Several weeks ago, Jason Kelce (football player, podcaster, Taylor Swift’s brother-in-law) starred in an ad asking Americans to pee in jars and send them to their nearest AI data center. The spot – a collaboration between Liquid Death and Kelce’s own Garage Beer – was nominally about water consumption. Mostly, though, it was evidence that hating on data centers has gone fully mainstream.
The polling backs this up. The University of Pennsylvania’s Annenberg Public Policy Center found that opposition to local data centers jumped twelve points in four months this year – from 49% in March to 61% in July. And an August Heatmap Pro poll pushed the number to 75%, with more than six in ten Americans strongly opposed – the most negative result since the outlet began tracking the question a year ago.
Americans are wary of AI data centers, and politicians have taken note (much faster than they usually do). In July, New York Governor Kathy Hochul imposed the country’s first statewide moratorium. Weeks later, Republican Texas Governor Greg Abbott – a longtime data center recruiter – paused new grid connections pending an audit. And just this month, Pennsylvania Governor Josh Shapiro – who celebrated Amazon’s data center investment as recently as last year – signed an executive order marketed as the nation’s strictest guardrails on the industry.
But while opposition to data centers grows, the labor movement remains divided, for obvious reasons. The data centers are creating jobs (at least, right now), while AI threatens to eliminate jobs in the future. And those jobs largely belong to different sets of people.
The Building Trades
So far, the building trades have been supportive. In May, IBEW urged its members to lobby Congress against any ban on data center construction. Several weeks ago, Don Slaiman of IBEW Local 26 took to the New York Times to argue that data centers have delivered tens of thousands of well-paid blue-collar careers, and that communities should be welcoming them. It’s not hard to see why. Data centers now account for 2.3% of all US construction spending, and spending has more than quadrupled since 2020.
Believe it or not, tech companies often use union labor on their projects (despite their best efforts not to; I wrote recently about Meta’s attempts to skirt union labor in data center construction). An Associated General Contractors survey estimates that data center construction is about one-third union, which is three times the construction industry’s overall union rate. And according to reports from the trades (so take it with a grain of salt), this is growing. Union locals report doubling apprentice classes and expanding training centers to keep up with demand, and North America’s Building Trades Unions (NABTU) hit record membership in 2025. In March, NABTU announced partnerships with OpenAI, BlackRock, and Meta; in April, NABTU signed a labor agreement with OpenAI and Oracle to build the multibillion-dollar Stargate campus. Roughly 700 union tradespeople are already working on the site in Michigan, with thousands more expected to join.
The leaders themselves aren’t shy about praising the job opportunities proffered by data centers. Robert Wilson of IBEW Local 7 in western Massachusetts described a proposed hyperscale facility as “a generational type project for our geographical area.” “At a time when educators are asking young people to consider the building trades over college,” he told me, jobs like these “could enable many young workers the ability to save for a down payment on their first home.”
Rob Bair, president of the Pennsylvania State Building and Construction Trades Council, made the same case: data center work “has allowed members steady employment with overtime work which allows them to live the middle-class lifestyle,” plus union-managed healthcare and a real pension waiting at the end. And Dan McConnell, business manager of the Minneapolis Building and Construction Trades Council, told me his members “take pride in building the infrastructure that keeps Minneapolis competitive.”
Unions Against Data Centers?
And still, there is another wing of the labor movement that wants to pump the brakes. National Nurses United formally endorsed the Sanders-AOC AI Data Center Moratorium Act. NNU co-president Jamie Brown called the unchecked growth of data centers “a public health crisis, an environmental crisis, and a workers’ rights crisis.” The American Association of University Professors endorsed the bill too. And the opposition runs deeper than national endorsements. Thomas Meyer – deputy political director of Food & Water Watch, which organizes against data center projects – rattled off local examples: a teachers union and UNITE HERE local in Seattle (UAW 4121 later joined), CWA locals in New Jersey, UNITE HERE in Philadelphia.
Caught in the middle are some of the labor movement’s most prominent leaders. Association of Flight Attendants-CWA president Sara Nelson stood with Sanders and expressed support for a slowdown, but AFA hasn’t formally endorsed the moratorium bill. American Federation of Teachers president Randi Weingarten stood at the same Sanders press conference, and then announced that her union does not officially support the moratorium. Two of the most recognizable leaders in American labor were willing to stand behind Bernie Sanders at a podium, but not behind his bill.
Colorado state senator Cathy Kipp has had a front-row seat to this divide. She recently proposed a bill regulating data center construction. The bill drew opposition from construction unions like the pipefitters and the IBEW; some derided it as the “Wyoming Jobs Act.” As in, the jobs would flee 40 miles north to Cheyenne. The state AFL-CIO opposed the bill too – though Kipp says the federation had factions pulling in different directions.
Her bill died several months ago. Since then, she told me, the conversation has shifted. At a conference earlier this summer in which local leaders were expressing support for new data center construction, Kipp stood up and cited a poll showing 91% of Coloradans want guardrails on data centers. “You guys cannot put me in the position of: is it you or my constituents?” she recalls telling the room. Several other legislators echoed her. Afterward, several labor leaders came up to thank her. Privately. “Some of them might be more nervous about saying it publicly, but it’s changing. A lot of people understand that their jobs are potentially going to be replaced,” she says.
To be clear, this has not turned into any kind of inter-labor war. At least, not publicly. Even Meyer – who professionally organizes against data center projects – told me he has “no quarrels with a union or any other organization working on behalf of their members.” McConnell, from the other side, expressed a similar sentiment: he respects the unions that support moratoria. In his words: “this should not be framed as one group of unions against another.”
In fairness to McConnell, when I began this research, I wondered if that’s what I might find. But I didn’t. The union leaders who support data center construction are about as wary of AI as the rest of us. But they’re also up against cratering public opinion, combined with some misinformation.
Consider the belief that AI data centers consume inordinate amounts of water – an idea so memed it made its way into a Jason Kelce ad. Wilson described a city council meeting in Westfield, Massachusetts, where a city councilor gave a presentation arguing that a proposed facility would destroy the aquifer beneath the city. The facility in question would be cooled by a closed loop of glycol – the same fluid, recirculating indefinitely – and wouldn’t draw water from the aquifer. When the public was invited to comment, “the developer for the proposed data center tried to address the concerns with water but the council president immediately shut him off and said the discussion could only be about the moratorium,” Wilson told me. “As an observer at the meeting it appeared the one year moratorium was going to be pushed through without allowing any debate.”
Wilson is correct: the concerns about water consumption are often overstated, though not negligible. The most widely cited estimate, from Lawrence Berkeley National Lab, puts data centers’ total water consumption at roughly 230 billion gallons in 2023 – a figure that includes the water consumed by the power plants feeding them, which is most of it. This amounts to less than 1% of national water consumption. As Robin Gaster of the Information Technology and Innovation Foundation wrote in July: “there is no universal national water problem.” The concern, of course, is that the number is growing, and that some areas are higher risk than others. We certainly cannot write off the concerns about AI’s water usage. Still, given how far the water consumption idea has spread, it’s understandable why a union head in an area that’s not at risk for drought would be frustrated.
Wilson also called worries about bright lights and heat islands “nonsense” (data centers, he noted, are windowless warehouses full of servers). McConnell made the same point: many of the impacts people fear – water consumption, noise, higher utility rates – are design choices, not inevitabilities, which is exactly why he wants enforceable standards. In his words: “Strong rules do not block good projects. They help us identify which projects are actually good.”
Several leaders also noted that data centers face an unfair critique about their impermanence. Construction work is, by nature, temporary. Nobody faults a hospital because the people who built it were eventually out of a job. As McConnell told me: “Lawyers, consultants and many other professionals build stable careers by moving from project to project; the building trades are no different.”
Support for Regulation
At the same time, the labor leaders I spoke with were hardly AI cheerleaders. Bair’s working groups in Pennsylvania are negotiating agreements to limit AI in call centers, hospitals, and classrooms. In his words: “We realize there is a time and place for AI but not at the expense of employees. AI is accelerating the displacement, but we’ve seen it coming in many shapes and forms.” McConnell told me his members raise concerns about AI, but that he doesn’t think moratoria are the solution. “Stopping construction by itself does not create an AI policy or protect workers from automation.” Even his members’ support for the buildout has limits, he was careful to add: “enthusiasm does not mean blind support.”
It’s important to mention that their skepticism collides with a lack of options. The federal government has spent decades declining to seriously invest in our country. Biden’s Bipartisan Infrastructure Law was a partial step, but it was nowhere near enough: even after it passed, the American Society of Civil Engineers still projected a $3.7 trillion gap over the next ten years between planned investment and “a state of good repair.” And that assumes current funding levels hold; the law’s authorizations expire in 2026. There is no New Deal on offer to employ the trades for a generation.
And so, the trades are stuck in an impossible position, one they’ve been in before. In a sense, none of this is new. Emergent technology often requires a significant upfront labor investment while threatening jobs in the long run. In the words of Todd Vachon, a labor studies professor at Rutgers: “Any legislation that phases out one form of technology in favor of another is going to draw opposition from the union supporting the technology being eliminated while the union representing workers manufacturing the new technology will be supportive of it.” Anne Lofaso – a law professor at the University of Cincinnati – cited coal as “a very tight analogy.” The well-paid union mining jobs were always partly myth: many weren’t union, many weren’t safe, and the jobs disappeared as the technology improved. “Data centers are similar,” she told me: a promise of good union jobs, hazards for the workers, environmental costs for the communities making the greatest sacrifices.
But there’s only so far historical analogies can go. In Vachon’s words: “What feels different with AI is that this friction is pretty broadly felt and breaks labor into two distinct categories of who is and who isn’t threatened by AI and it mostly falls along the line of blue-collar vs white-collar workers.”
Vachon hit upon something that’s always bothered me about the data center backlash. Technology – almost by definition – has always replaced jobs. And for centuries, the professional class largely filed it under progress. Inevitable, even. But now that the automation is coming for people who went to college, pausing it has become a mainstream idea. If lawyers were asked to give up a decade of work to protect electricians’ jobs, would they? History suggests probably not.
That doesn’t mean we don’t need a pause. But construction isn’t the root of the problem. The root of the problem is AI. People don’t like it, don’t trust it, and don’t know what it’s going to do to our society. Data centers have existed for decades; we don’t see widespread protests outside a hospital’s server room (the erroneous conflation of all data centers with AI data centers was cited by multiple union leaders). In the same Annenberg survey, 68% of Americans said the government has done “too little” to regulate AI. A July Emerson College poll found 63% of voters concerned about AI, versus 14% excited. For Meyer, the reason so many workers oppose data centers – including, per one recent poll, a majority of union households in New York – comes down to AI itself. Specifically, who gains and who pays. “The benefits are so concentrated and the costs are so massive,” he told me. “People don’t like big tech billionaires trying to bully or bribe to get what they want.” The people he talks to “feel like AI is being forced on all of society,” and see fighting data centers locally as “the best way to make an impact and exert some agency.”
People are scared of AI, and we’re so far away from even the most basic regulation. In July – just days after OpenAI disclosed that two of its models escaped a testing environment and hacked another AI company (cool!) – Congressmen Ted Lieu (D-CA) and Nathaniel Moran (R-TX) introduced the bipartisan AI Kill Switch Act, which is exactly what it sounds like: a requirement that the most powerful AI systems be built with an off switch. Imagine Ford trying to sell a car without a brake. I generally hate the term “common-sense regulation” (who doesn’t think their own ideas are common sense?), but come on. This is common sense. A month later, the bill has gone nowhere. It’s challenging to even talk about AI without answers to basic questions like “How do you turn it off?” “Is it a bubble?” “Will it tank the economy?” “Will it channel wealth exclusively to the top 1% while everyone else suffers?” “Will it kill us all?” AI regulation is about far more than just the existential threat: it’s also about whether it will perpetuate America’s destructive and expanding economic inequality.
There is no mystery about why we haven’t yet regulated the AI companies directly. In what I believe will be remembered as the most iconic image of Trump’s second term, tech CEOs lined the second row of his inauguration, ahead of his own incoming cabinet. They had paid for the view, a million dollars apiece to the inaugural fund, and the corruption has only continued since then. The return on investment has been excellent. Republicans tried twice to attach a ten-year moratorium on state AI laws to must-pass legislation; when Congress wouldn’t cooperate, Trump signed an executive order directing the DOJ to challenge state AI laws and threatened to withhold federal broadband money from states that regulate the technology. The federal government is not merely declining to put guardrails on AI; it is actively dismantling the ones states try to build.
We are a country understandably scared of AI, with a federal government unwilling to regulate it. And so, we’ve found the only available option: halting the AI data centers. It’s a smart strategy: stopping construction is famously easy in America, just ask Ezra Klein. Unlike regulating Trump’s buddies, it’s working: in the first quarter of this year alone, community opposition blocked or delayed at least 75 data center projects worth roughly $130 billion, about as much as all of last year combined. The industry is expanding fast enough that record construction and record obstruction are happening simultaneously. And public opposition keeps climbing, as support for a moratorium grows.
But it’s worth getting specific about what a moratorium entails. The Sanders-AOC bill would freeze construction of new large AI data centers until the government passes federal AI guardrails. And I definitely support those guardrails (I would go so far as to call them “common sense”), but we should be honest: realistically, that isn’t happening before January 2029. Under the proposed moratorium, existing facilities would keep running, and the white-collar workers at Anthropic and OpenAI would carry on. The pause lands entirely on the people who build. In short: construction workers are being asked to pay for the sins of Sam Altman, because the government is too corrupt to regulate him directly. McConnell calls the moratorium “a very blunt instrument,” and he’s right. The question is whether anything sharper is available.
A Kill Switch
I’m torn myself. I understand the building trades’ position, and I understand what a national moratorium would mean for their members. But I’m also very uneasy about AI, and I see the argument for using anything in our toolkit to slow it down, blunt or not. In Lofaso’s words: “As someone who likes to see scientific progress, I would like to see unions negotiate job retraining, social assistance, etc. But AI feels different and potentially more dangerous. I wish we could have a moratorium just to think through the ramifications of AI and how to create safe AI and protect workers and their jobs.”
Still, I found cause for hope. The version of the fight that ends well is already underway. The trades are asking for guardrails themselves. Bair’s council backed the principles behind PA Governor Josh Shapiro’s order regulating data center construction. Wilson wants rates and water regulated by law, not by promise. McConnell wants project labor agreements, ratepayer safeguards, and enforceable community benefits. The tech companies had a better chance of skirting union labor before their popularity cratered: our AI overlords are now on their back foot, scrambling to get communities and elected officials to support data center construction, which means it’s the ideal time for the labor unions to make demands – about safety, job protection, profit distribution, and more.
Vachon suggested something similar: that labor exchange its support for guardrails. In his words, they would say: “We’ll only build these if we can be reasonably assured they won’t kill us.” That sounds good to me; I would prefer AI not kill us all. I don’t want to suggest that communities should stop fighting unwelcome data centers. But union leadership can be an important ally in this battle, even leaders who don’t support a moratorium. This Congress appears unwilling to pass a bill requiring AI to come with an off switch. But the trades control something the industry can’t route around: whether the infrastructure gets built. Right now, that’s the closest thing to a kill switch this country has.
Ginny Hogan is an NYC-based writer. Her new book, "Obey in Advance," co-written with Carlos Greaves, is out in September.
More
Please share widely! And if you republish somewhere, please make sure to attribute the original to Labor Politics, thanks.
If you’re a worker, writer, or labor organizer interested in submitting a piece like this to Labor Politics, please shoot me a DM. Keep in mind that this blog is mostly interested in articles that deal with big strategic questions for unions or Left politics (not just reporting) and that unfortunately I have to decline many pieces because of a lack of fit or lack of space. But I’d love to make this a platform that labor folks and labor-inclined lefties can continue to periodically intervene and debate. And many thanks to Ginny for submitting such a well-researched and thoughtful piece!
i'm peripherally attached to the trades and this absolutely gets at the issues faced by local unions. part of what we need to see campaigners doing - especially when communicating with trades members - is making the point clear that our rulers are offering us either homes for GPUs that steal the work of society, cook the planet, and promise to destroy our jobs, or nothing. This is especially salient for tradesworkers who are told by capital, and the government that represents it, thanking making real the fever dream of data centers as far as the eye can see is more important than infrastructure to improve our housing, transportation, and energy systems.
Thank you for this article. I would add that data centers currently use a huge amount of energy much of which is created by burning fossil fuels. The resulting climate change impact (not to mention energy bill increase) is not good for working people (to put it mildly). Guardrails could theoretically solve this, if we can require data centers to provide clean energy and force them to pay for it.