Voters Reject AI Data Centres in Nearly Every Local Ballot Measure as the 2026 Midterms Deliver a Near-Sweep Against the AI Buildout

Introduction

On the night of November 3, 2026, voters across the United States did something the technology industry had not expected on anything like this scale. In town after town, and in city after city, they were asked a direct question about the giant buildings that power artificial intelligence, and in almost every place they answered no. The question took many forms. Sometimes it was a ban. Sometimes it was a moratorium. Sometimes it was a new rule saying that the public must get a vote before a large project can go ahead. But the direction of travel was the same almost everywhere, and by the morning after the election the pattern was clear enough that even the industry’s own supporters were describing it as a rout. The results were reported by Ballotpedia’s tracker and by CNBC, and they were still unofficial when the first counts were published, because local election offices were still certifying their totals. Yet the shape of the outcome was not in doubt. Voters in ten states weighed in on 41 local measures connected to data centres, and the measures that would limit or block the facilities won in 40 of them. Only one went the other way. The story of the night is not only about buildings and electricity. It is about how a technology that was once discussed mainly in boardrooms and research papers has become a neighbourhood issue, argued over in village halls and at kitchen tables, and decided by people who live next door to the proposed sites. The figures in this post are presented only as reported and attributed, and the results described here were still unofficial at the time of writing, so the final certified totals may differ slightly from the early counts. The key point of this report is that voters reject AI data centers midterm ballot 2026 in a near-sweep across ten states.

1. The scale of the rejection: 40 of 41

1.1 What Ballotpedia counted

The starting point for understanding the night is the count itself. Ballotpedia, which tracks ballot measures across the country, recorded 41 local measures on data centres in ten states on the November 3 ballot. A yes vote on 28 of those 41 would itself limit data centres, through bans, moratoriums, zoning rules or a requirement that the public vote on big projects. The other measures were framed the other way round, so a no vote was the vote that kept the limits in place. When the results were gathered, the tally reported by CNBC and by the tracker Data Centers on the Ballot was that voters sided against data centres in every measure except one. An updated count later put the figure at 40 of the 41 measures going the way of tighter limits, either by approving bans, moratoriums and new requirements, or by rejecting measures that would have helped data centres get built. The New York Post put the total higher still, reporting at least 50 measures with voters deciding against data centres in 47 cases and three outcomes still pending on Wednesday afternoon, a reminder that different outlets were counting slightly different sets of measures. What nobody disputed was the direction. Stateline reported that by Wednesday, 33 local ballot measures restricting data centres had passed or were leading in counts still under way, with only one measure going the way of the industry’s supporters. Counting the six votes decided earlier in the year, 46 of the 47 data-centre measures decided in 2026 had favoured limits. That is a record of near-total defeat for the industry at the ballot box in a single year, and it arrived in the same election in which the Democrats took control of both chambers of Congress.

1.2 The one exception in Harwood

The single exception to the pattern was Harwood, North Dakota, and even that exception was a narrow one. Voters there were asked whether to change the city charter so that large industrial projects, including data centres, would need an independent review and a public hearing before they could be adopted. They rejected that amendment by 325 votes to 105, with 75 percent of the 430 people who voted saying no to the extra review. The secretary of state’s office reported the result, and it was the only place on the night where the industry’s side of the argument appeared to win. Harwood matters because it shows what the night was not. It was not a case of voters being unable to find the question on the ballot, or of the measures failing for technical reasons. In Harwood the voters understood the proposal and chose the other way, which makes the results everywhere else look more like a deliberate choice and less like an accident. It also matters because North Dakota is one of the states where more votes are still to come, with a further vote expected in December, so the state’s story is not finished. But for one night at least, Harwood stood alone, and the rest of the map pointed the other way.

2. Ohio at the centre of the fight

2.1 Twenty-one measures in eighteen places

No state carried more of the fight than Ohio. The state accounted for 21 of the 41 measures, spread across 18 jurisdictions, more than any other state by a wide margin. Illinois came next with seven, followed by Maine with four and Kansas with three. The other states with measures were California, Florida, Nevada, New Jersey, North Dakota and Wisconsin. In Ohio, every one of the 19 jurisdictions with reported results voted against development, according to the counts published after the election. Around Columbus, voters went against data centres in Ashville, Pataskala, Grove City, Sunbury and Granville, according to unofficial counts reported by NBC4. In Wilmington, four rezoning proposals connected to data-centre projects were rejected by roughly 80 percent to 20 percent. In Ashville, more than 83 percent of voters backed a measure challenging the village’s negotiated terms with an EdgeConneX facility. In Hubbard, more than three-quarters of voters approved a ban on data-centre construction, according to Trumbull County election results. The pattern was not confined to one kind of place. It appeared in small villages and in larger suburbs, in communities that had been offered deals and in communities that had not, and the margins were often very large. Ohio’s prominence in the results is no accident, because the state has become one of the main battlegrounds for the buildout, with large campuses proposed in and around Columbus and in the south of the state.

2.2 The PORTS-Pike campus and the stakes

The largest single project hanging over Ohio is the PORTS-Pike Technology Campus, where OpenAI announced in August that it had secured 8 gigawatts of computing capacity, described as the largest planned AI data-centre campus in the world, with phased capacity expected from 2028. That announcement set the scale of what was at stake. Eight gigawatts is an enormous amount of power, enough to supply millions of homes, and the campus would sit in a region where the electricity grid, the water supply and the roads were never designed for a project of that size. Supporters of the campus argue that it will bring jobs, tax revenue and a leading position in the technology that is reshaping the economy. Opponents argue that the jobs are few relative to the size of the buildings, that the tax benefits are often reduced by incentives, and that the costs fall on local residents in the form of higher electricity bills, strained water supplies and construction traffic. The ballot measures did not decide the fate of PORTS-Pike directly, but they decided the rules under which projects like it will be judged, and they sent a message to every company planning a campus in Ohio that the welcome it once expected is no longer guaranteed. That message will be read in boardrooms far beyond Ohio, because the same companies are planning campuses in other states, and the same arguments are being made in other communities.

3. How voters decided, place by place

3.1 The biggest margins

The most striking feature of the results was not only how many measures went against data centres, but how large some of the margins were. In Alhambra, California, in Los Angeles County, eight miles from downtown Los Angeles, 82 percent of voters backed a ban on data centres throughout the city, in the only California jurisdiction with a data-centre question on the ballot this time. In Joliet, Illinois, 79 percent of voters said no to an advisory question asking the city to approve AI data centres, and in Morgan County, 88 percent backed restrictions. In El Dorado, Kansas, voters approved a 10-year ban on high-impact data centres and certain battery-storage facilities by 71 percent to 29 percent. In Edgewater, Florida, voters said no new data centres could be built there. Earlier in the year, the pattern had already been visible. In Monterey Park, California, voters backed a ban on the facilities by 88.3 percent, in what appeared to be the first vote of its kind in the United States, and in Port Washington, Wisconsin, voters had voted in April for an initiative asking whether voters should have to approve tax increment financing districts over 10 million dollars, which are also used for data centres. These are not close results. They are the kind of margins that suggest a settled view rather than a momentary mood, and they were repeated in place after place.

3.2 The closest calls

Not every result was a landslide, and the closest ones are worth noting because they show where the argument is still being fought. In Ohio, the zoning rules in Sprigg Township passed with 53 percent, and Grove City’s requirement for a public vote on big developments passed with 55 percent, the closest results in the state. Those figures matter because they show that the industry’s case is not entirely without support. In some places, the argument that a data centre brings investment and jobs still carries weight, and the result turned on a few hundred votes. But even the close results went the same way. No close result went the industry’s way except Harwood, and the overall picture was of a movement that won the big arguments and most of the small ones too. The closeness of Sprigg Township and Grove City also hints at what the next phase of the fight may look like. Where the margins are narrow, companies and their supporters may try again, with different proposals, different promises and different sites. Where the margins are wide, the message is harder to ignore, and the political cost of trying again is higher. For now, though, the record stands as a sweep, and the close calls are the exception that proves how broad it was.

4. Power, water and the cost of living

4.1 What ‘high-impact’ means in Kansas

To understand why voters reacted as they did, it helps to look at what the measures were actually about. In Kansas, voters adopted some of the most restrictive proposals in the country, and the state defined a high-impact facility as one requiring more than 10 megawatts of peak electricity or more than 100,000 gallons of water use per day. Those thresholds sound technical, but they describe the everyday reality of a modern data centre. A single large facility can use as much electricity as a small town, and it can use very large amounts of water to keep its servers cool. The Kansas measures were not aimed at small computer rooms in office buildings. They were aimed at the industrial-scale facilities that have become the standard for training and running artificial intelligence systems, and they were written in terms that local people could understand. The El Dorado ban, approved by 71 percent to 29 percent, covered high-impact data centres and certain battery-storage facilities for ten years, a period long enough to force a serious rethink of where such facilities belong. The Kansas example shows how the debate has moved from abstract arguments about technology to concrete questions about how much power and water a community is willing to give to a single building.

4.2 Bills, grids and neighbourhoods

The concerns that drove voters were practical ones. The first was electricity. Data centres draw power constantly, day and night, and in many parts of the country the grid is already under strain. Residents worried that their bills would rise to pay for the new transmission lines and substations that the facilities need, and that the reliability of their own supply would suffer when demand peaks. The second was water. In dry states and in communities that depend on aquifers, the idea of a single facility using hundreds of thousands of gallons a day was hard to accept, particularly when farmers and households were being asked to conserve. The third was the character of the place itself. A data centre is a large, windowless building with few workers, heavy truck traffic during construction, and a constant hum from its cooling systems. People who bought homes in quiet suburbs or farming villages did not expect to find one next door, and they resented being told that they had no say in the matter. The ballot measures gave them that say, and they used it. The result was a verdict not only on data centres but on the way decisions about them had been made, often by councils and developers, with little public consultation, and often with tax incentives that were negotiated behind closed doors.

5. The governor races that ran on data centres

5.1 Texas and Georgia

The ballot measures were not the only place where data centres shaped the election. In Texas, the state with the most planned AI data-centre capacity in the country, Democrat Gina Hinojosa beat Governor Greg Abbott after promising a moratorium on new data-centre construction. With about 93.5 percent of the expected vote counted, she had 52.8 percent to Abbott’s 45.7 percent, according to NBC News. In Georgia, Keisha Lance Bottoms won the governor’s race after calling for a pause on new data centres, leading Republican Rick Jackson by 54.2 percent to 45.8 percent with 98.2 percent of the expected vote counted, on NBC News’s count. These were not single-issue races, and it would be wrong to say that data centres alone decided them. Both states had other issues on the ballot, including the cost of living, education and the national political mood. But the fact that two winning candidates had made data centres a central part of their campaigns, and that both won in states where the industry had expected a friendly reception, is a sign of how far the issue has travelled. It is no longer a local planning dispute. It is a statewide political issue, and it is one that candidates now believe can win them elections.

5.2 Ohio’s governor race

Ohio told a similar story. Democrat Amy Acton beat Vivek Ramaswamy by 52 percent to 44.7 percent, according to NBC News, in a race in which the data-centre buildout was one of the live issues. Ramaswamy had been associated with the pro-growth, pro-technology case for the industry, and his defeat in a state where 21 measures on data centres were on the ballot on the same day is hard to read as a coincidence. The three governor results together suggest that the backlash against data centres is not confined to the places where a facility is proposed. It has become a broader political force, one that cuts across the usual lines of the debate. Voters who might disagree about many things appear to agree that the current pace and scale of the buildout, and the way its costs are shared, are not acceptable. That agreement is what makes the results of November 3 so significant. They were not the product of a single campaign or a single organisation. They were the product of a view that has formed in many places at once, and that found expression in many different kinds of vote on the same night.

6. What the polls said before election day

6.1 The August survey on living near a data centre

The results should not have come as a complete surprise, because the polling had been pointing in this direction for months. An August 2026 poll by Embold Research, of 2,045 registered voters, found that 75 percent opposed a new data centre near their home, according to Heatmap’s reporting of the survey. That figure is striking because it is not a narrow majority. It is three quarters of the electorate, and it was recorded before the campaigns over the ballot measures had reached their peak. The poll suggests that opposition to data centres is not mainly a matter of party or ideology. It is a matter of proximity. People may support artificial intelligence in general, and they may use its products every day, but they do not want the infrastructure that powers it built next to their homes, schools and farms. That gap between support for the technology and opposition to its physical presence is the central political problem for the industry, and it is one that no amount of advertising has so far solved. The poll also helps to explain why the ballot measures won so widely. They were not asking voters to take a position on an abstract question. They were asking them about something they had already decided.

6.2 Election-day voters on the issue

The polling on election day itself told a similar story. According to exit polls published by NBC, 56 percent of voters described AI and data centres as a crucial or major factor in their vote, and of those, 64 percent voted for Democrats. The CNBC voter poll reported the same 56 percent figure, and noted that Democrats ranked the issue higher than Republicans. That does not mean the issue belonged to one party. In many of the local measures, the votes against data centres came from communities that vote Republican in national elections as well as from communities that vote Democrat. But the exit polling does show that the issue mattered to a majority of voters, and that it mattered more to the voters who backed the Democrats, who won control of Congress on the same night. The combination of the August survey and the election-day polling paints a consistent picture. Voters had made up their minds about data centres before they went to the polls, they regarded the issue as important, and when they were given a direct say, they used it to set limits. The industry now has to reckon with a public that is informed, engaged and largely opposed to the current model of development.

7. What happens next for the industry

7.1 The December votes still to come

The story is not over, because more votes are scheduled before the year ends. Additional votes are expected in Arkansas and North Dakota in December 2026, including two referendums in Clarksville, Arkansas, on December 1, on bonds of up to 50 billion dollars and 5 billion dollars for data-centre projects, and a ban on data centres in North Dakota on December 15. Those votes will be watched closely, because they will show whether the November pattern holds or whether the industry can recover ground. The Clarksville votes are particularly interesting, because they involve public money on a very large scale, and they will test whether voters are willing to borrow to support the buildout or whether they see the bond proposals as another way of shifting the costs onto the public. The North Dakota vote will test whether the Harwood result was a sign of a different mood in that state or simply a local exception. Beyond December, the Ohio effort to put a constitutional amendment banning new data-centre construction on the ballot next year is already being discussed, and similar efforts may follow in other states. The November results have given those campaigns a model and a momentum they did not have before, and the industry will have to decide whether to fight them, to negotiate with them, or to change its approach.

7.2 What companies and councils must decide

For the companies building data centres and for the councils that host them, the results pose a set of practical questions. The first is where to build. If voters in ten states have said no to the current model, the industry will need to find places where the facilities are genuinely welcome, and that may mean smaller sites, sites further from homes, or sites where the power and water are already available without new infrastructure. The second is how to pay. If voters believe they are being asked to subsidise the industry through higher bills and tax incentives, the industry will need to show that it is paying its own way, and that the benefits it promises are real and shared. The third is how to talk to communities. The ballot measures were in part a response to a feeling that decisions had been made without the people who would live with them, and the companies that succeed in the next phase are likely to be those that consult early, share information openly and accept limits that the community sets. None of this means the buildout will stop. The demand for computing power is real, and the technology will continue to grow. But the results of November 3 suggest that it will grow on different terms, and in different places, than the industry had planned. The era in which a data centre could be built almost anywhere, with little public debate, appears to be over.

8. What remains unconfirmed and what to watch next

8.1 The counts still to be certified

Several things about the results remain unconfirmed, and they should be stated plainly. The first is the exact count. Different outlets reported different totals, with Ballotpedia and CNBC putting the figure at 41 measures in ten states, and the New York Post reporting at least 50, with three outcomes still pending on Wednesday afternoon. The difference reflects different definitions of what counts as a data-centre measure, and different cut-off points for the counting. The second is the final certified totals. Most of the results described here were still unofficial when they were first reported, and local election offices were still certifying their counts. The final figures may differ slightly from the early counts, and a small number of races could still change. The third is the precise effect of the measures that were advisory rather than binding. Some of the votes were advisory, which means they do not by themselves stop a project, and their effect will depend on what local elected officials do next. None of these uncertainties changes the overall picture, but they do mean that the numbers in this post should be read as the reported and attributed counts available at the time of writing, not as final certified results.

8.2 What to watch next

The next steps are clear enough to list. Watch the December votes in Clarksville, Arkansas, on December 1, and in North Dakota on December 15, to see whether the November pattern holds. Watch whether the Ohio constitutional amendment effort reaches the ballot next year, and whether similar efforts appear in other states. Watch what the companies do next, whether they pause projects, move them, or change their proposals to meet the concerns that voters raised about power, water and bills. Watch what local councils do with the advisory votes, and whether they treat them as instructions or as opinions to be noted and set aside. And watch the governor races that were decided on this issue, to see whether the promises made during the campaigns, including the moratorium in Texas and the pause in Georgia, are turned into policy. The results of November 3, 2026, were a verdict, but they were also the start of a longer argument about where artificial intelligence is built, who pays for it, and who gets to decide. The coming weeks will provide the answers.

Comments

comments