After the Data Centers
Data center opponents have assembled a potent political force. But without a more expansive vision, there is a risk that this techlash could be co-opted and fizzle out.
“What will it take to get Americans on board with data centers?” a Democratic Party staffer on Capitol Hill recently asked me. It was a sincere question, not a rhetorical one. Data centers have become a flash point in U.S. politics: Americans are sick of them, and organizers are attempting to block their construction, while a cadre of developers, financiers, tech companies, and energy producers push the deals forward, eager to reap the rewards. The failure to scale up these data centers, they claim, is the bottleneck standing in the way of society-changing artificial-intelligence-enabled prosperity.
This conflict has reoriented the political conversation on AI, which is now focused squarely on how the country should build the physical infrastructure on which the fate of the AI revolution hangs. Some grassroots challenges to these projects are succeeding; at least twenty proposed data center projects, accounting for nearly $42 billion in investment, were canceled in the first quarter of 2026 after facing public backlash. Yet it remains unclear how willing politicians are to stop AI entirely. Crafty elected officials, such as New York Governor Kathy Hochul, may well see election-year benefits to pausing an AI buildout that is disruptive in the short term precisely because it is (purportedly) inevitable in the long run.
A new structure of accommodation is taking shape, with politicians and tech companies converging on the idea that data centers should pay their “fair share,” with ratepayer protection pledges, community benefits agreements, and commitments by operators to bring their own power. Even Donald Trump’s White House recognizes Americans’ distaste for data centers, asking companies to commit to ensuring that AI infrastructure doesn’t raise electricity prices, exhaust local water supplies, or overload regional grids.
All of these developments point to the power of the data center opposition; its organizers have assembled a potent political force, channeling new and popular energies that are now impossible for politicians to ignore. But without a more expansive vision of what to do with this technology—some grander ambition in addition to “no data centers”—there is a risk that this “techlash” will fizzle out, captured by business elites and directed toward policies to manage, not challenge or repurpose, the growth of AI infrastructure.
Beyond the environmental and labor impacts, the bedrock of AI discontent is a feeling that collective life is increasingly unrecognizable, dominated by efforts to automate social relations and synthesize creative expression that seem disconnected from, if not contemptuous of, common human experience. We are told that we now have a breakthrough technology to carry us into a new modernity, even as many feel their own futures constrained by harsh material reality: unaffordability, downward mobility, and the climate crisis. This apparent contradiction prompts an urgent question: If few people like AI, why are there so many data centers?
Data centers weren’t always the focus of elected officials’ AI concerns. Joe Biden’s October 2023 executive order on AI, which he called “the most significant action any government anywhere in the world has ever taken on AI safety, security and trust,” mostly covered testing and protocols for government use. Executive action on AI infrastructure didn’t come until January 14, 2025, six days before Trump’s second inauguration, when Biden released plans to accelerate data center construction by leasing federal lands and streamlining environmental permitting.
The same forces that compelled the lame-duck Biden administration to embrace data center abundance have intensified under Trump, who is speeding up the construction of AI infrastructure in order to secure U.S. technological dominance. Only one day after Trump’s inauguration, the White House announced the Stargate project, a joint venture with OpenAI, Oracle, SoftBank, and MGX that committed up to $500 billion to build U.S. data centers. The flashy announcement was soon followed by a series of executive orders to ease financial and regulatory burdens on data center development and shift risks onto the public balance sheet.
Whatever the ultimate effects of AI, the investments being spent to facilitate the technology’s expansion are already having an impact. After decades of U.S. politicians griping about the bleeding of the American industrial base, private industry has stepped up with massive capital expenditure. Tech companies are on track to spend roughly $700 billion on AI infrastructure in 2026. In postindustrial regions where the jobs left long ago due to automation and outsourcing, the numbers are so eye-watering that the possible contradictions—bespoke industrial production for data centers may exhaust, not reinvigorate, overall industrial capacity—stay back of mind. Many localities and states, until the recent popular pushback, were all too eager to welcome the facilities, competing with one another by enticing developers with fast-track permitting and tax breaks. Although the subsidies trade away billions in sales tax revenue, and the claims about the jobs these developments would create have been severely exaggerated, there is some evidence that data centers can aid local economies by bringing in “capital investment, construction activity, and specialized employment,” according to a National Bureau of Economic Research paper.
In any event, Americans aren’t buying the claims of data centers’ benefits. Seven in ten oppose one being built near where they live, with 55 percent strongly opposed. Residents might reject data centers for a number of reasons. They require very large amounts of water. They emit noise pollution from the persistent humming of HVAC systems that disturbs wildlife and nearby ecosystems. By drawing large amounts of power from electrical grids, they raise residential energy bills. In May, the independent market monitor for PJM Interconnection, the largest power grid operator in the United States, reported that data centers have driven a 76 percent increase in the cost of energy. “The price impacts on customers have been very large and are not reversible,” the monitor said, and “will be even larger in the near term unless the issues associated with data center load are addressed in a timely manner.” Then there are the problems associated with a lack of democratic process. Locals sometimes learn about these projects only by the time they’re already in advanced talks, with city officials and developers sworn to secrecy due to nondisclosure agreements.
The data center opposition is made up of a diverse complex of interests. Riding a wave of unrest among homeowners, tenants, environmentalists, consumer watchdogs, AI haters, conservatives, socialists, and even multinational industrial companies, organizing has slowed down the frenzied construction. Politicized by the arrival of AI infrastructure in their communities, people have won real victories to address their immediate troubles. Not content to disrupt town hall meetings, they are putting pressure on politicians to enact local moratoria to temporarily ban new construction on data centers until people know more about the impacts of the technology and policymakers have had more time to evaluate risks and enact protective ordinances. As a stalling tactic, moratoria may also help communities hedge against an AI bubble, preventing the overbuilding of expensive computing infrastructure that could end up as stranded assets with little organic demand. This local opposition has created headaches for developers, but the larger effects remain unclear. For now, developers are following the path of least resistance, moving projects from sites in Wisconsin and Oregon to states like New Mexico and Texas that are friendlier to development, often feature weaker environmental standards, and suffer from greater water scarcity.
Backers of AI infrastructure have criticized their opponents as yet another NIMBY coalition wielding the esoterica of land-use grievances and environmental impact statements. (More imaginatively, some have described AI opposition as a Chinese psyop.) It is true that the movement shares some connective tissue with more diffuse resistance to economic development projects; a Utah county commission instituted a six-month moratorium on both data centers and utility-scale solar projects, citing concerns about “balanced and sustainable development standards.” Yet the movement has a much larger tent than earlier NIMBY or “slow growth” campaigns, with neither the dog-whistle racism nor the subordination of certain interest groups (such as renters). With a multiplicity of demands and red lines, how should we understand this new localized power?
We might characterize data center opposition as a coalition in media res, offering a real-time glimpse into what cultural theorist Stuart Hall called “the production of politics—politics as a production.” We are seeing how new sites of political contestation, formed out of a convergence of different forces, are “fundamentally open ended,” rarely reflecting “already unified collective political identities, already constituted forms of struggle.” Indeed, some protesters have used generative AI to make signs for anti–data center rallies, a hint that not everyone who is against AI infrastructure finds the technology useless. Facing internal contradictions, can the data center opposition constitute a durable political formation over time? For now, the nascent movement shows how “interests are not given but have to be politically and ideologically constructed,” as Hall argued. Many of their identities and preferences are still forming and “necessarily contradictory.” The tangled-up politics have already led to debates about whether data center resistance will become the next Occupy Wall Street or Tea Party. When the popular masses have such unstable interests, Hall cautioned, “it is possible to recruit them to very different political projects.”
Right on cue, business and political elites are attempting to co-opt popular animus against AI infrastructure, directing it to channels they deem more acceptable. Those channels, however—like developing infrastructure that is “green,” or features community benefits agreements, or puts well-paid union labor to work—are unlikely to address what is animating people’s discontent with AI. Popular resistance to data centers has built on a long-gestating feeling that used to be difficult to make out in a world of invisible code and algorithms, but now has corporeal form: There are forces that structure our lives against which many feel invisible and powerless, and those forces are now prioritizing the construction of overwhelmingly large facilities most people had never heard of until recently. One architecture critic has called AI data centers “the first real major post-human building type.” They are an empty canvas on which people can express their anxieties about job precarity, an unfair economy, surveillance, doom-scrolling slop, climate disaster, spacefaring billionaires, enshittification, and any number of other grievances that often feel beyond our capacity to control.
Of course, there is also the frontal assault that AI presents to labor, regardless of whether the technology actually works. Studies on generative AI’s impact on labor productivity remain mixed, but that hasn’t stopped companies from shedding workers. In April, Meta announced 8,000 layoffs due to the company’s investments in generative AI, while the CEO of Standard Chartered said the bank is planning to cut thousands of jobs because management wants to automate “lower-value human capital.” When business leaders use such cavalier language to describe human beings, is it any surprise that opinion polls show that Immigration and Customs Enforcement is viewed more favorably in the United States than AI?
The hype is beginning to wear thin, as more people recognize how narratives of technological marvel obscure a more straightforward story. As science and technology studies scholars describe it in a recent paper, “heterogeneous capitalist actors have seized the moment offered by a period of uncertainty to mobilize communications technologies and financial relationships to their advantage.” In the process, “they are re-structuring markets in their favor, and directing the flow and capture of capital, resources, data, materials, and labor between sites.”
The tech oligarchs, in other words, are doing economic planning. If AI is “inevitable,” it is neither market forces nor the immutable tides of progress making it so. The technology’s sudden and vast deployment is better understood as a product of planning, organized by and serving particular financial interests. A small number of firms are exploiting their economic and political power to impose a new technology—maybe even a very useful one—in the face of two developments: mass resentment and muted demand for paid (and unsubsidized) AI services. Put another way, the investments to build and deploy generative AI are happening without commensurate market signals of demand. AI lurches forward, as the supply side of the economy shushes broad dissatisfaction and downplays the boggling estimates of revenue needed to justify the financial commitments. It is a reminder, as observed by economists J.W. Mason and Arjun Jayadev, that market production is “in fact organized through highly centralized systems—the command-and-control economy of the corporation, on the one hand, and the planning embodied in the financial system, on the other.”
Having commandeered the market, highly capitalized tech companies are shaping the economy to their own developmentalist visions. If the data centers are a totem of these efforts to reorder collective life, the left has an opportunity to articulate more ambitious demands, focusing less on “AI guardrails” as such and instead building toward levers of economic control and coordination. Throwing sand in the gears through grassroots resistance to AI infrastructure is one thing; transforming that oppositional power to affirmatively shape the economy is another entirely.
What will it take to change the structure of AI production itself—the networks of private capital, critical minerals, semiconductor chips, cloud infrastructure, model developers, and workers—for the public’s benefit? Certainly, it will require massive renewal of state capacity, deployed to pursue more pluralistic ownership models within the AI sector. More than anything, it will require a willingness to transform the logic of private firm operations—something absent in the Trump administration’s much-discussed experiments in state ownership. Under Trump, the U.S. government has cut revenue-sharing export deals with chipmakers, made direct public investments across the technology supply chain, and taken equity stakes in Intel and rare-earth miners. It has debated starting a sovereign wealth fund, constituted by shares in AI model developers like OpenAI. Such actions may appear as proactive, even muscular, state interventions into private markets. But as political economist Ilias Alami has explained, the Trump mutation of state capitalism tends to fortify private capital, alloying it with explicit state support and a drive toward imperial competition. In this political context, developments in the global “AI race” are received as a matter of national security. Breakthrough advances in Chinese large language models, such as those at DeepSeek and Moonshot, compel American AI firms and political authorities to get even cozier to protect politically brokered rents. Absent some new transformational logic, such fusions of the state and private firms are only likely to further “encourage the irrational use and allocation of resources,” directing capital “to build data centres and other AI infrastructure instead of satisfying urgent human and ecological needs.”
Any reasonable alternative, seeking to organize broader forces of production toward the public interest and coordinated growth, will need to wrest planning power from private industry. If capital markets are underdelivering in providing societally beneficial uses of AI (and in productive non-AI investments across the board), the government should invest in and build its own innovation capacities directly instead of derisking private investments for data centers. To pursue technological advancements that are not being met by private capital, the state will also need a means to provide initial funding (as well as guarantee demand in the long run). A National Investment Authority, as a public alternative to venture capital, could make investment decisions to promote varied forms of innovation. Similarly, Aaron Benanav has called for sectoral investment boards, made up of workers, technical experts, and members of the public, to allocate funds across competing proposals. It would be unwise for the government to replicate the capital-intensive machine learning models coming out of Silicon Valley, but that is the point: Public funding should not play the same game but explore forms of technological benefit that are perceived as less commercially favorable and less entangled with material extraction and the exploitation of workers.
Meanwhile, the public can play a more active role in the overall direction of private investment. If private capital is driving the AI sector’s spectacular performance, the rest of the U.S. economy has underperformed. AI spending is crowding out other investments, with nearly 50 percent of corporate bond issuance in 2026 coming from AI companies, diverting demand for U.S. Treasuries, while industrial capacity is routed to accommodate various inputs in the AI tech stack. The result is inflationary pressure and a lopsided economy that teeters on the edge of a possible major market downturn. What’s more, if the world-altering bet on AI “succeeds,” wide deployment and adoption of AI is likely to produce an unequal and precarious labor market, uneven productivity growth (outside the tech sector), and a host of social harms we are only beginning to understand. Credit policy, the twin to industrial policy, could steer private credit flows in the direction of balanced growth of productive capacities, as well as transformative and socially beneficial ventures, while constraining the role of financial speculation.
Whatever the exact policy architecture—equity shares and corporate governance, public ownership of computing infrastructure, regulation of AI as a public utility, or varieties of financial repression—such institutions must provide the grounds for legitimate contestation and democratic participation. American industrial policy efforts, including the late Bidenomics, are technocratic affairs, shaped by the synthesis of priorities between industrialists and well-connected political elites and interest groups. A democratic industrial policy, like the one imagined by scholars Amy Kapczynski and Joel Michaels, requires two major features: administrative power to govern the economy on the one hand, and countervailing power to check the administrative state on the other. Both are needed to realize democratic claims, or to enable “the power of the people, through and beyond government, to meet their collective aims.” In more pragmatic terms, political programs will need to include “concrete hooks for mass movement organizations to make claims on government administrators and funding recipients.”
Such governance is sure to face political headwinds, seeing as it runs directly against the elite fetish for smoothing away frictions in government. Why bother “democratizing” debates about technology investments when no one knows in advance which technologies will prove societally useful? (Of course, this uncertainty cuts both ways. Past the hype and marketing, the AI companies do not know the long-term value of their products any more than the rest of us.) Yet as Matt Davies, a researcher at the Ada Lovelace Institute and London School of Economics, has argued, any effort to center the public in technology will demand moving beyond “the idea that there is some sort of neat policy fix that will eliminate conflict over technology and allow for stable and congenial deliberation.” Indeed, what’s exciting about the data center opposition is precisely its agonistic character. If, at times, it has expressed a rather parochial case against economic development, it has simultaneously articulated a clear-eyed critique of the constellation of corporate surveillance, fossil capital, and U.S. empire that animates our new tech infrastructure.
What does the AI revolt suggest about the future of American political life? Grassroots insurgencies are winning environmental and energy protections, while new and old blocs of capital—hyperscalers, asset managers, and real estate developers—absorb transitional demands as a cost of doing business.
As many await a potential AI crash, the more pressing issue, from the perspective of a democratic society, is the unprecedented mobilization of capital for a labor-automating technology that remains deeply unpopular in the United States. It urges on us questions not so much about this or that redistributive policy solution but instead about our whole conception of politics: If this is not the future we want, what are the institutional forms of power needed to broker and achieve an alternative?
Amid the emergence of AI, the possibility of democratic planning—as a long-term vision to complement organizing around immediate concerns—must be one of the principal hopes of the left. It is time to proclaim more audacious projects to discipline capital and coordinate growth across productive sectors, transforming not only the technology industry but the core of the U.S. economy.
Brian J. Chen is the Policy Director of Data & Society. He writes about AI, technology, and politics.