot independently verify how much grid cost they will absorb, how much water the facility will draw, or what was promised to whom, because the negotiations often happen under confidentiality. When a process is opaque, people do not assume the best case. They assume the version that explains why it was hidden.

That may also explain the nuclear comparison, and it should be read as a hypothesis, since the poll did not ask why. Whatever people think of nuclear power, a plant arrives through a siting process residents recognize: federal review, public hearings, a known institution to hold accountable. A data center can arrive through a negotiation that residents experience as concluded without them.

The fix is to stop treating community consent as a downstream communications problem and start treating it as an upstream design input.

• Put the community’s yes before the site commitment, not after it. Local approval as a precondition, as in Pennsylvania’s order, changes the incentive. A developer who knows consent is required designs the project to earn it instead of managing the opposition once capital is sunk.

• Lead with the costs residents are actually pricing, with binding commitments instead of narrative. Contractual protections on ratepayer exposure, such as the principle that the facility pays its own infrastructure costs, address the loss residents fear. A promise of economic benefit does not.

• Retire nondisclosure agreements in siting. If terms have to stay confidential, publish a summary of the ones that touch residents: grid costs, water draw, tax treatment. Concealment is the signal. Disclosure removes it.

• Match the concentration of benefits to the concentration of costs. If the costs fall on the neighbors, a meaningful share of the benefit should reach the neighbors directly, not only the county budget.

• Track opposition before the hearing, not at it. By the time a project reaches a public meeting, the trust judgment has already been made. The earliest reliable measure is what residents believe the costs will be, tested before the first filing.

• The surveys measured opposition, not reasons. The Annenberg researchers noted that their survey did not ask respondents why they opposed data centers. The mechanisms in this piece are inferences from the pattern of evidence, consistent with the data but not demonstrated by it.

• Stated opposition is not revealed behavior. A poll about a hypothetical local site can overstate what residents do when a real project, with real jobs and a real tax check, is on the table. Speaker Johnson noted that communities in Louisiana have welcomed these facilities, and plenty of projects are still being approved. The gap between what people say in a survey and what a local body ultimately votes is exactly the kind of gap this newsletter is skeptical of, and it applies here too.

• Binding commitments cost developers something, and the competitive argument is not frivolous. Requiring facilities to fund their own grid upgrades changes project economics. Trump’s warning that places declining projects will be left behind reflects a real constraint: capital can move to jurisdictions that ask for less, so strict conditions in one place can export the project, and its costs, elsewhere.

• Procedure cannot create consent that does not exist. Opposition stayed at 58 percent even after benefits messaging. A better process lowers the temperature of the fight, but some share of residents will oppose a project however fairly it is sited, and a local veto can block infrastructure with broad national value.

• Trust lost at the siting stage is slow to rebuild. Developers who ran fast-track and confidentiality first and pivot to transparency later are doing so with a community that has already drawn its conclusion about what the earlier process meant.

If your organization needs a community, a customer base, or a workforce to say yes to something you have already decided to build, when did you last test what they believe the costs are, separate from how well you have explained the benefits?