Data Center Backlash
The wave of local, state, and national opposition to AI data center construction that matured into a measurable political force in 2025–2026 — moratoria, ballot-box bans, recall campaigns, and a bipartisan polling profile that makes data centers one of the most unpopular infrastructure categories in the US. The backlash is best understood not as a technical dispute about water or decibels but as a renamed fight over inequality, precarity, and the social compact, with the data center as the most physical, least ignorable manifestation of the AI boom and of Big Tech generally. 1 ^[2026 — [@jasmine’s substack](https://jasmi.news/)
The phenomenon
Data centers are the most visible, physical manifestation of an otherwise “cloud”-seeming technology: hyperscale facilities span hundreds of acres, come wrapped in new transmission lines, and are — in Jasmine Sun’s phrase — “just greige.” That physicality makes them the available target for a diffuse set of anxieties about AI, Big Tech power, and economic order. ^[2026 — [@jasmine’s substack](https://jasmi.news/)
Scale of the response, as of mid-2026:
- At least 15 state legislatures introduced moratorium bills; ~100 localities approved their own pauses. New York imposed the first statewide moratorium (one year, ≥50 MW facilities, by executive order, July 2026); Maine’s legislature passed one (vetoed over a single-project carve-out); Pennsylvania’s governor imposed environmental/transparency safeguards, a local-approval requirement, and a ban on state NDAs with developers; Texas paused grid-interconnection approvals pending audit. Monterey Park, California became the first US city to ban data centers by ballot measure (June 2026). ^[Nicol Turner Lee, Darrell M. West 2026 — Data center moratoriums are not a substitute for oversight ^[2026 — Where are authorities restricting data centres amid AI boom?
- The phenomenon is international: Amsterdam (moratorium 2019, barred until at least 2030), a Dutch national hyperscale restriction (2022), Dublin’s grid-operator freeze (2021–Dec 2025, lifted with bring-your-own-power condition), Denmark’s draft bill giving data centers lowest grid-connection priority (backed by 80% of parliament), Australia’s planned “Office of AI” with siting/power/water rules. ^[2026 — Where are authorities restricting data centres amid AI boom?
- Polling shows roughly 30% for / 70% against, with Democrats and Republicans matched in opposition — comedian Charlie Berens called data centers “the most bipartisan issue since beer.” The coalition on the ground is correspondingly strange-bedfellowed: MAGA voters, democratic socialists, environmental advocates, retired bank executives, and unions landing on opposite sides of the question. ^[2026 — [@jasmine’s substack](https://jasmi.news/)
- At the federal level, Sanders/AOC introduced an AI data center moratorium bill (little traction), while House Energy & Commerce ranking member Pallone called for a national moratorium — even as both parties’ leaderships resist slowing buildout on China-competition grounds. ^[Nicol Turner Lee, Darrell M. West 2026 — Data center moratoriums are not a substitute for oversight
Competing explanations
The material account: data centers are big, ugly, noisy, energy-hungry, and locally disruptive during 2–6 year construction; genuine harms exist (xAI’s Memphis turbines worsening air quality in already-burdened neighborhoods; DTE’s 1.4 GW Saline-Stargate contract adding ~25% on top of total system demand with redacted power estimates). But by the numbers the impacts are comparable to factories or solar farms that attract nothing like this response — closed-loop cooling uses roughly golf-course quantities of water. The material account explains the location of fights but not their intensity or national character. ^[2026 — [@jasmine’s substack](https://jasmi.news/) ^[Nicol Turner Lee, Darrell M. West 2026 — Data center moratoriums are not a substitute for oversight
The process account: NDAs, fast-tracks, redacted filings, lawsuits against too-small-to-fight towns (Related Digital suing Saline Township, pop. 3,000, into a $14M settlement), and residents who feel “blindsided” — grassy field one day, cranes the next. The process failure is real, and the industry has started conceding it (Microsoft ended NDAs with local governments in 2026). ^[2026 — [@jasmine’s substack](https://jasmi.news/) ^[2026 — Where are authorities restricting data centres amid AI boom?
The relabeled-inequality account (Stokes): in places where “inequality” is coded as communist vocabulary, the same anxiety surfaces under a locally legible name — “datacenters.” The narrative structure is extraction: the few making proprietary the goods (water, farmland, quiet, privacy) that belong to the many. On this reading the factual inaccuracies in anti-datacenter propaganda (Stokes: “as gonzo as QANON was”) don’t matter, because facts aren’t what’s at issue; the fight is over the social compact. The pro-side’s pitch collapses into promising tax revenue — money for government services addressed to an electorate that distrusts government — because the facilities bring almost no permanent local jobs after construction. ^[2026 — ‘Here in Red Texas, You C…’ — Jon Stokes (@jon_stokes) Twitter/X thread
The thymos account (Sun, via Fukuyama): politics is not an optimization game and voter interests are not reducible to dollar amounts. A backroom data center deal is the negation of thymos — corrosive not because it’s uneconomical but because it demonstrates the demos’s buy-in was never important. This is why bigger benefit numbers can backfire (“I’m being bribed. There must be a catch.”) and why Sun’s overall verdict is that “AI populism is more about populism than about AI,” riding on historically low institutional trust. ^[2026 — [@jasmine’s substack](https://jasmi.news/)
These accounts are complementary: material impacts pick the sites, process failures supply the grievance’s evidence, and the inequality/thymos accounts explain why the grievance generalizes into a national movement rather than staying a series of local zoning disputes.
The trust asymmetry
Charles Franklin (Marquette Law Poll) observes you only get 30/70 sentiment when one side fails to make a case — and there is no pro-data-center faction: the companies are unsympathetic, permanent jobs are few, consumers don’t experience AI products as essential, and promised tax contributions are delayed and hypothetical in places burned before (Mount Pleasant’s Foxconn fiasco — $4B in incentives for 13,000 promised jobs that became ~1,000 — is the cautionary tale cited in conversation after conversation). The deindustrialized-town pattern repeats: a single big employer leaves, a contaminated or debt-saddled site remains, and the next big suitor is met by residents determined not to be fooled twice. ^[2026 — [@jasmine’s substack](https://jasmi.news/)
Sun’s summary: “If people are this angry at a building sitting and paying taxes on unused land, something fundamental has gone very wrong.” ^[2026 — [@jasmine’s substack](https://jasmi.news/)
Open questions
- Does the backlash survive benefit-sharing? Early counter-moves (ratepayer protection laws, community benefit agreements, New York’s proposed $1M-per-megawatt benefits floor, Meta’s Louisiana teacher bonuses) test whether the opposition is really priceable. The thymos account predicts partial failure: better deals don’t address the recognition deficit.
- Bubble exposure: if AI capex is a bubble, host communities fear stranded-asset risk (“another GM plant scenario”) — a rational deindustrialized-town prior, and one that makes opposition more risk-averse than anti-technology.
- Attribution politics: both OpenAI (blaming PRC-linked influence ops) and Republicans probing China ties have tried to externalize the backlash; Brookings and Sun both report the concern as genuine and domestic. The “astroturf” reading itself functions as evidence for the oligarchy story opponents tell.
- Whether the backlash ports to other AI infrastructure — generation buildout, transmission, mineral extraction — as data centers saturate available sites.
Related
- open-access-order — North, Wallis, and Weingast’s framework for understanding the backlash: the natural state returns when access to economic activity must be bargained for rather than governed by impersonal rules
- political-anger-emotions-policy-views — the 150-year anger series gives the backlash its emotional register; Stantcheva’s finding that online partisan anger rises acyclically matches the backlash’s both-parties-all-the-time profile
- abundance-agenda — the supply-side liberal answer (build, but govern the build) collides with a public whose objection is not primarily about supply
- bullshit-jobs — rival diagnosis of the same malaise; Stokes’s “permanent underclass, not evenly distributed” line rhymes with Graeber’s managerial-feudalism account of pointless work
- moral-economy-price-tag — the backlash as a moral-economy reaction: opposition to a transaction perceived to violate the moral order of the community, not merely to its price
- environmental-land-use-regulation-esa — the other US case where place-based regulation blocks nationally-desired construction; the ESA housing-permit findings are the mirror image (regulation as the blocker vs. regulation as the demanded remedy)
- butlerian-jihad-thesis — the endpoint Nathan flagged: if “datacenters” is where inequality anxiety gets expressed, throttling the buildout is the available political lever on AI itself
- ai-text-watermarking — the governance-half of AI legitimacy: provenance for outputs vs. consent for physical plant
- the-nearby-fujin — Xiang Biao’s lens: the backlash attaches to a visible local object because the far grievance (inequality) has no nearby scope; NIMBYism as the nearby’s pathology
Sources
- 2026 — ‘Here in Red Texas, You C…’ — Jon Stokes (@jon_stokes) Twitter/X thread
- 2026 — [@jasmine’s substack](https://jasmi.news/)
- Nicol Turner Lee, Darrell M. West 2026 — Data center moratoriums are not a substitute for oversight
- 2026 — Where are authorities restricting data centres amid AI boom?