takeoff.watch

Voters approve a ban or pause on frontier AI or new datacentres

Will voters in any country or US state approve a ballot measure that bans or pauses frontier AI development or the construction of new large datacentres?

Chance by Q4 2031

22%

Cumulative probability

0%25%50%75%100%
  • Latest forecast

Forecast history

Forecast made on 17 Oct 2026

Latest of 4 forecasts · Methodology v2.1

Model reasoning

Forecast made 17 Oct 2026 · Methodology v2.1

No statewide or national vote on banning data centres or advanced AI is scheduled this year; the dozens of such votes are all local and don't count. The main hope is a volunteer Ohio petition to ban large data centres, which has about a third of the signatures it needs and will probably miss the 2027 ballot. Other states could try in 2028 or 2030. A well-funded signature drive in Ohio or elsewhere would most raise the odds.

Where things stand (17 Oct 2026)

  • No qualifying measure is on any 2026 ballot. Ballotpedia's data-centre index counts 49 measures in 2026, and every one is local (municipal, township or county), so none can count. Ballotpedia's list of 2026 statewide measures has nothing on data centres or AI. No national referendum on either topic is scheduled anywhere I could find (I searched for Swiss popular initiatives and for Latin American and European referendums).
  • The only qualifying statewide effort is Ohio's "Prohibition of Construction of a Data Center" constitutional amendment. It would add Art. II §36a: "The construction of a data center is hereby prohibited," for facilities with more than 25 MW peak load, and it is self-executing. That clearly meets the question's definition. The Attorney General certified it on 26 Mar 2026, the Ballot Board cleared it in April, and it now targets 2 Nov 2027. The group behind it, Conserve Ohio, is all-volunteer (about 1,000 volunteers). It missed the 2026 deadline.
    • Signatures needed: 10% of the 2026 governor's-race vote, which is at least about 413k valid signatures, with a minimum of 5% of that vote in each of 44 counties. The group itself said it aims for about 600k raw signatures. The filing deadline is 30 Jun 2027.
    • Signatures collected: about 25k on 15 May, 70k on 17 Jun, 130k on 8 Sep and about 140k on 2 Oct.
    • Pace (my calculation from those counts): about 4.9k a week (Apr–May), 9.5k (May–Jun), 5.1k (Jun–Sep) and 2.9k (Sep), for a lifetime average of about 5.6k a week.
    • What it would take: making 2027 needs about 12–13k a week sustained. Making the 2028 deadline (5 Jul 2028) needs about 5–5.5k a week for 92 weeks, while older signatures go stale and legislators talk about limiting how long signatures stay valid.
    • Timing if it passes: Ohio certified the November 2023 results on 7 Dec 2023, so a November vote would resolve in Q4.
  • Other measures that don't count:
    • Massachusetts' 2028 data-centre initiative only sets conditions plus a two-thirds local vote, which is an excluded category.
    • California initiative 25-0034A1 is inactive, and it would only have created a regulator with discretionary power.
    • Maine's statutory moratorium (LD 307) was vetoed. New York's moratorium and Texas's permit halt came from legislators and governors, with no vote of the people.
  • Market context: Polymarket's "Ohio enacts data center moratorium" contract trades at 15.5% by June 2027, 20% by end-2027 and 26.5% by end-2028. Most of that is the legislative route, which does not count here. The small rise across late 2027 fits a ballot-route chance of only a few percent.

Reference class and adjustments

Past statewide votes to ban a class of energy or industrial development after a local backlash:

  • They usually fail against heavy industry spending: anti-nuclear initiatives in 1976 (all failed) and Colorado's 2018 fracking-setback measure Prop 112 (43%).
  • But some pass once a backlash is broad and lasting: Montana I-80 (1978), Oregon's 1980 nuclear moratorium, Montana I-137's 1998 cyanide-mining ban, and Ohio's 2006 smoking ban.

Opinion today is unusually hostile:

  • Bowling Green State University poll: 63% of Ohioans oppose data centres anywhere in Ohio.
  • Gallup and Fox polls: about 70% of Americans oppose one in their area.
  • Against that, 52% nationally (Fox) back building them in remote places, and opposition is starting to split along party lines.

I put the chance of passing, if on the ballot, at about 40%.

The model (hazards computed in Python, then multiplied through)

  • Q4 2027: Ohio qualifies about 5%, times 0.40 passing, plus about 0.4% from other routes. Quarterly hazard 2.4%.
  • Q4 2028: Ohio reaches the ballot about 13% (including a possible switch to the cheaper statute route) × 0.40; other initiative states such as Michigan, Arizona, Montana, Nebraska, Oregon and Maine about 5%; a frontier-AI or superintelligence-ban initiative about 1%; voters keeping a legislated moratorium in a referendum about 0.7%. Hazard 11.5%.
  • Q4 2029 and Q4 2031: about 1% each. Q4 2030: about 5.7%. Other quarters: 0.1–0.25%.
  • Cumulative: about 2.8% by end-2027, 14% by end-2028 and 22% by end-2031.

Strongest case against my estimate

The signature drive is losing pace and has no paid circulators. Industry and building-trades unions (who have formed a coalition with OpenAI, SoftBank and Blackstone) would spend without limit. Legislatures and governors are absorbing the backlash with their own moratoria and rules. An AI investment slowdown or federal preemption could remove the issue's urgency. All of this argues for something nearer 10%.

The case for higher: opposition crosses party lines, the Ohio signatures never expire, there are about 25 initiative states and three general elections in the window, and temporary moratoria and "pause until a condition is met" measures both count. I land at about 20–22%.

How I read the question: conditional-permit or local-vote measures (like Massachusetts') don't count; Ohio's ban would; the date that matters is the December certification.

Sources

Resolution criteria

Opened 16 Sep 2026Not yet resolved

Question. By the last day of each quarter, have voters in any country or US state approved a ballot measure that bans or pauses frontier AI development or the construction of new large datacentres?

Definitions.

  • Country or US state: a national electorate or the statewide electorate of one of the 50 US states. Municipal, county, district, territorial, and non-US subnational votes do not count.
  • Ballot measure: a legally binding initiative, referendum, or constitutional amendment submitted to that electorate. Approval means that the competent electoral authority has certified that all applicable passage requirements were met, including any turnout or supermajority requirements.
  • Bans or pauses: the adopted text itself imposes a prohibition or moratorium, permanently, for a specified positive duration, or until a stated condition is met. It must apply throughout the country or state to a class of development or construction, rather than only to a named project or site. A future effective date counts; the prohibition need not yet have taken effect.
  • Frontier AI development: training or otherwise developing advanced general-purpose AI models covered by the measure's explicit frontier/advanced-model definition or a training-compute or general-capability threshold. A prohibition on all AI development also counts. Restrictions only on deployment, particular applications, or use do not count. The developer need not be a covered developer.
  • New large datacentres: new facilities or integrated campuses primarily housing computing, storage, and networking equipment, with planned aggregate electrical demand or peak load exceeding 25 MW. A general datacentre construction ban includes large datacentres and counts without a size threshold in the text. A ban covering only facilities above a higher size threshold also counts. AI use or an AI-related motivation is not required.

Resolves YES if certified approval and the adopted measure's official text establish either qualifying prohibition. Approval of a binding referendum retaining an existing qualifying prohibition counts. Later repeal, judicial invalidation, or failure to implement the measure does not reverse YES.

Exclusions (resolve NO).

  • Petition circulation, signature certification, ballot qualification, unofficial returns, projections, or legislative passage without voter approval.
  • Advisory votes, opinion polls, or measures merely requesting later legislation.
  • Measures that only establish a regulator, reporting obligations, licensing or safety-review procedures, taxes, or discretionary power to halt development; ordinary compliance requirements are not themselves a moratorium. An express interim moratorium pending a review does count.
  • Requirements for further local voter approval, withdrawal of subsidies, individual project or rezoning decisions, and bans limited to expansions of existing facilities.
  • Local votes, including Monterey Park's June 2026 Measure NDC, irrespective of their substantive prohibition.

Resolving source. The competent national electoral commission, referendum authority, US state secretary of state, or state canvassing board's published certified results, together with the official adopted text published by the electoral authority or official gazette. Ballotpedia's data-centre ballot-measure index may locate measures but cannot replace official certification. Relevant examples of official measure documentation include the Ohio Attorney General's proposed construction-ban text and the California Secretary of State's description of initiative 25-0034A1; proposal documents alone do not resolve YES. This question uses these official sources rather than Public confirmation (A2). Apply the Resolution window and Resolution date (A3): the qualifying publication is the certification of passage, not election day or a later republication of an old certification.

Download all forecasts (JSON)