What did US voters decide about AI data centers in the November 3, 2026 ballot measures?

On Tuesday, November 3, 2026, the US midterm elections put data centers on the ballot more often than ever before. Ballotpedia, the nonpartisan encyclopedia of American elections, counts 41 local measures about data centers in 10 states that day. By Wednesday, its unofficial tally showed voters choosing the side that limits data center development in 40 of them. CNBC, working from Ballotpedia's analysis plus results it gathered itself, and Scientific American both report the same picture: one win for development, in Harwood, North Dakota.

These were local votes, not state or federal laws. Each one applies to a single city, village, township or county, and some were only advisory. Taken together, though, they are the clearest measure so far of how residents feel about the large computing buildings behind AI. The guide's article on why communities oppose AI data centers explains the reasons behind that resistance; this one covers what happened at the ballot box.

Where the votes took place

Ohio dominated the list. The other nine states had between one and seven measures each.

Ballotpedia counts at least 50 such local measures across 2026 as a whole. Six came earlier in the year, and all six went against data centers too. Three more are scheduled: two bond votes in Clarksville, Arkansas, on December 1, and a proposed ban on sites of 25 megawatts or more in Mandan, North Dakota, on December 15.

What kinds of questions voters answered

The measures were not all alike. They fall into a few families:

• Bans. A permanent prohibition in the city's charter or zoning rules. Several bans in Ohio and Kansas use a threshold of 25 megawatts of electrical demand, a level that leaves small server rooms alone but blocks the very large sites used for AI. A megawatt is a million watts.
• Moratoriums. A temporary pause on permits, usually 180 days or a year, to give officials time to write rules. Wells, Maine approved a 180-day pause.
• Rezoning and project votes. A yes or no on one specific site. In Wilmington, Ohio, voters were asked four times whether to keep land rezoned from rural residential to light industrial.
• Process rules. Requirements that large projects go to a public vote, or that developers seeking tax breaks post financial guarantees.
• Advisory questions. Non-binding votes that tell officials what residents want, like the seven in Illinois.

Most of the November 3 measures started with residents. In Ballotpedia's list, 30 of the 41 are citizen initiatives or referendums, which reach the ballot by petition; the other 11 were referred to voters by a local council.

How to read the results: "yes" does not always mean "against"

Ballot wording matters. On some measures a yes vote limits data centers; on others a yes vote supports them. To compare results, you have to ask what each vote did, not just whether it passed. Here is a worked example with unofficial counts as Ballotpedia listed them on November 4:

Read row by row, the table shows why simple headlines can mislead. In Boulder City and Harwood, the "no" side won, yet the results point in opposite directions. Grove City shows that not every vote was a landslide: 55% to 45% is a real contest. Scientific American notes that construction of Polaris Forge 2, a $3 billion data center, is underway near Harwood, which gives local context for that result.

How solid are the numbers?

All of these results are unofficial. Counties and towns certify their results in the days and weeks after an election, and late ballots can shift a count. A few points to keep in mind:

• The total. CNBC, Scientific American and Ballotpedia all give 41 measures in 10 states for November 3. Ballotpedia's 2026 page counts 50 for the full year, in 13 states, because it also includes the earlier and later votes.
• Timing of the count. On Wednesday, November 4, CNBC described voters backing restrictions in nearly every contest "with results available," and Scientific American wrote that most results were in. Ballotpedia's page, as of November 4, lists 47 measures decided for the year, 46 of them against development and one for.
• Small margins in small places. Some votes involve a few hundred ballots. In Harwood, 430 votes were cast on the question, so a handful of late ballots could move the percentage, though not likely the outcome.

The wider election

Data centers also came up in races at the top of the ticket. CNBC reports that Democratic candidates who promised to pause or regulate new sites won several major contests, including the governor's offices in Texas and Ohio. Ohio's governor-elect, Amy Acton, supports what she calls a conditional moratorium, under which data centers would have to cover all the utility costs they cause. A CNBC voter poll found that 56% of voters rated data centers as a major or crucial factor in their choice.

Ohio matters to the AI buildup in particular. CNBC notes that OpenAI has lined up 8 gigawatts of planned capacity in the state, at a campus called PORTS-Pike, a project developed by SoftBank and SB Energy with Nvidia's financial backing. That project is not on a local ballot, but its home state just elected a governor who wants tighter conditions on such sites.

What do these votes actually change for AI data center projects?

The effect depends on the type of measure. A ban written into a city charter is binding inside that city's limits until voters change it again. A moratorium is a pause, usually for six months or a year, while officials draft zoning rules; projects can move ahead once those rules are in place. Advisory votes bind no one, though councils often cite them. And a vote in one town does not stop a developer from looking at the next county, which is one reason state-level rules draw so much attention.

For the AI companies that need these buildings, the practical message from November 3 is about time and conditions. Sites that need a rezoning or a public vote now face a real chance of rejection, and political leaders in large data center states have promised closer scrutiny of power bills and water use. Developers can still build where zoning already allows it, as in Harwood. What remains to be seen is whether the three votes still scheduled for December follow the same pattern, and whether state legislatures turn local resistance into broader rules.