US Voters Rejected AI Data Centers in Nearly Every November Ballot Measure

Last updated: 6 November 2026. Figures below are sourced to the primary ballot trackers and reporting linked inline; where trackers disagree on the exact count, that is flagged explicitly rather than smoothed over.

The 60-second version

• In the 3 November US midterms, voters in dozens of local and state contests rejected AI data centre expansion almost everywhere it was on the ballot — trackers put the margin between 37-to-1 and 46-to-1 depending on what is counted, with “43-1” the figure most widely repeated in headlines.
• Ohio had the most measures of any state (at least 21), and voters there also recalled a mayor and city commissioners over a Meta project, while electing Amy Acton as governor.
• In Texas, Democrat Gina Hinojosa unseated incumbent Governor Greg Abbott after campaigning on a statewide data-centre moratorium — ending more than three decades of Republican control of that office.
• Amazon, Google and Microsoft had already started offering concessions — ending non-disclosure agreements with local officials, covering power-grid costs, paying full property taxes — weeks before the vote, suggesting the industry saw this coming.
• NBC exit polling found a majority of voters called AI and data centres a “crucial or major” factor in their congressional vote, and roughly four in five opposed a data centre in their own neighbourhood.

Key numbers, by tracker

What actually happened on 3 November

Ohio was the epicentre: at least 21 data-centre-related measures on local ballots — zoning bans, building moratoriums, new permitting hurdles — spread across 18 jurisdictions, according to Ballotpedia’s tracker. Most passed. Piqua recalled its mayor and two city commissioners who had backed a proposed Meta facility; Urbana recalled its mayor on similar grounds; a county where Meta had planned a 700,000-square-foot data centre passed an outright ban. Ohio also elected Amy Acton, a Democrat, as governor; Morgan Stanley analysts said her win “creates significant risk of data center project delays in the state,” which is notable given that Ohio is also where OpenAI has reportedly been negotiating a roughly $500 billion, Nvidia-backed data-centre lease of its own — a far larger commitment than anything Anthropic has disclosed in the state.

The bigger shock came in Texas, where Democrat Gina Hinojosa defeated Governor Greg Abbott after centring her campaign on a statewide data-centre moratorium — the first Democratic governor Texas has elected in decades. South Carolina’s Annie Andrews flipped a Republican-held Senate seat on the same issue, and in Ohio’s Senate race Sherrod Brown’s campaign reportedly cast his opponent as “the face of data centres in Ohio,” per the Christian Science Monitor. This wasn’t a single-issue election — inflation and healthcare featured too — but data-centre siting was a costly issue for candidates who ignored it.

Why the pass/fail ratio keeps changing depending on who you ask

Three different numbers circulating this week for how lopsided the vote was isn’t sloppy reporting — it’s three trackers measuring three different things. Ballotpedia’s count is the most granular: 50 data-centre measures across 13 states for the full 2026 calendar year, of which 41 landed on the 3 November ballot specifically, across 10 states. As of 4 November it had 47 of those 50 decided, with 46 favouring new restrictions and exactly one — a town of fewer than 900 people in North Dakota — declining to add extra review. Redbud Consulting’s count, reported by Fortune, covers local contests with complete counts only and arrives at 37-to-1. Axios’s “43-1,” the figure that has travelled furthest in headlines, has no published methodology in the reporting we could find, and may blend ballot measures with candidate races that turned on the same issue.

Every credible tracker agrees on the direction — restrictions won almost everywhere they were on a ballot, by a double-digit-to-one margin at minimum. None agrees on the precise denominator, because “a data-centre ballot measure” isn’t one standardised category: some are outright bans, some 180-day moratoriums, some just a public hearing requirement or a water-use disclosure rule. Treating 43-1 as a citable statistic, rather than headline shorthand, is the first thing most coverage this week gets wrong.

What’s actually driving the backlash

Exit polling points to pocketbook concerns ahead of abstract AI anxiety. Nadine Smith, president and CEO of Color of Change, framed it to TechCrunch as a question of who absorbs the cost: “Families should never be subsidizing the richest corporations in history through their utility bills.” That matches the specific grievances in local campaigns — electricity rate increases tied to new grid capacity, water drawn for cooling, property-tax abatements that shift the local burden, and NDAs that kept councils from telling residents what was being built nearby. It echoes, from a different angle, what Apollo’s economists argued in October, when they said AI spending had delivered close to zero measurable payoff outside the tech sector itself — if the economic case for hosting a hyperscale facility isn’t visible to the household paying higher bills, the political case collapses with it.

Data & Society’s Maia Woluchem called the pattern bipartisan “frustration with tech power at minimum,” and the geography backs that up: rural Ohio, small-town Illinois, coastal Maine and conservative pockets of Kansas all produced restriction votes. Tech and policy researcher Aniketh Koneru’s read — “it is clear that money does not buy outcomes” — is borne out by the numbers: the pro-data-centre group Build American AI, tied to the Leading the Future super PAC, spent at least $2.54 million on Ohio TV ads alone and confirmed more than $5 million combined across Ohio, Wisconsin and Kansas, according to DeSmog — and voters in all three states went against data-centre development anyway.

Big Tech saw it coming — and started conceding before the votes were counted

What’s easy to miss in the “voters revolt” framing is that most of the concessions now reported as a response to the election were actually announced beforehand. Amazon’s “Built Together” programme — over $1 billion across five years for host communities, annual energy and water disclosure, lower-emission backup generators, and a promise to pay enough for its own power that other customers’ bills don’t rise — was announced in October. Microsoft said in January it would cover grid-infrastructure costs for its own facilities and stop seeking local property-tax breaks, then rolled out a five-point “Community-First AI Infrastructure” plan. Google has pledged to pay full property taxes on new campuses and says local-investment spending will reach $1 billion by the end of 2027; Google’s Devi Thomas told the Associated Press that “early public consultation is increasingly a very important part of the process.” Meta, via VP of public policy Lindsay Elin, said it is “reviewing our use of non-disclosure agreements.” Amazon, Microsoft and Google have all dropped NDA requirements for local officials.

The dollar figures attached to this capex are hard to pin down precisely, because outlets group different companies over different horizons: the Associated Press puts Amazon, Google, Meta and Microsoft’s combined spending at “more than $600 billion this year”; Fortune puts Amazon, Microsoft, Alphabet and Meta’s 2026 capital expenditure at $720–740 billion. Both are consistent with the scale described in Anthropic’s multibillion-dollar compute deals, where Nvidia and its partners increasingly function as landlords for frontier labs rather than hardware vendors — infrastructure at that scale was always going to collide with local politics eventually. Dell’Oro’s Baron Fung called the path ahead “two or three steps forward, one step back” rather than a full stop; Macquarie analysts framed the real effect as hyperscalers paying more for the “social licence to operate,” not hard restriction.

What most coverage is leaving out

Two things get lost in the 43-1 headline. First, recalls — not ballot measures — may be the more durable signal. Piqua and Urbana, Ohio, Yukon, Oklahoma, and a council seat in Mason, Michigan, all saw officials removed specifically over data-centre support, and CSMonitor counted as many as 30 recall attempts tied to the issue across 2026. A recalled official is a sharper deterrent to the next council weighing a project than a one-off ballot measure, because it demonstrates personal political risk rather than abstract sentiment. Second, almost none of this cycle’s coverage distinguishes opposition to AI as a technology from opposition to AI infrastructure as a neighbour. The same NBC exit poll that found roughly 80% opposed a data centre near their own home found no comparable hostility to AI tools themselves — this is a siting and cost-allocation fight, not evidence the public has turned against AI products, a distinction that matters for anyone forecasting what regulators do next. It also exposes a gap: the voluntary federal safety accord picked apart at October’s NYC Council hearing has essentially nothing to say about grid costs, water use or local disclosure — precisely the terrain where this fight is actually happening, county by county, not in Washington.

Practical takeaways for builders and publishers

For anyone commissioning, financing or writing about AI infrastructure, a few things look close to certain. Siting risk needs modelling market by market, not as a uniform regulatory environment — a county in Ohio and a county in Texas can land in opposite places within the same state in the same year. Transparency commitments that were optional a year ago — public energy and water disclosures, no-NDA clauses with local officials, published community-investment figures — are fast becoming the baseline expectation for getting a project through a local vote; a developer that skips them should expect a harder fight. Anyone citing a single ratio like “43-1” in reporting or investor materials should name the tracker it comes from, since the number moves with methodology, and conflating ballot measures with recalls or candidate races overstates a precision that doesn’t exist yet. Finally, the Ohio pattern — buildout concentrated in a handful of counties, intense localised opposition, a state government that can swing either way next cycle — is likely to repeat in Georgia, Virginia and other data-centre-dense states, and is worth tracking as a leading indicator ahead of capex decisions, not a lagging one.

What we still don’t know

There is no single, reconciled national tally yet — Ballotpedia’s own count is still showing three pending measures, and none of the trackers above has published a methodology detailed enough to let an outside analyst merge them cleanly. It’s also unclear how durable this year’s concessions will prove: pledges to end NDAs or cover grid costs were mostly announced as goodwill gestures ahead of a vote, and nothing yet compels a company to keep them once a project has been approved. Whether the backlash meaningfully slows the pace of AI compute buildout, as opposed to simply redirecting it toward friendlier counties, states or countries, is genuinely contested among the analysts quoted this week — Goldman Sachs expects construction delays in Georgia and Texas specifically, while Dell’Oro’s framing suggests a bumpier but largely unchanged trajectory. And it remains to be seen whether Congress responds at all: both the Sanders–Ocasio-Cortez moratorium proposal and Senator Heinrich’s GRID Savings Act were introduced before the election and neither has had a floor vote.

FAQ

How many AI data-centre ballot measures were decided in the November 2026 election?

It depends which tracker you use. Ballotpedia counted 41 measures on the 3 November ballot across 10 states (50 for the full 2026 calendar year, across 13 states), with 46 of 47 decided measures nationwide favouring restrictions. Redbud Consulting’s count of completed local contests put it at 37-to-1. The “43-1” figure in widely shared headlines appears to use a different, undisclosed methodology.

Did voters reject AI itself, or just data centres near them?

The available exit polling points mainly to the latter. Roughly 80% of respondents opposed a data centre in their own neighbourhood, but there’s no comparably strong signal of opposition to AI tools or products generally — this looks like a siting, cost and transparency fight more than a referendum on AI as a technology.

Which company has offered the biggest concessions?

Amazon’s “Built Together” programme is the largest publicly quantified commitment — more than $1 billion over five years for host communities — alongside pledges on energy and water disclosure and backup-generator emissions. Microsoft and Google have made comparable, if differently structured, commitments on tax breaks and infrastructure costs.

Will this actually slow the AI buildout?

Analysts disagree. Goldman Sachs expects specific construction delays in Georgia and Texas tied to the election results; others, like Dell’Oro’s Baron Fung, expect a bumpier but ultimately continued buildout rather than a reversal.

Sources

• Ballotpedia — Data center-related ballot measures, 2026 (primary tracker)
• Christian Science Monitor — “Voters send a resounding message against data centers”
• TechCrunch — “After the midterms, a bigger clash over data centers could be coming”
• Fortune — “After the midterms, Big Tech’s AI data center problem is only getting bigger”
• Associated Press via ABC News — “Facing pushback on data centers, tech giants are promising hosts more money and fewer secrets”
• Washington Post — Texas governor election results, 3 November 2026
• Axios — “Tech giants turn to cash, concessions for AI rebrand after voters pummeled data centers 43-1”
• DeSmog — “Revealed: Pro-AI Data Center Group Spending At Least $2.54M on TV Ads to Influence Ohio Election”

Please follow and like us: