The artificial intelligence data center boom hit a roadblock on election night, further complicating the tech industry's historic expansion plans.

There were 41 data center-related ballot measures up for a vote in the 2026 midterm elections in 10 states. Voters backed restrictions or rejected pro-development measures in nearly every contest with results available Wednesday, with Harwood, North Dakota, the exception.

Ohio alone accounted for more than half of the measures, with 21 on the ballot, according to CNBC's reporting.

Counting the measures

The tallies differ by source and by the hour they were compiled, which is worth stating plainly. CNBC, using Ballotpedia and its own collection of results, counted 41 measures across 10 states and a single exception in Harwood, where voters rejected an added review of industrial projects including data centers.

CBS News, citing an analysis from political research firm Redbud Consulting, reported that voters had approved 38 ballot measures restricting data centers in communities across California, Illinois, Maine, Ohio, New York and other states. Residents in Harwood, North Dakota, were the only voters to approve a data center.

Both accounts point the same way: in almost every local vote where a result was available, the side arguing for limits won. The difference in totals reflects how many contests had been called when each outlet compiled its numbers.

Ohio as the battleground

The issue also played out at the top of the ticket, where Democrat Sherrod Brown ousted Republican Sen. Jon Husted after casting the incumbent as "the face of data centers."

Ohio has become the focal point because of scale. OpenAI announced in August that it had secured 8 gigawatts of computing capacity at the PORTS-Pike Technology Campus in the state, the largest planned AI data center campus in the world, with construction due to begin this year and capacity arriving in phases from 2028.

SoftBank and SB Energy are developing the campus with the US Department of Energy, and Nvidia is backstopping a project that has become a test of whether a community will accept the load, the water and the transmission lines that come with it.

The Texas question

Texas accounts for 25% to 30% of the data center capacity expected to come online by 2030, and its governor's race turned on the issue. Democrat Gina Hinojosa, who defeated Republican incumbent Greg Abbott, proposed a temporary ban on data center construction during her campaign.

Stephen Dover, chief market strategist at Franklin Templeton, predicted the election results would lead to a "slowdown in data centers being built", bringing new regulations and costs for developers. Dan Ives of Yorkville Ives & Co. put it more bluntly when asked whether the votes would slow the pace of the 800 to 1,000 data centers planned across the US: he said the answer was a clear yes.

Hinojosa's win could stall the addition of 13 gigawatts of data center power in Texas, according to Gordon Johnson of GLJ Research. Abbott had already suspended data center approvals, which makes it easier for the new governor to extend a moratorium rather than create one.

What Wall Street expects

The polling backdrop explains why the measures passed. By a margin of more than 2-to-1, more Americans oppose than favour building a new data center in their area, a CBS News poll found this summer.

That is a difficult environment for a buildout that needs local consent for land, power and water. Analysts are now modelling delays rather than cancellations, because the demand from AI training and inference has not fallen, while the permits have become harder to obtain.

The votes also changed who is making the decision. A project that once needed a county board now sits inside a governor's agenda, and the states competing hardest for the investment are the ones where the politics of it just became more expensive.

The measures themselves are narrow, covering zoning, setbacks, noise limits and water rules, but their cumulative effect is a longer queue of approvals. Developers can absorb one county saying no; they cannot absorb dozens saying no at once, because the transmission lines and water systems they need cross those boundaries.

That is the part the market is pricing. A data center needs a power connection years in advance, and a project that loses its local consent does not simply move to the next state, because the grid capacity is where the grid capacity is. The industry's response has been to promise community payments and local hiring, which are the standard tools for a project that needs consent, and to argue that the load brings tax revenue that would otherwise go elsewhere. Neither argument addresses the two things voters cited most often in local hearings: water use in dry counties and the noise of cooling equipment at the edge of a subdivision. Both are local grievances that a national AI argument does not answer, which is why the measures passed by wide margins rather than along party lines.

The wider picture includes the AI trade that lifted markets this autumn, the historic firsts in this week’s results and the order against election officials accused of intimidation.