CASE STUDY

Data Centers Are Contested in 24 States, and the Friction Curve Is Steepening

Municipal Alpha tracks data-center zoning and siting activity across US municipalities, surfacing it from planning-board minutes and committee agendas months before national reporting. As of May 2026, 80 distinct municipal and state records across 47 jurisdictions in 24 states showed the pattern forming.

Updated August 2026. Original analysis (below) current to May 14, 2026, based on 80 distinct municipal and state records then flagged in the production pipeline as data-center-zoning-related, spanning 47 jurisdictions across 24 states. See "The August 2026 Update" for what changed since.

What This Piece Is, and What It Isn't

This is not an attempt to catalog every data center zoning action in America. The corpus is incrementally built, weighted toward where buyer interest has pulled coverage, and there are states we have not yet onboarded deeply. The signal density is therefore not a true denominator. What this piece does claim is something narrower and more useful for an institutional reader: when the same regulatory pattern is forming in real time across diverse jurisdictions, the early evidence of it sits in municipal meeting agendas, planning board minutes, and committee work sessions, months before the pattern reaches national reporting. The job of the corpus is to find that first, not to find it all.

A note on the headline numbers up front. We deliberately separate two metrics that are easy to conflate:

1. The apples-to-apples market-acceleration view. Within the cohort of jurisdictions that were already in our pipeline as of March 1, 2026, data center zoning records went from 3 in February to roughly 41 in April, with continued activity into May. That growth is on a fixed coverage set, so it reflects real change in municipal behavior rather than corpus expansion.
2. The broader regulatory-reach view. Adding jurisdictions newly onboarded since March 1, the same pattern is now visible in 80 distinct records across 47 jurisdictions in 24 states. Most of the new jurisdictions came online because the pattern itself drew attention to them, which is its own form of evidence, but it is not the same as saying market activity grew by the same factor.

Both numbers are useful. Treating them as one number would not be.

The Acceleration, Within the Stable Cohort

This is the cleaner read. We hold the coverage set fixed at jurisdictions present in the pipeline before March 1, then count data center zoning records by month.

Month	Records in cohort	Cohort jurisdictions active	Cohort states active
Feb 2026	3	2	2
Mar 2026	7	6	4
Apr 2026	41	4	2
May 2026 (through 14th)	19	3	1

Three observations.

First, the absolute numbers are small. With 18 distinct underlying documents in the stable cohort, we are talking about a thin signal, not a deluge. That is consistent with reading municipal calendars closely rather than crawling broadly.

Second, the April spike (~6x over March) coincides with the legislative drama around Maine's LD307. The bill would have been the country's first state-level moratorium on data centers requiring more than 20 megawatts. The House passed it 82-62 on April 6. Governor Mills vetoed it on April 24. The House attempted an override on April 29 and failed 72-65 (10 votes short of the supermajority and 10 votes flipped from the original passage). On the same day Mills also signed LD713, which prohibits data centers from claiming tax incentives under Maine's business development programs, and issued an executive order establishing a Maine Data Center Advisory Council. The April spike in municipal records across other states maps to this full arc, not just the House vote. Net Maine outcome: no moratorium, but a different shape of restriction landed in parallel.

Third, the cohort state count narrows over time (4 in March, 2 in April, 1 in May) while record count rises. That is what depth looks like in the data: fewer states, more documents per state, more committees per municipality engaging the same question.

The Broader Reach, Including Newly Onboarded Coverage

Across the full current pipeline, the same topic now surfaces 80 distinct documents across 47 jurisdictions in 24 states. The split between the cohort and the broader view is meaningful: 18 documents in the stable cohort, 62 documents from jurisdictions added since March 1. The broader pattern is real, but a portion of the count reflects coverage extension, not market expansion.

Top states by distinct documents in the current dataset:

State	Approx. distinct records	Jurisdictions
ME	~24	8
NH	~17	5
IA	2	2
MA	~5	3
ID	~3	3
GA	~3	3
AZ	~3	3
Seventeen other states	~23	20

The Northeast concentration is partly real (the regional grid is constrained, and state public-records frameworks make the discussion legible) and partly coverage-side (our corpus has the most depth in that region because that is where buyer interest has pulled it). We have not yet run a stratified scan across all 50 states with uniform sampling density, so we cannot fully separate those two factors. That is a real limit and we name it.

The August 2026 Update

A platform decision-maker at a renewables developer described the moment his own business usually finds out too late to matter, once a data-center developer has already signed the utility load application:

"He's too far past the ball already. ... I want to see something before that."

That complaint is really about a town's disposition toward the project, whether it will fight it or compete for it, and that decision is visible earlier still. It shows up in committee, in zoning language, in incentive votes, months before any load application, any lawsuit, or any press coverage.

That data centers are running into local resistance is not news; it is a national story by now. What national coverage does not do is the drill-down: which board, which meeting, which vote, and how long before the moratorium or the lawsuit reaches the press. This piece tracks that disposition in both directions, towns moving to restrict, and towns moving to compete.

What happened in August

• Utica, Wisconsin. The town board voted 3-0 on August 10 to approve Winnebago County's data-center text amendment and 12-month moratorium for the town. The chairman, in the minutes: "it's keeping them out of our town."
• Scarborough, Maine. Its moratorium is now in court. A developer is suing over both the moratorium and the town's "incomplete application" finding; the town's motion to dismiss has already been denied, and the case is in discovery.
• Brunswick, Maine. Council reviewed a tiered zoning framework built around the 180-day moratorium it adopted in June.
• Lockport, New York (the Town, not the neighboring City). Noticed an 18-month moratorium on data-center permitting for an August 5 public hearing.

None of it started in August

• In Utica, a resident asked the board directly on July 13 whether it would recommend a moratorium "next month." The chairman said the town would look into it. The board adopted one unanimously at its very next meeting, August 10.
• In Brunswick, the item to set a June 1 public hearing on the moratorium was already on the town's May 18 agenda, two weeks ahead of the vote.
• Across Maine more broadly, the May edition of this analysis had already logged eight Maine jurisdictions building this cluster, months before any of August's outcomes landed.

The claim this corpus supports: for each of the August outcomes tracked here, the moratoria and the lawsuit alike, the decision was already sitting in a committee record earlier in the year, for anyone reading municipal agendas closely enough to see it. That is a statement about these cases, not a law of the corpus. A moratorium can surface without a visible run-up, and where that happens we will say so.

The state layer, for balance

Municipal moratoria are not the only thing moving. Pennsylvania is the most aggressive state right now, though not with an outright ban. On August 18, Governor Josh Shapiro signed Executive Order 2026-05, which blocks new data-center permitting statewide unless a developer meets new electricity and water limits, hires locally, and has already secured local government approval; it also pulls AI data centers out of the state's fast-track review and bars developers from using NDAs on these projects (state government sources: pa.gov, reported by The Philadelphia Inquirer and The Washington Post). The legislature has bills moving in parallel: a bipartisan regulatory framework (PUC oversight, clean-energy minimums) already passed the House, and a Senate bill letting municipalities, not the state, impose their own 18-month moratoriums cleared committee in July. Nothing at the state level has passed as an outright ban.

That executive order is this analysis's own thesis, one level up. Pennsylvania just made local government approval a precondition for state permitting, which means the record this piece tracks is now, formally, part of the state's own process too.

Where the welcome mat is out

As restriction rises, state by state and town by town, the harder question for a developer is not only where not to build. It is which town is actively competing to win the project.

• Sand Springs, Oklahoma is running a formal economic-development process for a $1 billion-per-phase, multi-phase data center, structured around three Incentive Districts offering a full, 25-year exemption from property tax. Two public hearings ran in May, and on May 15 the Council approved the project plan, all three districts, and a development agreement, 5-0, with Google's representatives on the record.
• Los Lunas, New Mexico already won one, and is still winning more. The village's 2022 comprehensive plan credits its Local Economic Development Act incentives with helping land Meta's data center, the state's largest, alongside an Amazon warehouse and two solar sites on the same 2,000-acre business park. In March 2026, the Village Council approved a further LEDA agreement 4-0, funding an expansion of the existing data center there from local gross receipts tax revenue.

Same corpus, same method, the other half of the picture: not just which board is about to say no, but which one already said yes, and on what terms.

The welcome-mat pipeline, stage by stage

The leading part of this pipeline runs the same way the moratorium side does: visible in committee records months earlier. These are softer signals than an adopted moratorium. An ordinance is intent, a staff note is a conversation, and only the incentive vote is a binding act; read them in that order.

• Institutional readiness, before any project is named. Hermitage, Pennsylvania drafted a full data-center zoning ordinance, definitions, siting standards, setbacks, environmental review, decommissioning terms, and noticed it for a final vote in July, with no developer named anywhere in the ordinance. A town does not write decommissioning rules for a use case it is not expecting to show up.
• Early-stage talks. McAlester, Oklahoma's city manager reported to Council in June 2025: "New Data Center: Early-stage talks with a potential data center." Many months before Sand Springs and Los Lunas reached their public votes, that language, a staff report flagging a conversation with no name attached yet, was already sitting in a routine monthly update.
• Formal commitment. Sand Springs' incentive districts and Los Lunas' unanimous vote, above, both public, both on the record, months before either reached this piece.

Three towns, three stages of the same read the moratorium side already proves: the record shows the decision before it is news.

One more data point, of a different kind. Stillwater, Oklahoma's March 2026 building-permit report lists a new data-center building among three new commercial permits totaling $293 million that month. That is not a leading signal, it is the opposite: the point where a project stops being a signal and becomes a fact. Its value here is different, it is proof the earlier stages are real. A town that shows institutional readiness, then early talks, then a formal incentive vote, is a town where construction follows.

What's next

Shelbyville, Kentucky's council reached consensus on July 16 to recommend a moratorium, with a first reading set for August 6. Bellefonte, Pennsylvania and Douglas County, Wisconsin are a step earlier, where Utica was in July: the question is on a committee agenda, not yet a vote. Bellefonte's ordinance committee spent July on how the state's pending moratorium bills would apply to a borough with little land left to zone, leaning toward an intermunicipal agreement for hyperscale sites and local conditional-use rules for smaller ones; Douglas County's board chair asked the land-and-water committee to list its environmental concerns before it takes up either a moratorium or a data-center rezoning. Read them the same way, as evidence of what September's outcomes will look like, not as news yet.

The Iowa Specifics: One Ordinance, Many Meetings

A useful clarification on the data shape. Clayton County, Iowa shows up in the pipeline with 47 signal emissions, but those trace to just two distinct source documents: a recurring Planning and Zoning Commission workshop agenda that has appeared across five consecutive months as the commission works through a single proposed data center ordinance. The 47 number is therefore not 47 separate things happening. It is one ongoing policy initiative producing two underlying documents that recur on the pipeline as the meetings continue.

That distinction matters because it is a general property of the corpus: a single sustained policy effort produces many signal-table emissions over time. We are working on clustering these emissions into named initiatives so the headline counts reflect distinct activity rather than signal-table mechanics. The honest version of the Iowa story is not "49 things happened in Clayton County." It is "one ordinance has been actively in development for five months in a single Iowa county, and the work is visible at the document level in real time."

That kind of granular policy-writing visibility is the actual product of the corpus.

The Friction Points

Across the 47 jurisdictions, the discussions converge on the same four pressure points, with regional variation in which one dominates:

Water consumption. Cooling water draw is the single most-cited concern in jurisdictions outside the Northeast. Alpharetta, Georgia's Natural Resources Commission studied tree preservation alongside water impacts. Gunnison County, Colorado raised water usage when reviewing Colorado HB 1030.

Electricity costs and grid capacity. Maine's LD912 (passed) limits the amount of electricity that may be provided to data centers on commercial or industrial sites. Rhode Island's H7331 requires data centers requiring significant electricity to fund the associated infrastructure improvements, rather than socializing the cost to other ratepayers. Maryland's HB1595 grants counties the authority to establish special tax subclasses for data center personal property. The structural question underneath these bills is the same one: who pays for the grid the data center needs.

Land use and zoning ordinance gaps. Existing municipal zoning codes were written before data centers were a contested land use. Clayton County, IA is rewriting from scratch. Hot Springs, AR added data center conditional use permits to its industrial district code. Newton County, GA paused all data center applications to study the impact before proceeding. The pattern is consistent: when existing zoning does not anticipate a use case, the response is some combination of moratorium, conditional use permit, or new ordinance category.

Bundled-moratorium framework. A subset of jurisdictions are addressing data centers inside broader development-moratorium ordinances rather than calling them out specifically. Webster, NY proposed a six-month temporary development moratorium that explicitly named data centers alongside large-scale residential and commercial. This is a cleaner regulatory path than a data-center-specific ban, and it suggests other municipalities will follow.

The State Legislative Wave

State-level bills with current pipeline visibility:

State	Bill	Direction
ME	LD307	Moratorium proposal (House passed Apr 6, Governor vetoed Apr 24, House override failed Apr 29)
ME	LD713	Tax-incentive denial (signed Apr 29), data centers prohibited from claiming state business-development tax incentives
ME	LD912	Electricity caps (passed), limits supply to commercial/industrial sites
MA	H630	Education-to-career data center initiative
MD	HB1595	County tax subclass authority for data center personal property
NH	SB439	Municipal zoning authority over data centers
RI	H7331	Data centers must fund associated infrastructure
SC	H5286	Data center moratorium proposal
SD	HB1038	PUC authority to assess actual costs to data centers (passed committee)
WA	HB2655	Sales tax exemption for data centers east of the Cascades

Ten bills across nine states observed in the current pull, with directions ranging from outright moratorium attempt (Maine, South Carolina) to active incentivization (Washington). The Maine sequence is instructive: a moratorium-shaped bill can fail while a parallel restriction (tax-incentive denial via LD713) plus an executive-order governance framework lands on the same day. The state legislative layer is the lagging indicator. The municipal layer is where the policy gets written first, which is the value of reading the records as they accrue.

What This Means for the Domains the Reader May Care About

We do not make trade recommendations. What we do is surface where signal that reaches your domain is appearing in public documents. A few of the visible threads:

Regional grid capex paths. The states pushing back on data centers (Maine, Vermont, Maryland, Rhode Island) are concentrated in grid regions that are already constrained. The states courting them (Washington east of the Cascades, Idaho, Nevada) have lower marginal cost of power and surplus interconnection capacity. The geographic split that is visible in 47 municipal records implies a corresponding split in where the next decade of compute-driven transmission and substation investment lands.

Data-center REIT and developer geography. A jurisdiction with a moratorium and a jurisdiction with a tax abatement are roughly two years apart in build timelines. The current pipeline visibility on which way a given municipality is leaning is, in many cases, weeks or months before the formal ordinance vote. That informational lead is the value here, not the eventual outcome.

Municipal credit and rate-setting. When a state requires data centers to fund their own infrastructure (Rhode Island's H7331, Maine's LD912), the credit story shifts. The infrastructure cost becomes a contract obligation between the data center operator and the utility or municipality, rather than a public capex burden socialized to existing ratepayers. The credit impact is non-trivial and the language showing up in municipal minutes is specific enough to model.

Utility procurement and interconnection. Pittsfield, MA city council agendas in April carried Eversource grid-interconnection requests in the same packet as data center zoning discussions. That kind of paired visibility does not show up in earnings calls but does show up in a Wednesday-night council meeting agenda.

What Would Falsify This Read

The current thesis is that the data center zoning pattern follows the same shape as cell towers in the 1990s and cannabis dispensaries in the 2010s: a new contested land use category, written into ordinance language one jurisdiction at a time, with eventual state-level codification. The thesis would be falsified by any of four observations:

• Cohort-restricted signal accrual reverts toward zero over a sustained period (more than two months). The current curve is steepening, not flattening, but the within-cohort numbers are small and the next two months will be telling.
• State legislative response stalls. We track 9 bills across 9 states currently; if that count drops while municipal activity continues, the policy-diffusion thesis weakens.
• A major federal preemption event removes municipal authority over data center siting. The current legislative direction, including New Hampshire's SB439, is the opposite: explicitly affirming municipal zoning authority.
• A major municipal document-hosting-platform consolidation event removes our ability to see the pattern as it forms. A small number of private platforms host the public-facing document infrastructure for thousands of municipalities. If one of those platforms changes its architecture, puts documents behind authentication, or otherwise blocks public access, a meaningful portion of the input layer would disappear from view. That risk is concentrated in private tech vendors rather than in the government bodies themselves.

We will continue surfacing this pattern as it evolves.

Methodology and Known Limits

The records counted here are documents flagged by the production pipeline as related to data center regulation: minutes, agendas, ordinance drafts, planning-board memos, state legislative documents, and contagion events across the corpus.

Signal emissions versus distinct documents. Earlier versions of this analysis quoted a "signal count" (58 in March, growing). What we counted there were emissions in the pipeline's signals table, which over-count distinct underlying documents by roughly 3 to 4x because a single document recurring across multiple scans produces multiple emissions. The current piece quotes distinct documents wherever the comparison matters. The 80-document corpus-wide figure is the honest version.

Cohort definition. The "stable cohort" is the set of (municipality, state) pairs with at least one document in the pipeline before March 1, 2026. Restricting the time series to that cohort gives the apples-to-apples view of market activity, controlling for the corpus continuing to onboard new jurisdictions.

Coverage limits. Coverage is weighted toward the Northeast (where the corpus has the most depth) and toward jurisdictions where we have established active coverage. We have not yet run a stratified scan across all 50 states with uniform sampling density. Without that test, claims about systematic geographic skew remain inference rather than evidence. Signal counts in less-covered states should be read as a floor, not a ceiling.

Initiative clustering, not yet implemented. A single ongoing policy initiative (a planning commission working through one ordinance over several months) currently produces multiple records in the pipeline. We are building toward clustering those into named initiatives so headline counts reflect distinct policy work rather than document recurrence. The Clayton County example is called out explicitly in this piece for that reason.

State-legislative outcomes, partial coverage. The pipeline captured Maine's LD307 House vote (April 6) via earlier LegiScan extraction. It did not capture the subsequent Governor's veto (April 24), the failed House override (April 29), or the parallel LD713 signing (April 29) at the time those events happened. LegiScan now serves a Cloudflare challenge to headless server-side requests, which is the platform-consolidation shape we list in the falsification section above. Live status checks need either a JavaScript-executing approach (the ScraperAPI rendering tier or a residential-proxy browser) or the paid LegiScan API. We are building a bill-watchlist mechanism that polls status changes on the subset of bills already flagged through municipal contagion signal, rather than blanket state-legislative tracking. Until that lands, state-level outcomes on bills in this piece should be verified against primary sources (state government, established reporting outlets) at read time.

All quoted document text is sourced. Every link goes to the original municipal, state-government, or news document. Original analysis data current to May 14, 2026; the "August 2026 Update" section is current to August 26, 2026 and is sourced by a full-text corpus search rather than the production pipeline's data-center-zoning tag, so its figures are a different measurement lens on the same pattern, not a continuation of the 80/47/24 count above.

If You'd Like the Underlying Records

We can produce a region-specific or portfolio-specific cut from the underlying corpus on request. State-by-state breakdowns, doc-type drilldowns, and municipality-level chains for any of the 47 jurisdictions are available. Email matt@municipalalpha.com with the region or use case in scope.