Social License Is the Unpriced Constraint in the AI Buildout
The binding constraint on the American AI buildout is not chips, capital, or grid. It is social license — and the Gallup poll released this month confirms that the constraint has already bound (link in footnote).
Seventy-one percent of Americans oppose AI data centers being built in their local area. Forty-eight percent oppose strongly. The opposition is bipartisan — 56% of Democrats, 39% of Republicans, 48% of independents say strongly opposed (Gallup, fielded March 2–18, 2026, n=1,000, ±4 pts). There is no meaningful split by age, race, education, income, or urbanicity, with the gender gap (women 55% vs men 43% strongly opposed) the only meaningful demographic divide. Seventy-one percent of Americans do not agree the sky is blue. They have agreed on this.
Barely a quarter favor these projects; only 7% strongly. That is not erosion. It is the political equivalent of a closed door. Any siting plan that assumes communities will accept the current deal structure — tax abatement, vague jobs promises, ratepayer-funded transmission, water rights traded away in side letters — is planning against a wall the planners have not yet seen.
This essay argues three claims that follow from that wall. First, the regulatory framework around data-center siting is not absent; it is stale, and capital has learned to arbitrage the staleness. Second, local government — zoning boards, county commissions, utility commission interveners, water districts — is the venue where the 71% expresses itself, and the federal-versus-local asymmetry is now the most important political fact in the sector. Third, the next playbook is a deal-structure question, and the developers who get there first will outpace the developers still running the 2022 pitch.
The framework is stale, not absent
Data centers are not unregulated. They sit inside local zoning codes, state utility commissions, federal interconnection queues, and — where there is a federal nexus — environmental review. The framework exists. It was written for a different era of industrial siting and a different scale of energy demand, and it was not designed for a customer that needs five hundred megawatts in eighteen months.
Hyperscalers have learned to arbitrage that staleness. They locate where tax abatements are largest. They negotiate interconnects in queues already congested with renewable projects. They socialize the marginal cost of new transmission and generation through the ratepayer base. The pitch to communities is jobs. The actual local effect is concentrated environmental burden and diffuse cost socialization, which is the same effect industrial siting produced in the 1970s before the federal environmental statutes were rewritten.
Half of opponents name "excessive use of resources" as their primary concern — 18% each cite water and energy specifically, and another 16% cite pollution. That is a resource-burden coalition, not a NIMBY reflex. It maps directly onto what the deal structure actually delivers locally.
This is not a failure of capitalism in some moral sense. It is what capital does when the rules are stale and the upside is large. The honest framing is that the framework needs rebuilding for the current scale and customer type — not that the framework does not exist. Anyone arguing that data centers are "unregulated" is making it easier for capital to keep running the arbitrage, because they are picking a fight the developer cannot lose.
The federal-versus-local asymmetry is now the operative political fact
The federal posture on AI infrastructure is and will remain supportive. The White House has named AI a national priority. Congress is moving on permitting reform. The Department of Energy is funding grid upgrades. Federal agencies will continue to cheerlead.
None of that matters at the venue where data-center permits actually get approved or denied. County commissioners hold the conditional-use permit. State utility commissioners approve the special-rate tariff. Water districts hold the withdrawal allocation. Zoning boards hold the setback variance. Each of those venues is staffed by elected or appointed officials who read local polls, attend local meetings, and face local voters.
When opposition is bipartisan and crosses every demographic except gender, the political math at the local level is unambiguous. A commissioner who approves a data center on the old terms in 2026 is taking on a career-limiting vote, and the commissioners who survive the next cycle will be the ones who learned that lesson early.
What this means for developers is operationally specific. The federal interconnection queue and the federal permitting reform conversation are no longer the binding constraint on project delivery. The binding constraint is whether the deal you can offer the host community survives a six-hour public hearing where half the attendees are organized opposition.
Local opposition has gotten better. The resource-burden frame — water, energy, pollution, ratepayer cost — gives organizers a coalition that holds across party lines, and the bipartisan poll numbers tell every local elected official that the coalition is real. The federal cheerleading does not reach the room where the hearing is being held.
The next playbook is a deal-structure question
The current siting playbook is finished. Communities have learned what the deal actually looks like, and they are voting against it in the venues where their votes still count. The interesting question is what replaces it, because something will — the underlying economic demand for compute is not going away.
The next playbook will be a deal-structure question, and four elements appear in nearly every credible proposal being drafted by counsel for community-side coalitions:
Explicit revenue share to host communities. Not jobs promises, not tax-abatement clawbacks. A percentage of facility revenue paid directly to the county or municipality, indexed to compute output, and audited.
Ratepayer ringfencing. Grid upgrades attributable to a specific data-center load are billed to that load, not socialized onto residential and small-commercial ratepayers. This is the issue that has united consumer advocates and conservative tax groups in the same intervention filings.
Enforceable water guarantees. Not "the facility will use best practices." A specific withdrawal cap, a specific reuse target, a specific drought-curtailment trigger, and a specific penalty schedule that survives the operator's bankruptcy.
Decommissioning bonds that outlive the operator. The chip generation a facility is built around has a ten-to-fifteen-year economic life. The deal structure has to handle what happens when the facility is obsolete and the operator has moved on or wound down. The 1980s mining-reclamation statutes already answered this question for a different industry; the substrate is available.
A developer who walks into a county hearing with those four elements drafted into the proposal — and the third-party verification framework named — will run a faster project than a developer offering the 2022 deal. The faster project will pencil, even with the revenue share, because the time-to-energization savings dominate the deal economics in a compute market priced against scarcity.
Implication for capital allocators
The thesis above has direct implications for anyone underwriting AI infrastructure as an asset class.
For private equity and infrastructure funds with exposure to data-center development, the unpriced risk on existing portfolio assets is litigation and re-permitting risk on facilities approved under the old deal structure. The risk is concentrated in jurisdictions where the local political composition has flipped during the build cycle. Allocators should be asking sponsors which assets carry that exposure and what the re-permitting cost would look like.
For new deployments, the question is which developers have internalized the new playbook and which are still pitching 2022. The developers running the new playbook will be slower to break ground and faster to energize, because they will lose fewer projects to community opposition in years two and three. That trade-off favors the new playbook on a risk-adjusted basis, and underwriters who do not price it that way are mispricing the asset.
For policy-adjacent allocators — pension funds with ESG mandates, sovereign wealth funds, climate-aligned infrastructure vehicles — the social-license question is also a fiduciary question. A facility that lacks community consent is a facility with a non-trivial probability of operating disruption, regulatory reversal, or asset stranding. That probability belongs in the pro forma.
The Gallup poll is not a story about citizen anger, though it is partly that. It is a story about which infrastructure model the country will consent to fund, host, and protect — and the model being built is not that one. The developers who reprice on that finding will move faster. The allocators who reprice on it will lose less.
Reprice the asset.
Source: Gallup, "Most Americans Oppose Data Centers in Their Local Area," fielded March 2–18, 2026 (n=1,000; ±4 pts), with open-ended follow-up April 1–15, 2026 (n=2,054 Gallup Panel).
Jonathan Lyons, CVA, MBA — Founder, Among & Between.