Ambitions to expand AI infrastructure in the United States face a growing local challenge after voters backed restrictions on data centers in the midterm elections, reinforcing what experts describe as an emerging constraint on development.
On Nov 3, voters favored limiting data center development in 40 of 41 local ballot measures, according to Ballotpedia. The measures included outright bans, temporary moratoriums and voter approval requirements.
The results reflect growing public concerns over the noise, land use and environmental impact of large data centers, as well as their impact on electricity and water bills. They also contrast with the federal policy push.
In July last year, US President Donald Trump signed an executive order prioritizing the rapid construction of AI data centers and supporting energy infrastructure through streamlined federal permitting and access to federal land.
However, public opposition has grown rapidly. "I've never seen an issue where the public sentiment has moved as quickly as it has as opposition to data centers in the past year," said David Sandalow, inaugural fellow at Columbia University's Center on Global Energy Policy, at an energy conference before the midterms.
Sandalow cited a December study he coauthored examining state legislation on data centers. It found that incentives to attract the facilities dominated the laws enacted. Last year, more state legislatures were trying to attract data centers than pursuing any other policy related to the facilities, he said.
The political landscape has since shifted. With Democrats wresting control of both chambers of Congress from Republicans in a "blue wave", experts said the election results could give incoming officials greater political support to tighten development rules. Democrats have pledged to clamp down on the massive, energy-hungry facilities.
The potential policy changes could extend across national, state and local levels, according to a Brookings Institution analysis of the election's implications. It identified possible measures, including extended moratoriums, mandatory protections against higher electricity costs, a reconsideration of tax exemptions, and greater scrutiny of projects previously eligible for automatic approval.
Texas offers an example of how opposition to data centers is reshaping political debate. Republican Governor Greg Abbott announced a moratorium on data center construction this year before losing his reelection bid to Democrat Gina Hinojosa.
Hinojosa, who made data centers a key part of her campaign, has outlined several planned restrictions on the facilities, including voiding contracts that affect water supplies and requiring data centers to pay taxes on electricity.
"It's a very dramatic change," Sandalow said of the opposition to the facilities at the recent US-China Green Energy Summit at Stanford University. "This is going to be the constraint going forward. And part of this is driven by data centers moving into these suburban areas."
The issue has also emerged as an early test for Congress. The Bipartisan American Affordability and Jobs Act is designed to speed up permitting reviews while requiring large data centers to pay for the energy they use. The Senate held a procedural vote on Monday to gauge support for advancing the legislation.
However, the bill faces a more difficult path following the midterms. Investment firm TD Cowen has lowered its estimate of the likelihood of its passage from 40 percent before the elections to 25 percent.
Investment hurdles
The election results could also create additional challenges for AI infrastructure investment, particularly as newly elected state officials consider restrictions on data center development, according to a post-election analysis by Morgan Stanley.
Decisions governing US data center projects rest largely with state utility commissions, legislatures and county boards. Additional state regulations could create a more fragmented environment, the analysis said.
Sandalow said the changing nature of AI computing could further intensify local opposition.
In the next five years or so, a greater share of computing is expected to be devoted to inference — when trained models can make predictions or decisions on new, previously unseen data — rather than large-scale model training, he said.
Applications such as autonomous vehicles require relatively low latency, meaning data centers need to be located closer to where their computing capacity is used.
"To the extent that we're locating data centers in suburban areas, that's going to cause a lot of opposition," he said.
