top of page
Glossy 3D 48PEGASUS Logo.png

Data Centers, Energy Use, and Local Backlash

Ballots and public response

The ballot box became a siting signal

NOVEMBER 3, 2026 46 of 47 in updated tracking · 40 of 41 in first report · Measures, not national voters

The count that changed

On November 6, Ballotpedia reported that outcomes favored limiting data center development in 40 of the 41 local measures it had identified for the November 3 U.S. midterms. Its expanded scorecard, with an analysis dated November 14, identified 47 measures on the same day and coded 46 as favoring limits. The change is coverage of the election, not evidence that six additional elections happened after November 6. The exception in both snapshots remained Harwood, North Dakota. [10] [11]

Outside the November election, the November 14 scorecard listed six earlier decisions, each coded in favor of limits, and four later measures not yet decided at that cutoff. That yields 52 of 53 decided contests favoring limits; 57 measures were tracked in 13 states across the calendar, including future contests. [11]

Why this matters

Projects that look bankable on power, fiber and capital can still face a local decision about where industrial demand belongs, what conditions apply and which risks hosts will accept. For a developer, that decision can change design, sequencing and cost before electricity demand ever reaches the grid.

The tally is a count of local questions, not a national plebiscite, a poll, a list of canceled projects or a count of votes cast. It gives the same weight to a tiny town’s advisory measure and a large city’s binding charter amendment. Some votes addressed only one parcel or industrial threshold; others addressed wider land-use policy. The right takeaway is not that the United States voted to ban data centers. It is that locally contested proposals repeatedly encountered support for restriction or tighter oversight when the question was put on these tracked ballots. [11]

How the 40 of 41 count was made

METHODOLOGY Code direction, not yes/no · Use dated snapshots · Count one contest per ballot line

Definition of a restrictive outcome

The original November 6 count included 27 successful measures introducing limits or expressing opposition, and 13 unsuccessful measures that would have facilitated or expressed support for development. Both outcomes fall on the same side of a policy-direction code. That does not make an advisory statement legally equivalent to a charter ban. A 'no' vote on an industrial rezoning may also advance the restrictive side without changing broader zoning law. [10]

The November 3 preview identified 28 measures where a yes vote limited development and 13 where yes would facilitate it. With 27 of 28 restrictive propositions passing and all 13 facilitative propositions failing, the numerator is 27 + 13 = 40. The one remaining case was defeat of a restrictive initiative. This reproduces the arithmetic of the original snapshot; it does not establish that every possible local contest was captured. [12] [10]

Definition of an observable result

A reproducible ledger should store the date, jurisdiction, exact ballot text, direction of a yes vote, whether the measure binds, official results or reporting snapshot, and caveats. Archive links and access dates when possible; do not silently substitute an updated table for an earlier version cited in the narrative.

The denominator is contests, and the unit is a ballot question, not a facility or a jurisdiction. Four separate Wilmington zoning measures do not represent four independent towns deciding comparable policies. Ballot selection also favors places with enough controversy, petitioning capacity or institutional access to put an issue before voters. Public meetings, negotiated changes and projects approved administratively sit outside this denominator. The count cannot be extrapolated as an industry-wide cancellation rate or interpreted as a binomial sample of U.S. opinion. [11]

Audit of the expanded 47-question count

ELECTION RECONCILIATION 33 restrictive yes outcomes · 13 facilitative no outcomes · 1 restrictive no outcome

Break down the first half

The scorecard identifies 33 passed restrictions or opposition measures: five full bans, 13 partial bans, seven six-month moratoriums, two applications of zoning rules, three nonbinding opposition measures, two voter-approval requirements and one financial-guarantee requirement. These add to 33. Some partial bans preserve narrow exceptions; not every rule uses the same power threshold. [11]

The expanded breakdown has six more restrictive approvals than the November 6 write-up: one additional partial ban, four additional moratoriums and one additional advisory opposition measure at the category level. The tally of failed facilitative measures remains 13. These comparisons establish the change in reported coverage. Without a retained contest-by-contest copy of the first table, they do not establish every mapping between versions. [10] [11]

Break down the second half

The 13 failed facilitative measures comprise seven advisory support questions, four Wilmington rezoning questions, one Boulder City approved-use question and one Ashville term-summary referendum. The 47th contest, Harwood’s failed additional-review initiative, is the one development-favoring result by the scorecard’s coding. Neither a defeat nor an advisory vote should be translated into an awarded construction permit. [11]

There are different kinds of official evidence: the election authority supplies the yes and no counts; the ballot and adopted text establish the policy change; later records show implementation. The nationwide tally is attributed to the original tracker, with selected counts independently checked where available. It is not presented as a separate canvass of all 47. A concise reporting form is: “46 of 47 outcomes in Ballotpedia’s expanded Nov. 14 accounting of Nov. 3 local data center measures favored limits or expressed opposition.” [11]

What local measures actually govern

LEGAL MECHANISMS Ban ≠ moratorium · Referendum ≠ permit · Advisory ≠ binding

Four categories of constraint

A full ban generally removes a defined use within the jurisdiction, subject to exact definitions, exemptions, effective dates and applicable law. A partial ban may exclude only facilities above a load threshold or include a narrow carveout. A moratorium delays approvals or applications under the written terms; whether it applies to pending or previously approved work depends on the instrument. Additional approvals preserve a possible route while adding cost, process or calendar risk.

The ballot language matters more than a label. One measure can regulate storage and data center uses together, whereas another amends broad industrial policy and may name data centers only as one example. Terms like “capacity,” “monthly demand” and “peak load” should not be treated as interchangeable when determining whether a proposed campus qualifies. [11] [17]

Match legal effect to the text

Parcel-specific rezoning can fail without prohibiting the same use on already entitled land. A vote against a development term summary may change negotiations without independently invalidating every separate prerequisite. A community-consent requirement may add a later vote rather than denying a particular proposal immediately. Advisory votes document a political instruction or preference but may leave formal authority elsewhere.

An analyst should retain separate fields for the public outcome, immediate legal action, practical project effect and outstanding approvals. Follow adoption, effective date, potential litigation, amendments, implementing rules and the next docket entry. The correct status after voting may be conditional, delayed, redirected or unchanged rather than permanently canceled. This distinction protects elected officials, residents, lenders and readers from treating a headline as a complete statement of power, permit or land rights. [11]

Harwood: the exception did not approve a campus

NORTH DAKOTA Additional-review initiative failed · 105 yes / 326 no in tracker · About 76% against

What voters considered

Harwood was the only November 3 result coded as development-favoring in both Ballotpedia releases. The tracked count of 105 yes and 326 no corresponds to 75.64% against among 431 reported yes/no votes. The initiative would have added review before city approval of certain industrial development, including data centers. Voters declined that added requirement. The question did not itself authorize a particular campus. [11] [13]

North Dakota’s Secretary of State election portal distinguished night reporting from official canvassing. The Harwood totals used here are the figures in the published tracker; they are not represented as independently certified by this paper. State results sites can contain multiple similarly named questions, so the analyst should match the home-rule amendment to the actual additional-review text before transcribing. [14]

What the result does not tell us

Some voters may have opposed an extra layer of procedure but also opposed a specific development. Others may have favored economic growth, trusted the existing process or rejected a detail in the proposed language. The vote does not reveal which motivation applied, whether views differ by party, or whether another measure with different safeguards would have passed.

For a pipeline forecast, the right status change is narrower: remove the proposed review requirement as an expected hurdle in this jurisdiction unless a future action revives it. Keep outstanding land-use, utility, water and public approvals on separate lines. This is also why describing Harwood as “the one community that voted to build” misstates the evidence; it was the lone outcome favoring development under the tracker’s classification, not a direct affirmative vote to issue approval. [11] [13]

Ohio: the count was geographically concentrated

OHIO State concentration matters · Screen charter provisions · Do not double-count projects

Multiple policy choices in one state

Ballotpedia’s original election-day preview identified 21 Ohio questions among its 41 then-tracked contests. This concentration matters for interpretation: neither the original nor expanded count is a geographically balanced U.S. sample. One instrument cannot be used to describe the entire Ohio set. Illustrative choices ranged from local bans to applying zoning, refusing a term summary, rejecting rezoning and adding guarantees for developers seeking incentives. [12] [11]

Piqua appeared in the general election’s issue results; Miami County separately lists a Piqua special election. The two files should not be conflated. Ballotpedia’s issue page lists 4,118 yes and 2,066 no for the charter initiative. The measure defines covered facilities by aggregate monthly demand or peak load greater than 25 MW, so the summary label “prohibit data center construction” alone is incomplete. [19] [18]

Four votes, one place

Wilmington supplies a clear denominator example. Four measures separately addressed specific rezoning proposals that could allow data center construction. Each failed and each counts in the scorecard. Four failures are four ballot decisions but do not establish four independent labor markets, four distinct operators or four canceled campus projects. Track parcel and project identifiers separately. [11]

Conversely, Grove City introduced a decision gate rather than adopting a general data center ban, while Perry targeted the financial risk of certain incentives. An Ohio comparison should normalize the measure types before comparing margins or estimating implications. The useful output is a register of restrictions by boundary and legal mechanism, not a single statewide vote share. Interpret the possible flow of investment as contingent: a different parcel might still be eligible, yet it may require different power or network infrastructure and a fresh approvals timeline. [11] [17]

Grove City: development consent, not prohibition

OHIO CASE STUDY Issue 28 · 11,317 yes / 9,411 no · 54.60% in Nov. 4 county report

What the county reported

Franklin County’s November 4 unofficial return gave the Issue 28 charter amendment 11,317 yes and 9,411 no votes, or 54.60% yes among votes recorded for or against the issue. All 30 applicable precincts were shown reporting; the report also listed 638 undervotes and two overvotes. Reporting completeness is not identical to certification, and the denominator for vote share excludes the over- and undervotes. [15]

Grove City’s public information and ballot records describe an additional citywide approval requirement for covered developments, expansions or transfers. The thresholds were at least 50 contiguous acres, more than 20 MW of peak electrical demand, or more than 500,000 gallons of water per day. Exact text controls; a summary should not replace it. [17] [16]

What changes in the model

This kind of trigger adds an official decision after ordinary technical evaluation and may also affect other large developments. It is not a ban on all computing facilities, and it does not prove approval for projects that fall below one threshold; other triggers and permitting rules may still apply. The city had separately announced a one-year data center moratorium approved by council in June, a distinct instrument rather than the November result. [17]

For underwriting, anchor the project timeline to the applicable future election and preserve independent entries for zoning, interconnection, construction permits and community-impact standards. Test whether proposed phasing changes the measured peak or only changes the marketing description; authorization should hinge on the actual scope and controlling law. For public reporting, compare the likely burden with and without the new vote while being clear about which residents, uses and districts are covered by the charter text.

Two narrow votes: Ashville and Perry

DEVELOPMENT TERMS Ashville: no on upholding summary · Perry: yes on guarantees · Both coded restrictive

Ashville: negotiations were the target

The Ashville referendum asked whether to uphold a term summary guiding negotiations between the village and EdgeConneX for two data centers. Ballotpedia reports 331 votes in favor and 1,469 against. A no vote therefore registers as a restrictive outcome. The wording matters: the vote was about a particular negotiated instrument, not a referendum on all computing facilities or a completed construction contract. [11]

To assess the development implications, identify which provisions depended on that summary and which obligations or rights already existed under separate documents. Do not assume that an instrument guiding future negotiations guarantees an economic benefit, or that a vote against it automatically voids every land and utility arrangement.

Perry: the incentive was the leverage point

Perry’s Issue 16 required financial guarantees from hyperscale developers seeking certain incentives, according to the measure register. The tracked result was 946 yes to 140 no. The difference in vote directions is the simplest demonstration of why one cannot sum “no” votes as opposition: Ashville’s restrictive outcome came through a no vote, while Perry’s came through a yes vote. [11]

A financial guarantee can make certain obligations collectible if a developer falls short, but its value depends on the amount, trigger, issuer, beneficiary and expiration. It is not evidence on its own that no local public costs remain. Evaluate whether security covers stranded infrastructure, promised payments, road restoration or other defined exposures. A useful public reporting template identifies who can call the guarantee and when, what performance triggers release, and how much exposure would remain after collection.

Maine: communities chose time to write rules

MAINE Seven six-month moratoriums in tracker · Wells Article 5 · Temporary pause, not ban

A policy deadline, not permanent rejection

Seven Maine measures in the expanded November 3 count adopted six-month moratoriums. That cluster often gets summarized as seven communities saying no to data centers, but the specific policy was a pause. A moratorium changes the time during which a defined application may be considered or approved. It creates an interval for rulemaking and study, not proof of which ordinance will replace it. [11]

In Wells, the special town meeting warrant set out a retroactive moratorium rather than simply asking whether residents supported data centers. The town clerk certified 5,818 yes, 1,282 no and 459 blank on Article 5, confirming the tracker’s count. The warrant is the authority for the measure’s scope and timing. [20] [21] [11]

What good use of a pause looks like

A practical pause should have a defined work plan. Assign responsibility for reviewing noise and emergency equipment, quantifying water and power requirements, analyzing compatible land uses, and proposing measurable standards with enforcement. Publish a realistic date for public review and a plan for unfinished work. If the temporary policy has a retroactive reach, identify how previously filed applications will be handled instead of telling readers all work stopped.

For investors, track the expiration and the next policy hearing as separate schedule events. A facility can remain an active opportunity when construction is not immediately permitted. For residents, an end date is valuable only if it produces readable rules on the exposures that prompted the pause. A policy can be time-limited yet still materially affect financing, procurement and project sequencing. Do not turn temporary votes into permanent losses of electricity demand in a forecast.

Boulder City: the geography of authority

NEVADA Question 1 · 2,165 yes / 5,227 no in tracker · Specified area only

Read the boundary explicitly

Boulder City’s published Question 1 asked whether data center facilities should be an approved land use in the Eldorado Valley Transfer Area outside the specified habitat conservation easement. Ballotpedia recorded 2,165 yes and 5,227 no, roughly 71% opposed among yes/no votes. The area named in the question determines the decision’s reach. A rejection there cannot be described as a citywide ban on every possible project. [22] [11]

A land-use decision can affect one application pathway while leaving others unresolved. Municipal, county, state and federal decisions may involve different territories and records. The same metropolitan project name can be used for separate parcels or proposals. When mapping project exposure, treat parcel, controlling title or easement, jurisdiction, permitting agency and stage as distinct fields instead of assigning a single regional status.

A test for a misleading schedule

Ask whether an alternative site can actually substitute the same service date. A new parcel may need a different substation, transmission upgrade, cooling design, water agreement or environmental review. An asset that is portable on a national map may not be portable as a contracted power load or operating campus. If the developer has only an option on alternative land, record that option separately from a deliverable plan.

Do not infer the outcome of other nearby proceedings from the city ballot. Track them independently and report which body can decide each question. A reader should be able to distinguish three claims: local voters opposed an approved-use designation at the defined site; a separate project needed another government’s action; and actual operating capacity will depend on completion of its own approvals. Only the first follows from Question 1 itself. [22]

Illinois: why advisory votes count but do not bind

ILLINOIS Different question wording · Advisory results are directional · Check implementing action

Not every yes vote meant the same thing

The expanded Ballotpedia table records Bloomington’s advisory large-scale moratorium measure with 22,119 yes and 6,376 no. In Joliet, an advisory question asking voters whether the city should approve AI data center construction instead received 3,883 yes and 14,599 no. Each result reflected restriction or opposition in the tracker, but neither was itself a binding land-use ordinance. [11]

Other Illinois advisory votes appeared in the same count, including measures in Madison County townships and Morgan County. Separate maps and district boundaries are essential; a township’s advisory result is not necessarily the result for all residents of an overlapping municipality or county. Broad headline labels can obscure who was entitled to cast a vote and which officials received the advice.

Evidence of policy preference, not completion

An advisory result can be an early warning for council actions long before a final ordinance is enacted. It may change political incentives, prompt a moratorium or motivate a developer to redesign, withdraw or defer. None of those outcomes should be tracked as a fact until a decision, filing or company disclosure demonstrates it.

A useful issue register records the deadline for a response, if any; the authority to enact subsequent policy; whether planning staff released recommendations; and the date of the next hearing or council vote. Good governance does not require treating the question as if it already changed law, nor should it mean ignoring the result. A responsible executive can acknowledge the preference, document the specific burdens residents raised, and explain the next actual decision where revisions can still occur. [11]

Public opinion: three instruments, different questions

NATIONAL POLLS Pew: comfort · [*REDACTED*]: opposition · Quinnipiac: opposition · Do not pool the results

Use the right denominator

[*REDACTED*]’s March 2–18 survey found 71% opposition to constructing an AI data center locally, including 48% strongly opposed. Quinnipiac’s September 24–27 poll of 1,202 adults found 72% opposed, 21% supportive and 7% without an opinion about building an AI data center in their community. Its March release reported 65%, 24% and 11% respectively. Quinnipiac reported an overall September sampling margin of ±3.5 percentage points including the design effect. [25] [24]

Pew’s July 20–August 9 survey asked about comfort with a new local data center. Sixty percent said they would be not too or not at all comfortable; the question and response categories differ from the other polls. Annenberg’s June 16–July 19 panel found 61% opposed to new area data centers, compared with 49% in its spring wave. [23] [26]

Agreement without false precision

These independently designed surveys show widespread national concern about local development. They do not establish that a particular proposal would lose at the same rate. Question framing, locally negotiated conditions, response options, sample design and field dates vary. An average of the four rates would be numerically possible but substantively weak, and neither a [*REDACTED*]–Pew difference nor a cross-poll month-by-month line is a clean trend.

Quinnipiac’s September release contains a separate 73% result about concern over AI’s potential threat to human survival. The local data-center opposition result is 72%, not 73%. Sources should be read by question and figure, rather than relying on a headline centered on a different topic. Large-sample precision also cannot eliminate framing or survey nonresponse limitations. [24]

Party: broad concern does not mean equal intensity

POLITICAL INTERPRETATION 54% Republicans / leaners uncomfortable · 68% Democrats / leaners uncomfortable · Survey association, not causation

What the subgroup data show

Pew’s summer scenario found that 54% of Republicans and Republican-leaning independents and 68% of Democrats and Democratic-leaning independents were not too or not at all comfortable with a new local data center. [*REDACTED*] in March also found majority opposition across party groups, with strong opposition more common among Democrats (56%) than Republicans (39%), and independents at 48%. These are distinct samples and questions, but they independently support breadth of concern with unequal intensity. [23] [25]

Pew found majority discomfort in urban, suburban and rural groups. It also reported more negative assessments in both party groups since January on several effects. That is a reason to test assumptions about acceptance in every market, not a reason to label all residents in a given partisan or geographic category as opponents. [23]

What election returns cannot show

A ballot result cannot identify who voted no by party or why. Even comparing precinct returns with congressional votes cannot establish that the same voters supported restrictions and switched candidates; that would require suitable individual-level evidence, and selection and issue-salience problems would remain. A candidate who runs on utility affordability may also address inflation or unrelated local concerns; winning does not isolate one election driver.

Publish a party comparison as survey evidence about reported attitudes, not an assertion about turnout, campaign effects or the national midterm outcome. If a report codes elected officials’ positions, archive campaign statements with dates and separate strict prohibition, utility cost protection, site-specific opposition and conditional support. Candidate positions belong in a different table from voter returns. Neither table should be filled from partisan stereotypes.

Why residents object, and why some support projects

PUBLIC MOTIVES Open-ended responses · Cost, land and trust · Projects may differ materially

Concerns are not interchangeable

[*REDACTED*]’s April open-ended follow-up found that local opponents often pointed to resource use, with water and energy each mentioned by 18% of opponents; other concerns included pollution, quality of life and economic costs. Among supporters, roughly two-thirds cited economic benefits and 55% mentioned job opportunities. The denominators are opponents and supporters in the follow-up, not all U.S. adults. These are reported reasons, not direct measurements of environmental or economic effects. [25]

Pew’s data show increasing awareness as well as more negative expectations across several surveyed effects. Awareness can travel through announcements, news and opposition efforts; a rising survey rate does not identify which channel changed individual opinions. Public questions about jobs, tax revenue or electricity bills often call for different evidence than questions about changing the character of a place. [23]

Resolve facts before debating values

The engineering assessment should describe peak draw, annual consumption, cooling system, emergency generation and construction schedule. The economic assessment should specify who gets which fiscal and employment benefits and when. The public record should then separately address values that numbers cannot eliminate: acceptable land-use change, community autonomy, privacy and perceived fairness in who absorbs burdens.

Do not treat acknowledgement of concerns as proof that a predicted impact will occur. A specific facility might use substantially less direct water than another design, shift noise with different setbacks or pay for certain upgrades. Equally, a generic promise of sustainability does not answer whether an individual utility or neighborhood would face a cumulative burden. Credible engagement starts by showing where there is evidence, what remains contested, and what decisions residents can still influence. [25] [23]

The missing denominator: opposition outside elections

SELECTION BIAS Petitions are not ballots · Approvals can occur off-ballot · Governance limits vary

What the election ledger excludes

Citizens can oppose a project through hearings, administrative challenges, public records, petitions, initiative campaigns or other mechanisms. Some petitions never reach a ballot because of signatures, deadlines or legal authority. Ballotpedia’s tracker separately catalogs proposed measures that did not qualify, including disputes in jurisdictions not represented in the election results. Such cases are not additional “no votes.” They can still alter the project’s expected timing and risk. [11]

Similarly, approvals made administratively, projects redesigned before a vote, and facilities built in industrial districts without a new ballot question are not observations in the November count. Nor is a withdrawn proposal with unresolved financing necessarily canceled because of backlash. Attribution requires a record connecting the withdrawal to the cause.

Compare governance, not just headlines

A full pipeline ledger should begin with proposed developments, not elections. Record a date and source for announcement, application, permit decision, formal challenge, pause, redesign, capital commitment and operation. If the project has no verified public name or location, label it unresolved rather than guessing. Maintain a separate inventory of jurisdiction rules and an event log for petitions and hearings.

This structure supports two different metrics: the share of tracked ballot questions favoring limits and the share of contemporaneous projects facing documented changes. The second needs a defensible project denominator and status definitions. Depending on the question, counts can be weighted by committed or expected MW, but that creates a different measure and adds uncertainty. The 46-of-47 finding cannot serve as that denominator. It belongs as a strong data point within the particular institutional path in which it was observed. [11]

From engagement to enforceable conditions

COMMUNITY GOVERNANCE Publish early · Name who bears costs · Define compliance and remedies

Start before the site is irreversible

Publish a dated project description and a nontechnical summary while residents and agencies can still change material features. Identify land parcels, physical scale, construction windows, operating phase, power arrangements, cooling concept and emergency equipment. Separate target values from contractually binding commitments. For proprietary details, explain why they cannot be made public and provide independently reviewable ranges where appropriate rather than leaving the relevant burden unknown.

Create a comment-response register by topic. Permit several participation channels, distinguish questions from formal objections, and avoid claiming that turnout represents all voters. A process can be transparent without guaranteeing agreement. The governance objective is informed consent where the law requires it and credible accountability whenever substantial local impacts are possible.

Translate the pledge into a condition

For each material term specify a metric, limit, measurement method, responsible party, reporting cadence, verification source and remedy. For example, a proposed particulate or noise requirement needs a boundary, measurement conditions and exception policy; a water use commitment needs its withdrawal and consumption definitions and treatment of seasonal peaks. Identify who pays for monitoring, who investigates noncompliance and whether a commitment attaches to a successor operator.

Use enforceable instruments for obligations that matter to underwriting or public affordability. A public summary should identify the governing agreement or permit, the authority empowered to enforce it, and the limits of that authority. Community benefits funded only under one scenario should not be represented as guaranteed in all scenarios. This checklist is a recommended framework, not a statement that any jurisdiction has adopted the same requirements.

Local benefit models need downside cases

FISCAL DILIGENCE Net revenues, not press releases · Jobs by phase · Model downside and rebound

Make the headline numbers comparable

Construction employment should be reported as labor-years, duration and peak headcount where available. Operating employment belongs in a separate row with expected full-time positions and an estimated start date. An operator’s payroll figure is not the same as local wage income unless worker residence and subcontracting are known. Jobs announced for the last construction phase should not be attributed to the first facility.

Compute net effects for each government entity. Include tax abatements, payment-in-lieu terms, credits, agreements to reimburse infrastructure, expected public service costs, and which district collects each dollar. An increase in total assessed value does not necessarily increase funding for every entity if allocation rules or incentives redirect revenue.

Test the project that never reaches plan

Model phased, delayed and canceled cases. Ask who paid for road, substation or water capacity that is now unused and whether deposits, collateral or minimum bills cover it. Distinguish private ownership of equipment from pledged municipal obligations and explain which commitments actually trigger. A published bond authorization cap is not necessarily debt issued, taxpayer liability, investment already spent or final assessed value.

Set an explicit time horizon, phase schedule and discount assumptions. Show the no-project counterfactual and plausible alternative uses of the site. An alternative use may bring less electricity demand but more durable employment, or impose its own public costs; this is a location-specific question, not a predetermined finding. The result is an auditable framework for comparing exposures. It prevents a large nominal total from doing the work of a credible local fiscal analysis.

Upcoming votes: different measures, same discipline

AS OF NOVEMBER 19, 2026 Dec. 1 Clarksville x2 · Dec. 8 Hollis · Dec. 15 Mandan · Schedules may change

Four pending questions in the tracker

The November 14 Ballotpedia scorecard lists two December 1 Clarksville, Arkansas referenda, each about a distinct industrial development bond ordinance and related leasing and payment-in-lieu arrangements. The ceilings are up to $50 billion for [*REDACTED*] and up to $5 billion for SF ARK1. They are caps in the described authorizations, not evidence that $55 billion has been spent or borrowed. Read the separate provisions before adding figures. [11]

The same tracker lists a December 8 Hollis, Maine measure to establish a 180-day moratorium and a December 15 Mandan, North Dakota measure addressing data centers with anticipated capacity or demand of at least 25 MW. These were pending as of the cutoff. The key facts to monitor are the official elections calendar, final ballot wording, results and separate implementation records. [11]

How to avoid an erroneous update

Do not append these four contests to the numerator or denominator for November 3, and do not label them “defeated” based on the preceding trend. For each event, start with its power to change the project. One might affect a bond arrangement without directly authorizing construction; another may temporarily close an application window; the third may remove qualifying projects as a permitted use under the actual legal instrument.

Use separate time stamps for scheduled date, reported votes, canvass, effective date and implementation. If the results fail to answer whether a developer will proceed, preserve the gap in the record. A dated forward watchlist is different from a prediction. The report’s purpose is to say which decision would change which dependency, not to substitute a forecast of residents’ preferences or to tie national electricity growth to the result of one pending vote.

Decision protocol: how election risk reaches a project

SITING RISK CHECK Confirm authority · Identify decision gate · Model delay and relocation · Verify actual outcomes

Read the sequence

First identify the scope: current proposal, usable parcels, all unsettled approvals, largest requested utility commitments and phase schedule. Then link each election or policy to a dependency: zoning, ballot consent, financial incentive, utility rate protection, or a temporary delay. Record whether the dependency is necessary for the current concept or can be substituted. A restriction might move investment out of one neighborhood without immediately changing national compute demand or might slow demand until a different site is built.

Assign a scenario to each credible result. If a vote is advisory, the next decision lies elsewhere and its schedule belongs in the model. If the measure is binding and applies, identify the earliest verified effective date and alternative pathway. Include potential redesign, smaller phasing, revised guarantees and withdrawal, without assuming any one response is the developer’s plan. [11]

Treat public acceptance as a measurable operational issue

Set triggers for reforecasting: a certified vote, permit denial, withdrawal, adopted ordinance, signed infrastructure agreement or verified phase revision. Track public complaints and formal conditions separately; a surge in public engagement is informative before it changes legal rights. Compare the implied local outcome with actual land, construction and utility commitments.

In the November dataset, the central finding is robust to the tracker’s expanded coverage, but its interpretation remains bounded. It demonstrates organized local political resistance in a self-selected set of contests, not a national ban or a deterministic election effect. It creates a case for quantified terms, independent checks, honest polling interpretation and distinct project-stage reporting. If uncertainty remains after the vote, the responsible conclusion is to name the missing approval or evidence, not to force a binary “built” or “blocked” verdict. [10] [11]

Exhibit 03 · The tracked tally expanded

The Nov. 6 initial snapshot reported 40 of 41. The later November 3 set is 46 of 47; the year-to-date count is 52 of 53 as of Nov. 14. These are overlapping snapshots, not three independent election samples or estimates of national opinion.

Ballotpedia, 2026 local data-center measure scorecard and November 3 analysis: primary source

Exhibit 04 · Approved limits were not all bans

These 33 approvals are distinct from 13 defeats of pro-development questions; both contributed to the 46 restrictive or opposing outcomes in the expanded November 3 tally.

Ballotpedia, 2026 local data-center measure scorecard and November 3 analysis: primary source

Exhibit 05 · From ballot result to legal effect

Recover the actual question and canvass, classify its direction and authority, and verify what changed afterward. An advisory vote is not a permit denial.

48PEGASUS Research original conceptual exhibit. Not empirical data.

Exhibit 06 · Local opposition crossed party lines

September 24–27, 2026 national survey of 1,032 self-identified registered voters. Overall sampling margin ±3.8 percentage points; party subsamples have greater uncertainty. Question asked about building an AI data center in the respondent’s community, not a specific proposal or nationwide ban. Remaining share is calculated from rounded reported percentages.

Quinnipiac University registered-voter poll, September 29, 2026: primary source

Previous chapter

Next chapter

Back to the report

Browse: White Papers · Reports · Downloads

bottom of page