Meta Warns Midterm Backlash Will Make Building AI Data Centres Harder as Voters Reject Projects Across America

Meta has warned that the backlash against data centres unleashed by this week’s midterm elections will make it harder to build the infrastructure behind artificial intelligence in the United States, in the clearest admission yet from a technology giant that voters have become a serious obstacle to the AI building boom.

Dina Powell McCormick, the president of Meta, told the Wall Street Journal’s Tech conference on Wednesday that the Democratic victories in the midterms would complicate construction of new data centres, and that the industry had failed to explain to the public why the vast, power-hungry facilities were worth hosting. “I think it’s going to be harder,” she said, according to the Wall Street Journal. “It has been hard, I should say, but I also think that it’s becoming clear what models work, and the states want those models that work.”

Her warning came two days after voters across America delivered a sweeping verdict against data centre expansion. According to an analysis by Ballotpedia, with results compiled by CNBC, voters in 10 states backed restrictions on data centres or rejected pro-development measures in 40 of the 41 local ballot measures decided on Tuesday. The New York Post, using a broader count of all data centre measures decided in 2026, reported that voters had decided against the centres in 47 cases, with three votes still pending. The scale of the defeat stunned an industry that had treated local approval as a formality.

Powell McCormick said the backlash was about more than planning permission. “The visible sort of manifestation of AI is a data centre, and it’s something that people can say, ‘I don’t want this in my community,'” she said, according to TechCrunch. “I think there’s an underlying bigger issue, which is how people feel about AI in general.”

Meta, which is spending up to $145 billion on capital expenditure this year according to the Wall Street Journal, is racing to build enough computing power to run its personal AI agents, called Muse, for its 3.6 billion users. The company already operates data centres in 23 states, Powell McCormick said, and five of its projects, including one in Texas, are planned at one gigawatt or more. It is building a huge data centre in Louisiana called Hyperion, another in Ohio called Prometheus, and a third in Texas as part of a venture with the investment firm BlackRock.

But the political ground has shifted underneath those plans. The governors-elect of Ohio and Texas have both called for conditional freezes on new data centre construction, according to the Wall Street Journal, and the governor-elect of Texas, Gina Hinojosa, campaigned on a temporary moratorium on new data centres and an end to the tax incentives that lure them. In Georgia, the incoming governor also ran on tougher rules. Powell McCormick said she expected “a lot more review” of new projects. “What I mean by that is to start new data centre build-outs today in the United States will be very challenging,” she said.

She offered one note of hope for the industry. A governor who won on an anti-data-centre platform had, she said, told Meta: “As soon as we get past the inauguration, let’s meet because we don’t want to lose this investment in our town. We don’t want these jobs to go elsewhere.” Expanding existing projects, she added, would be easier than starting new ones, “because you’ve had a major track record for a few years, and the community has felt those benefits. They haven’t felt cost. They’ve actually felt benefits.”

The ballot-box revolt

The numbers from Tuesday’s vote explain why Meta is worried. At least 50 local ballot measures dealing with data centres were on ballots around the country in 2026, according to Ballotpedia, and 41 of them were decided on election day itself. Voters backed the restrictive side in all but one of those 41, according to the Ballotpedia and CNBC analysis reported by Fox Business. The Washington Post separately counted at least 38 jurisdictions that passed measures limiting or blocking data centres.

Ohio was the epicentre. Twenty-one ballot measures related to data centres were put to voters in the state, the most anywhere in the country, according to the New York Post. In Defiance, voters approved a charter amendment restricting data centre construction by 64.73 per cent to 35.27 per cent. In Pataskala the margin was 73.62 to 26.38, and in Granville 82.45 to 17.55. In Ashville, 81.61 per cent voted against a data centre referendum. The only place where the pro-development side won was Harwood, North Dakota, where voters rejected a measure that would have imposed additional review requirements on industrial sites.

A separate tracker, Data Centers on the Ballot, reported that the side favouring restrictions led in 44 of the 45 measures with results by midday on Wednesday, taking 72.8 per cent of all votes cast. Those figures are unofficial and the final tallies may change as counting continues, and different counts use different scopes: some count only the 41 measures decided on Tuesday, others count all 50 measures decided across 2026. But every count points in the same direction, and by a wide margin.

The revolt was not confined to the ballot box. According to Techstrong IT, New York Governor Kathy Hochul has enacted a one-year construction moratorium on facilities larger than 50 megawatts, while Pennsylvania Governor Josh Shapiro has stripped data centres from expedited permitting channels and banned state-level non-disclosure agreements. In Texas, Governor Greg Abbott has halted new grid interconnection approvals and required full disclosures on energy and water use. Monterey Park in California enacted a permanent municipal ban on the facilities earlier this year.

The opposition had been building long before election day. More than 145 data centre projects were delayed or cancelled through early August, according to figures from S&P Global cited by the Washington Post, with public opposition behind most of them. In the second quarter of 2026 alone, at least 45 projects worth $88 billion faced some form of local opposition, according to the research group Data Center Watch.

OpenAI’s answer: pay the power bill, drop the secrecy

The industry’s response came within 48 hours of the vote. On Thursday, OpenAI published a National Community Compact and a Compute Campus Policy Blueprint, promising a new approach to building the facilities it calls “compute campuses”. The company said it would stop asking public officials to sign non-disclosure agreements when negotiating or discussing future sites, publish construction plans and assessments of the effects on traffic, noise, lighting, water and energy, and pay the full cost of the electricity its facilities use, along with the grid infrastructure needed to serve them.

“We’ll pay the full cost of the electricity we use and the infrastructure needed to serve us. Before major spending begins, we’ll provide financial assurances so other customers are protected if a project is delayed, reduced, or canceled,” OpenAI said in its announcement, according to reporting by Bloomberg. The company also promised to cut power use during peak demand, send electricity back to the grid, set local hiring goals, and publicly report its progress.

Chris Lehane, OpenAI’s chief global affairs officer, said the company had been working on the plan since August, after months of conversations with residents and officials in Georgia, Ohio, Texas and other states. He told reporters that the facilities should be seen as economic engines rather than the silent “black boxes in a pasture” that many residents imagine, pointing to a planned project in southern Ohio projected to support 2,500 long-term jobs averaging $80,000 a year. OpenAI’s president, Greg Brockman, acknowledged at a separate event on Wednesday that the company had not explained itself well enough.

OpenAI also pointed to money it has already committed to host communities: $80 million for community projects in Effingham County, Georgia, including $4.8 million for countywide school meals and $2 million for a local park, and $40 million in Pike County, Ohio. Its agreement with Georgia Power allows the utility to reduce the power delivered to its planned Effingham campus by up to 1,000 megawatts during periods of high demand. The company said it had voided non-disclosure agreements in Effingham County in July.

But the compact comes with limits that its critics are already pointing out. It applies only to campuses that OpenAI itself owns and operates, and it sets out no penalties for breaking its promises and no independent verification of the figures it publishes. Much of the infrastructure reserved for OpenAI is built and owned by partners, including Oracle, according to Reuters, and the company has not said whether the compact automatically covers those partner-owned facilities. The blueprint, by contrast, asks lawmakers to turn many of the same rules into law for the whole industry, including large-load electricity tariffs that would force big power users to pay the extra grid costs they create.

Meta, for its part, says it already follows a version of the approach OpenAI is now promising. Powell McCormick said successful projects see Meta paying for its own electricity, directing tax surpluses from the data centres towards teachers and first responders, keeping the workforce local, and winning bipartisan support. “The stakes of who builds the technology that cures cancer, that gives access to education, that is also a very fundamental difference,” she said. “Mark [Zuckerberg, Meta CEO] is very ‘long human,’ and he’s betting on humans more than machines.”

A boom built on borrowed time

The political backlash arrives at a moment when the economics of the AI boom are already under strain. The Nasdaq Composite fell into correction territory on Thursday, dropping more than 10 per cent below its record close of 6 October, in a sell-off driven by chipmakers and fears that the Democrats’ midterm sweep would bring tighter regulation of the AI build-out. Meta itself completed a $15 billion senior-notes offering on 5 November, according to a securities filing, to help fund its data centre plans just as the market turned nervous about all that spending.

The sums involved are enormous. The four largest hyperscalers — Meta, Microsoft, Amazon and Google — are expected to spend roughly $800 billion on capital expenditure in 2026, rising to about $1.2 trillion in 2027, according to Goldman Sachs figures reported earlier this week. Enterprise spending on AI is forecast to reach $176.8 billion in 2026, more than five times last year’s level, according to the venture firm Menlo Ventures.

Yet even the industry’s own leaders admit they cannot see clearly where all that spending leads. At a financial summit in Hong Kong earlier this week, Goldman Sachs chief executive David Solomon said demand for computing power vastly exceeded supply for the next couple of years, but that beyond that “nobody knows how it will play out”. Citadel Securities chief executive Peng Zhao warned that the technology was developing faster than the investment cycle, so that small forecasting errors could lead to huge miscalculations. Investors, he said, were “suddenly confronting how little they can know with confidence”.

Powell McCormick struck a similar note about the value of what Meta is building even if its consumer AI agent, Muse, never reaches the billions of users the company hopes for. Muse has about 10 million users today. “We’re getting calls every day to lease that compute [we’ve built], and so one way or another, we’re seeing a path to monetization,” she said. The computing power itself, in other words, will be valuable whatever happens to the products that run on it.

That argument may not reassure the communities being asked to host the buildings. A data centre’s neighbours see the costs first: the electricity demand that can push up household bills, the water drawn for cooling, the noise, the traffic, the land taken up, and the relatively small number of permanent jobs once construction ends. The benefits — cheaper AI tools, medical research, economic growth — are real but distant and diffuse, while the costs are immediate and local. As one industry executive put it to Techstrong IT: “Voters were not rejecting computing. They were rejecting the idea that households should pay for the grid upgrades and the power that hyperscale campuses need.”

The backlash is also spreading beyond the United States. In Europe, Amsterdam has barred new municipal data centres through 2030, Denmark has introduced rules demoting data centre operators to the lowest tier of grid connection priority, and Ireland now requires new facilities to supply their own on-site power generation. Portugal and Australia are advancing legal frameworks requiring green power sourcing, alternative cooling technologies and local oversight. The pattern is the same everywhere: the infrastructure that AI needs is colliding with grids, water supplies and neighbourhoods that were never designed for it.

Analysis: Why It Matters

The ballot box has become AI’s first real regulator

For years, the debate about regulating artificial intelligence has taken place in Washington, Brussels and Beijing, in hearings about model safety, copyright and national security. Tuesday’s vote showed that the most effective regulation may come from town halls and county ballots instead. Voters did not need to understand transformer architectures or training runs to pass judgement on the industry. They understood a building the size of several football fields, a power bill that might rise, and a water supply that might fall, and they said no — in 40 of 41 cases, by margins that in some Ohio towns exceeded four to one.

That is a different kind of check on the industry from anything Congress or the European Union has produced, and in some ways a more powerful one. A federal AI safety law can be lobbied, delayed and watered down over years. A local referendum kills a project in a single evening. And because the results cut across party lines — the revolt was strongest in Ohio, a state that has voted Republican in recent presidential elections — no party can safely claim the issue as its own or dismiss it as the other side’s politics. The industry now faces a political map in which the opposition is everywhere and organised, and in which its own customers’ neighbours are its most effective critics.

The economics changed: who pays for the grid

Strip away the politics and the argument at the heart of the backlash is about money, specifically about who pays for the electricity grid. A single gigawatt-scale data centre consumes roughly as much power as a mid-sized city, and connecting it to the grid can require new transmission lines, substations and generation capacity that cost billions. For years, utilities spread those costs across all customers, which meant households quietly subsidised the AI boom through their power bills. Voters have now worked that out, and they do not like it.

This is why the most important sentence in OpenAI’s compact is not the one about non-disclosure agreements but the one about paying “the full cost of the electricity we use and the infrastructure needed to serve us”. It is also why Meta’s Powell McCormick emphasised that Meta pays for its own electricity. Both companies understand that the fight will be won or lost on the question of whether residents’ bills rise. The blueprint’s proposal for large-load tariffs, which would legally require big power users to pay the extra grid costs they create, may be the most consequential idea in either document, because it would turn a voluntary promise into a rule that applies to every builder, including the ones that never publish a compact.

The industry’s charm offensive — and its limits

OpenAI’s compact and Meta’s public messaging amount to the same strategy: convince communities that data centres are good neighbours by making the benefits visible and the costs somebody else’s problem. There is genuine substance in the promises — published water and power figures, independent noise studies, local hiring targets, money for school meals and parks — and the fact that both companies are making them at the same time suggests the industry has finally accepted that secrecy and take-it-or-leave-it deals no longer work.

But there are limits that should not be overlooked. A compact is a promise, not a permit condition, and OpenAI’s own document sets out no penalties for breaking it and no independent check on the numbers it reports. It covers only the campuses OpenAI owns and operates, not the partner-built facilities that supply much of its computing power. And the timing invites scepticism: OpenAI says the work began in August, and its chief global affairs officer told Axios on 1 November that a policy push would follow the midterms, but the announcement still landed two days after the voters spoke. Communities will judge these promises by what happens at the next planning meeting, not by what was published in a press release. The first real test will be Effingham County, where OpenAI must now produce a local compact and an independent noise study before its campus starts running.

A race against the clock: compute demand vs. permitting time

The deeper problem for the industry is time. Demand for computing power is growing faster than anyone can build for it, and the companies are spending hundreds of billions of dollars a year in a race to have the most capacity first. Every month a project spends in review, every referendum lost, every moratorium imposed, is a month in which a rival might get ahead. Powell McCormick’s warning that hyperscalers are now considering building outside the United States is the logical response: if America will not host the infrastructure, the companies will look for countries that will.

That threat should not be overstated. Moving a gigawatt-scale data centre abroad is not like moving a call centre, and the United States still offers the deepest capital markets, the best engineers and the largest customers. But the warning itself is a measure of how seriously the industry now takes the political risk. A year ago, the question was whether there was enough power and enough money to build the AI future. Now a third question has joined them: whether there are enough communities willing to host it. The answer, on Tuesday’s evidence, is far fewer than the industry assumed.

The global spread: this is not only an American story

Finally, it is worth noting that the American backlash is part of a wider pattern. Amsterdam, Dublin, Copenhagen and Lisbon are all tightening the rules in their own ways, for the same reasons: grids under strain, water under pressure, residents asking why their bills should rise to power somebody else’s AI. The technology is global, but its infrastructure is stubbornly local, and every country faces the same negotiation between national ambition and neighbourhood consent.

That negotiation will define the next phase of the AI era as surely as any breakthrough in model design. The companies that win it will be the ones that can show a community, in plain numbers, that hosting a data centre leaves residents better off: bills that do not rise, water that is protected, jobs that are real, and a say in what gets built. The companies that lose it will discover what Meta and OpenAI discovered this week — that the most expensive infrastructure in history is worthless if nobody will let you build it.

What to Watch Next

1. Effingham County, Georgia: OpenAI must publish a local compact for its planned campus and an independent noise study before the site starts operating. Whether it does so, and how residents react, will be the first test of the National Community Compact.
2. The new governors: Texas governor-elect Gina Hinojosa and the governors-elect of Ohio and Georgia take office in January. Watch whether campaign promises of moratoriums and conditional freezes become actual policy, and whether Meta’s hoped-for post-inauguration negotiations materialise.
3. Meta’s spending: Meta has just raised $15 billion in debt to fund its data centre build-out. Its next earnings report and capital expenditure guidance will show whether the political backlash is slowing the pace of construction or merely redirecting it.
4. The new Congress: OpenAI’s blueprint asks lawmakers to turn its voluntary promises into law, including large-load electricity tariffs and a ban on developer NDAs. The Democratic-controlled Congress convenes in January, and the blueprint gives its members a ready-made starting point.
5. The next ballots: Three data centre measures from 2026 are still undecided, and more are already being planned for next year. Every one of them will be watched as a test of whether Tuesday’s revolt was a one-off protest vote or the start of a lasting political movement.

Sources

• The Wall Street Journal — Meta President Says Midterm Results Will Make Building Data Centers Harder
• TechCrunch — Meta warns that election backlash against data centers could make AI expansion harder
• Bloomberg via BusinessMirror — OpenAI to change how it handles data center efforts after outcry
• Techstrong IT — OpenAI Tries Transparency Push as Midterm Backlash and Global Regulations Threaten AI Data Center Boom
• Ballotpedia — Data center-related ballot measures, 2026
• New York Post — Voters reject data centers in nearly 50 ballot initiatives during midterms