AI First: How the Federal Government is Prioritizing AI Over People and Planet

Executive summary

The recent explosion of artificial intelligence (AI) and the construction of data centers threatens to swallow the US economy and transform society. The stock market and its AI-driven rally has dazzled investors with an endless wave of asset price appreciation. Yet despite all of its science fiction hype, the actual implications of the AI buildout are increasingly socially and ecologically destructive. Coupled with what projects to be long-term global economic and supply chain disruptions stemming from the US-Israeli war on Iran, the rise of AI portends increasing precarity for the global working class.

We believe that AI’s emergence shows the potential for state capacity to be oriented toward a different mission that centers the ambitious creation of socially useful green infrastructure like clean energy, healthy schools, libraries, social housing, and public transit. While we recognize that it is tremendously difficult to mobilize sufficient capital toward pro-social ends, the state-supported AI buildout offers a glimpse of the possibility that a transformed state could orient its significant market-shaping and worldmaking capacity toward the public good. 

The AI and data center buildout is not the spontaneous outcome of an imagined “free market.” It is the result of political decisions past and present: subsidies, tax rates, infrastructure, deregulation, and an absence of regulation in the first place. Instead of anything resembling public control over economic and industrial planning, we are witnessing a corporate takeover of state capacity. 

In this report we detail three key ways in which the federal government is facilitating the AI boom:

As data centers spread across the United States, inequality rises, and climate and economic crises build, the ginned-up urgency of winning an AI race that most of us do not even want to be running in the first place needs to be named for what it is: an industry-led hype cycle that aims to supplant democratic control with investor prerogatives. Wresting the power of worldmaking away from the billionaires and into the hands of working people—and a government that puts people before profits—is urgent and necessary. 

We reject the “AI race” as a productive and socially beneficial use of resources. Instead, we advocate for a “race” toward a liveable world (ecologically and economically) where abundance is not measured in asset prices alone, but in a more capacious, multi-dimensional sense of what working people actually need and want. 

Across the United States, there is robust, well-organized, and successful organizing in local fights against AI data centers. In 2025 alone, local opposition contributed to the blocking or stalling of 48 projects worth $156 billion, often on the grounds of opposing additional pollution and lavish public subsidies to feed corporate profits. The alignment between public consciousness and the early but already widely documented harms of unfettered AI and data center deployment means the moment is ripe for a more transformative vision that drastically increases the federal government’s capacity to make choices for working people as opposed to working for AI and elite investors. CCI’s “Green Economic Populism” framework is an example of such an approach, one which aligns policies that bring immediate relief to working people with building a dynamic public sector that can enact the robust regulations and enormous green investments needed to realize a better future for all.²

As the data center boom shows, a state-capital partnership can rapidly roll out massive, world-shaping infrastructure when interests and motivations align. We want a government that directs collective resources and energy toward an agenda that serves people and the planet. This will require governing power and authority wielded for the public interest, with a focus on jobs with dignity, environmental stewardship and a green transition, and robust social welfare programs. This can only be accomplished with a means of democratizing investment decisions and developing a planning calculus that is insulated from corporate influence and so as to balance multiple, sometimes conflicting, objectives. 

We conclude by sketching out five pillars of a green and democratically planned policy agenda that abides by shared environmentally and socially just values and prioritizes human dignity.

Introduction

Concerns around rising electricity prices, air and noise pollution, and strains on local resources are precipitating a groundswell of campaigns to resist the construction of gargantuan, planet-warming data centers. There is increasing backlash against data centers and AI emerging from communities that are being asked to bear these burdens in exchange for local economic activity with a handful of permanent jobs while being offered little say in the matter. And what for? Behind the science fiction hype that AI will somehow birth a superhuman intelligence, workers fear and capitalists hope that AI will displace, disempower, or deskill wide swathes of the labor force. AI is getting generous public subsidies to erode public institutions and debase critical thinking capacities while locking the United States into massive, planet-warming energy uses that will make it even more difficult to transition to a sustainable future. 

Aided by generous federal policy, major tech companies have inflated in valuation in the last few years, and the relatively small number of publicly traded, AI-focused companies have accounted for a large portion of recent stock market gains. As JP Morgan’s Michael Cembalest explained, “The market cap of four hyperscalers and the semiconductor ecosystem companies they rely on has grown from $3 trillion to $18 trillion in just a few years, and a broader group of 42 AI related companies has generated 65%–75% of S&P 500’s earnings, profits and capital spending since ChatGPT’s launch in November 2022.”³

Meanwhile, wide swaths of institutions and firms (including the federal government) are racing to implement AI throughout their operations. This rush toward adoption, often without regard for proven utility or effectiveness, is largely driven by industry hype, cost-cutting desires, and groupthink among investors and management. The widespread incorporation of AI tools into workflows is generating nowhere near enough revenue to make up for the massive sums that AI firms are spending, yet valuations continue to inflate. People with significant appreciating assets in stocks, investments, and real estate may be mostly content with their rising portfolios, but the working class is increasingly exposed to the precarity of deteriorating labor market conditions, the loss of personal freedom through increased surveillance, and a host of increasingly unaffordable essentials.⁴

This is all occurring as the Trump administration pours gasoline on our climate and ecological crises with its unabashedly polluter-friendly agenda. The administration has repealed most of the climate and energy investments of the Inflation Reduction Act (IRA) and moved to stymie renewable energy construction, while gutting environmental regulations on greenhouse gases, environmental review, hazardous chemicals, air pollution, land use, and more. These massive AI data centers—hungry for land, energy, and water—are being built at a time when important environmental laws are being weakened and their enforcement is deliberately being neglected. The result is that federal policy is letting data centers and their supply chains shift more health, ecological, and infrastructure burden onto workers and frontline communities. Tech companies have been happy to take advantage of the opportunity to operate without social or environmental guardrails.

The confluence of corporate interest and favorable government policy is driving hype for these technologies. The AI industry is spending lavishly on a sector-wide marketing push to justify the expansion of AI into every facet of life, frequently invoking the specter of “artificial general intelligence” and claiming that AI will solve all manner of social problems (from cures to cancer to “solving” climate change). At the same time, the federal government backs up the industry’s claims to indispensability with rhetoric about how winning the “AI race” is essential to national security. But underneath the sales pitch lies an all-out effort waged by a segment of the ruling class to legitimize and entrench their dominance through sowing widespread dependence on technological tools that they control. They are both usurping and destroying already-strained public institutions and public infrastructure and swallowing up finite natural resources, clearing the pathway for AI deployment that these private companies have chosen as one that undermines the public interest. 

The federal government’s support for AI is wide-ranging, from procurements and financial derisking to regulatory streamlining and cutting jobs, strong-arming state and local compliance, and dismantling institutional oversight. Some of these tactics go back decades, while others have emerged in unique form with the Trump administration going all-in on AI as a core part of their “America First” strategy under the guise of national security and geopolitical dominance.⁵ The ways that the government is prioritizing AI reshapes who bears costs and who controls resources. It also influences whose claims on the government to steer industrial policy, investments, and resource allocation are deemed legitimate.

There seem to be endless ways to manufacture financing for Big Tech’s ambitious plans.⁶ But it is much harder to “manufacture” energy infrastructure. Access to electricity is emerging as the main obstacle to data center expansion, and Big Tech is quickly becoming agnostic about its energy sources as it rushes to get tons of new natural gas generation capacity built to power its data centers.⁷ Not long ago, the titans of Silicon Valley were championing clean energy and climate action; now Google has shifted from already shaky claims of climate neutrality, while Microsoft’s emissions have grown by a third since 2020.⁸

What is AI, and what is a data center?

The AI industry is an amalgamation of different sectors and technologies, the biggest names of which include Amazon, Meta, Nvidia, OpenAI, Anthrophic, and Google/Alphabet. The “AI technology stack” is oriented around business models where success hinges on military and domestic surveillance, the automation and algorithmic decision-making of public services, and widespread workplace deskilling and labor-force reductions.

The term “artificial intelligence” itself is ambiguous and is often used to refer to a wide range of different technologies that entail algorithmic collection and processing of data to generate outputs. This can include text-generating chatbots, automated pricing algorithms, facial recognition systems, and much more. While AI development has been ongoing for decades, the current boom arose as a result of significant advances in semiconductors and improvements in generative AI, specifically the commercial emergence of large language models (LLMs) and image generators like OpenAI’s ChatGPT, Meta’s Llama, Anthropic’s Claude, and Google’s Gemini. The training and running of these AI systems occurs in hyperscale data centers, which are warehouse-like buildings that house rows of server racks loaded with computer chips, memory, and networking, power, and cooling equipment. Conventional data centers generally use central processing units (CPUs) to process data, but AI computing primarily relies on more powerful graphics processing units (GPUs)—though AI data centers are also securing whatever CPU supplies they can get. The production of these more technologically complex chips requires substantial amounts of water, energy, and transition minerals, and shorter chip lifespans (two- to six-year active lifespan when used for intensive AI models depending on use case).

The overbuilding of AI data centers

Size matters. There are estimated to be more than 4,000 data centers currently operating in the United States, with thousands more under development.⁹ While data centers have been around for years providing various cloud network services and data storage, they are increasingly intended for AI training and usage. This means that AI data centers are significantly larger in size, energy consumption workloads, operational complexity, and cost.¹⁰ A subset of these are operated by “hyperscalers,” which are “large-scale cloud computing providers that operate massive data centers to support global digital services,” such as Microsoft, Google, and Amazon.¹¹ Conventional data centers have an energy demand of approximately 5–10 megawatts (MW) while a hyperscale, AI-focused data center requires at least 100 MW, consuming as much electricity annually as approximately 80,000 households.¹² Meta’s Hyperion AI data center campus in Richland Parish, Louisiana, is 5 GW, which is more than enough electricity for every home in the state of New Jersey.¹³ The company is paying to power the facility with 10 new gas-fired power plants.¹⁴

While conventional data centers demand up to 10 MW of energy, Meta’s latest AI data center demands 500 times that amount.

As access to energy becomes the most formidable hurdle for hyperscalers, projections and financing deals for new data centers continue as if there are no physical constraints. While no official data is yet available, estimates put data center electricity usage around 4 to 5 percent of total US consumption today, and that number is projected to increase to between 9 to 17 percent by 2030 according to the Electric Power Research Institute.¹⁶ The data center buildout is creating years-long backlogs in key energy infrastructure such as high-voltage and distribution transformers and switchgear systems.¹⁷

Without something akin to a “wartime mobilization,” there is no conceivable way for the scale of projected data center growth to be serviced by the current US electricity grid. In addition, core industrial backlogs have been exacerbated by the war on Iran, with shortages in large-frame gas turbines and other precision machines needed for processing and generating energy from natural gas.¹⁸ Manufacturers were already at full capacity with backlogged orders, and now must add replacements for destroyed Gulf infrastructure to their queues. In fact, even with sufficient intent, it is not clear that the physical industrial capacity exists to meaningfully speed up production of some of these key infrastructural elements.

AI that deskills, rather than empowers

While AI has begun to generate billions of dollars in revenue for key firms, the sector is still spending far more than it is taking in with the hope that massive investments will secure market-dominant positions that eventually pay off in the long term.¹⁹ One estimate assesses that “hyperscalers have invested over $560 billion into AI technology and data centers between 2024 and 2025 and have reported revenues of just $35 billion.”²⁰ Their anticipated profits, should they materialize, will require a customer base willing to pay significantly more for AI services (or willing to accept a lesser product overrun with ads). When it comes to enterprise customers, these costs will only be justified if AI delivers either expanded productivity and revenues or reductions in labor force expenses. This last point cannot be emphasized enough: a core part of the AI value proposition to its customers (and investors) is a promise to displace workers with AI programs that can do the same work—or close enough—for lower costs. Once dependency on these programs is firmly established, particularly within firms, public organizations, and other enterprise customers, AI service providers can ratchet up prices.

If successful, AI firms will be responsible for putting many people out of work, whether by outright replacing them or through mass deskilling that leaves fewer workers forced to deploy AI tools over larger swaths of responsibilities. Instead of one journalist writing one thoroughly and carefully reported article—supported by a team of editors, fact-checkers, designers, and more—imagine one person frantically editing dozens of AI-generated pieces of dubious value and churning them out in the same amount of time for less pay. This elimination of jobs and erosion of worker power could occur even if these tools underdeliver. 

This is not the only possible direction that AI development could take. But without any significant social or ethical guardrails, the most socially beneficial applications of AI technologies will remain few and far between.²¹ Instead, it is likelier that we will continue to see AI deployed in ways that further concentrate power in the hands of those who already have it—entities like employers, landlords, law enforcement, and dominant corporations—while pushing negative externalities like increased air, water, and noise pollution onto communities, and raising consumer costs of everything from electricity to algorithmically priced-up products of all sorts.²² The tech industry has chosen a path for AI development that values scale and market domination at escalating costs. With its ballooning energy demands overwhelmingly met by polluting energy sources, the AI industry has proven its willingness to sacrifice climate change mitigation and the clean energy transition to its own rapacious appetite for power.²³

Meanwhile, misinformation, psychosis, erosion of critical thinking abilities, mass surveillance, the further degradation of the quality and usability of the internet and software programs, and other harms are proliferating under the AI industry’s current business model.²⁴ Merriam-Webster’s 2025 word of the year was “slop,” a reference to the increasingly ubiquitous LLM-generated content polluting the internet. AI is adding to and intensifying labor for workers as they increasingly take on a greater range of tasks with less focus and are incentivized to produce “workslop,” or “AI generated work content that masquerades as good work, but lacks the substance to meaningfully advance a given task.”²⁵

With such a potently negative (current and promised) impact on the lives of most people, it is not surprising that concern about and disdain for AI and data centers is one of the few truly working class issues that spans the political spectrum.²⁶ Polling consistently finds that US residents are much more concerned about AI than excited about it and desire greater regulation and control over these technologies, with opposition to data centers rising dramatically.²⁷ A recent NBC News survey found that 46 percent of voters feel negatively about AI while only 26 percent feel positively, and 57 percent believe that the risks of AI outweigh the benefits while 34 percent believe the benefits outweigh the risks.²⁸

So long as AI is owned, operated, and developed by profit-maximizing firms with an absence of a countervailing force of strong and social-movement-oriented labor organizations, productivity improvements will primarily benefit employers who can use them to justify workforce reductions. While tech moguls will likely continue opining about a guaranteed basic income or shorter work weeks, it is difficult to imagine, for example, that if roads become packed with self-driving trucks and delivery vehicles, their owners will pay drivers to stay home and assist with job transition. Even if there was some way that displaced workers were justly compensated with a guaranteed income or other form of social wage, it is hard to deny the likelihood that any benefits from this trajectory of AI development can be derailed by the unintended creation of a largely uninhabitable world.

AI First: manufacturing a global AI race that most people do not want to be running

In some respects, support for AI is not out of line with the United States’s long history of policy enabling and facilitating the development of cutting-edge digital and information technologies and encouraging the consolidation of economic power in a small set of oligopolistic firms. This active shaping of technologies and markets sets the stage for the tech giants at the heart of the AI and data center boom to claim that AI and the technologies that manage, make sense of, and disseminate information are essential to military and economic dominance. This has intensified since the first Trump administration, as Silicon Valley has become an increasingly core plank of the US economy, and US foreign policy has turned to even more technologically complex forms of weaponry.

The industrial mobilization that supported World War II, the space race, the internet, biotechnology, and the fracking boom all represent different chapters in the long story of state-capital collaborations to create the technologies that shape the world. AI is the latest iteration of this process. What seems unique to this moment is how this new state-supported technology’s most palpable effects are to increase the cost of living; degrade the quality of the creative economy, information environment, and education system; and threaten large numbers of workers across sectors with wage stagnation, slowed hiring, reduced bargaining power, deskilling, or unemployment.

AI’s value proposition lies in its potential ability to “surveil, discipline, and displace” working people.²⁹ The government's interest in these technologies deepens ties with dystopian visions of surveillance and autonomous warfare capacities under the auspices of continuing US technological dominance and hegemony, as starkly evidenced by how the Trump administration has used AI extensively in its brutal campaigns of mass deportation and the war on Iran. While the US government’s post-9/11 affinity for mass surveillance and curtailing of individual freedoms and rights are well-established and predate this current technology cycle (e.g., the Patriot Act), AI is undeniably an accelerant. The “AI race” framing is doing important ideological work toward the erosion of civil liberties and democratic institutions, with serious material consequences to people’s everyday lives.³⁰

As economist Mariana Mazzucato has shown, there is a long history of state support for technological innovations that ultimately end up benefiting private entities.³¹ What we see happening now goes much deeper than simply relying on the state as a source of seed funding. Some of the most lucrative and foundational forms of state support for AI involve the integration and coordinating control of venture capital with the state, leading to a situation where those most heavily invested in the AI technology stack can simply help lobby, shape, legislate and regulate their market into existence. Writer Sinead O’Sullivan calls this “venture developmentalism,” a process “where the instruments of capitalism serve as extensions of industrial strategy, and industrial strategy doubles as asset protection.”³² The result is that dominant firms and their investors direct national resources toward their preferred technological future instead of the public interest.

The AI industry has done a remarkable job at positioning itself as an influential player at the heart of federal policymaking. Part of this strategy is to entrench their AI technologies in all aspects of government operations, from bureaucratic management to military infrastructure.³³ For example, OpenAI’s Chief Global Affairs Officer wrote two open letters in 2025 outlining their firm’s desired federal AI policies.³⁴ OpenAI wants the US government to provide them financing, skilled workers, and electrons, stating that they want to “ensure that frontier AI systems protect American national security interests, including through federal agency adoption.”³⁵ OpenAI’s “freedom-focused policy proposals” are structured around a supposedly existential battle between the United States and China’s “[Chinese‬‭ Communist Party]-built autocratic, authoritarian AI.”³⁶ Firms like OpenAI are positioning themselves as key players in a new “space race” that the United States cannot afford to lose in exchange for license to continue their unbridled growth. With significant financial and political power, and a national economic strategy pinned entirely on AI infrastructure, these companies might very well gain the deregulation and agenda-setting concessions they are seeking.

The industry’s capture of government policy and public investments/expenditures is evident in the evolving federal framework and signals to market actors that there will be continued and unquestioned government support for AI. Major investors in AI firms helped write significant Trump administration executive orders related to AI and have been appointed to senior White House roles, while venture capital firm Andreessen Horowitz’s lobbying spree and massive presence in DC has earned it “veto power over any AI-related proposals.”³⁷

The government has been shown to be completely tamed by the investors calling the shots and defining our socio-technical futures. As we will explore in more detail below, this great derisking involves de- or non-regulation in antitrust, environmental, and social policies—paired with fast-tracking of access to public resources in energy, land, and water in the name of national dominance. Anticipated future market dominance by a select few firms seem all the more likely.

The AI technology stack has been dominated by a small number of astronomically large and powerful corporations who have jettisoned their veneer of ethical and humane ambitions—from safety to climate targets to bans on military applications—to fall in line with the raw power-grabbing geopolitics of the Trump administration.³⁸ As these corporations accumulate economy-warping power, the government is largely forgoing any opportunities to set consumer, environmental, or public safety guardrails on these technologies. 

Fiscal tools—from a whole-of-government approach to embed and procure AI across the federal government to direct equity investments in technology companies and resource development—are currently being mobilized under the banner of building a new digital empire: AI dominance facilitated by a “Big AI State.”³⁹ Framed as a means to cost-containment, geopolitical strategy, and national economic progress, this shift in government policy and practice functions as a massive transfer of public resources and power to private firms and investors. 

In particular, AI’s dominance is being cemented through the degradation of an already deeply flawed approach to antitrust regulation. This approach has focused on consumer welfare (and specifically, low prices) as the sole arbiter of healthy market competition, and it has generally ignored predatory practices to shut out competitors and establish impenetrable market dominance as Amazon and Google have done.⁴⁰

The Biden administration made some relatively modest efforts to check the power of large tech firms. Specifically, they revitalized enforcement authority by the Federal Trade Commission (FTC) and Department of Justice (DOJ) Antitrust Division, including litigation against Apple, Amazon, Microsoft, Facebook and Google (which they won) on monopoly and anticompetitive charges.⁴¹ However, these efforts were hampered not only by litigation timelines that outlast administrations, but also by a sometimes conflicted judiciary and outdated interpretations of antitrust law. Up against time, an entrenched status quo, and acquiescent Democrats in Congress, the Biden administration made limited strides in undoing decades of corporate and ideological capture of the US government to target the tech oligopoly.⁴²

Since taking office, the Trump administration has opted for a softer approach to antitrust and other corporate regulation, including policy related to AI. This includes an executive order, “Removing Barriers to American Leadership in Artificial Intelligence,” on the fourth day of the term and a “National Policy Framework for Artificial Intelligence” released in March 2026 with legislative recommendations intended to accelerate AI development and “remove barriers” for industry.⁴³ The administration also launched a DOJ task force in January 2026 to challenge state AI regulation.⁴⁴ The administration is arguing that unraveling Big Tech’s power could see the AI boom grind to a halt, allowing international competitors to gain footing while “[threatening] to stifle innovation.”⁴⁵

The Trump administration has also championed market dominance (i.e., monopoly power) as a strategic national advantage—benefitting some of the president’s largest donors and most powerful supporters.⁴⁶ Big Tech CEOs have lavished Trump with money and praise. With around 70 percent of global AI supercomputing power currently located in the United States, and US-based AI firms being uniquely committed to the scale-above-all approach to AI deployment, they have a lot riding on the president’s favor.⁴⁷

The argument that Big Tech companies are too vital for US national security and economic interests to be allowed to fail, or “too strategic to fail,” is not new. In 2019, the first Trump administration's DOJ and Department of Energy both stepped in to argue that a monopoly ruling against Qualcomm would “put [the United States’] security at risk.”⁴⁸ Google brought up a similar argument in its antitrust case, suggesting that it would be too risky to break it up amidst the major technological transformation (toward AI) that they are helping to drive.⁴⁹

Indeed, controlling bigger firms with clear monopolistic bottlenecks in the AI supply chain offers the state leverage in its geopolitical machinations. For instance, the Biden administration used this power to weaponize exclusion from the most advanced chips against China to try to slow their AI development via strict export controls. The Trump administration has relaxed many of these policies, opting instead to weaponize access to chips to foster Chinese dependence on American AI platforms while expanding markets for US firms.⁵⁰ Of course, the opposite is also true: bigger and fewer firms with oligopolistic control over AI means owners and financiers of these firms have leverage to bend and transform the objectives of the state to meet their economic interests.⁵¹

The mythmaking of AI for national security represents the latest iteration in the tech industry’s deep, longstanding relationship with the Pentagon.⁵² Big Tech firms were already becoming significant defense contractors before the AI boom. For example, the Pentagon inked a joint $9 billion contract with Amazon, Google, Microsoft, and Oracle for cloud computing in 2022.⁵³ However, some of the largest AI companies were at first resistant to US military use of their AI tools, in part due to internal worker organizing. As scholar Nick Srnicek documented, “At the start of 2024, Anthropic, Google, Meta, and OpenAI were united against military use of their AI tools. But over the next 12 months, something changed.”⁵⁴ All four companies reversed their previous stances and joined unabashedly jingoistic firms like Palantir and Anduril in the AI weapons contracting business.⁵⁵

Of course, this “surrender” comes with lucrative long-term contracts. In other words, surveillance and military applications might materialize some of the requisite demand for Big Tech’s data center spending hopes. As Srnicek noted, “The soft budget constraints and long-term nature of defense contracting, combined with the often blurry metrics of success, make the military a highly desirable customer for new technologies.”⁵⁶ While AI-related contracts have so far only represented modest sums relative to the total (and anticipated) revenue of the large AI firms, their potential value far surpasses the immediate returns. The contracts represent a deepening of state lock-in for service providers, with the goal of making their infrastructure so inextricably interwoven with government systems so as to prove very difficult to replace—and potentially positioning the companies as too important to let fail.⁵⁷

As AI firms pursue greater involvement in procurement, they also seem to be struggling to maintain already weak ethical standards. In a recent and seemingly contradictory example, Anthropic has put its $200 million federal contract in jeopardy by refusing Secretary of War Pete Hegseth’s demand that the company jettison contractual terms prohibiting the government from using Claude for fully autonomous weaponry or mass surveillance of US citizens.⁵⁸ The Trump administration has responded by designating Anthropic a “supply chain risk,” which would prevent the company from doing business with the government and potentially even other federal contractors.⁵⁹ However, this clash obscures the fact that surveillance abroad and killing people with weapons that are not fully autonomous do not violate Anthropic’s principles, that Anthropic’s Claude model has already been used extensively by the US military in strikes in Iran, and that CEO Dario Amodei has framed AI as an existential battle with China—with all of the cold war and militarist implications that entails.⁶⁰ As of this writing, the courts have sided with Anthropic, saying the supply chain risk designation was likely an infringement on freedom of speech.⁶¹ Anthropic’s gambit could lead to a better position with future Democratic administrations despite its hawkish and militarist orientation. In the meantime, Claude vaulted from relative obscurity to become the most-downloaded app in the United States in the immediate aftermath of Hegseth’s escalation.⁶² 

Defense contracting is particularly enticing because of the nearly $1 trillion US military budget—which the Trump administration hopes to increase by a preposterous 50 percent.⁶³ As long as this level of federal spending persists, so will the military-industrial complex and the attendant allocation of resources and technological development toward violent ends rather than socially useful ones. Now that we have established the core linkage between AI data centers and distinctly antisocial purposes of death and destruction, we will turn our attention toward how the state systematically pushes to insulate data center developers and their financiers from any popular challenges on environmental protection.

Antisocial infrastructure: denying social and ecological priorities to facilitate highly polluting development

While ramping up militarism and going to war around the world, the Trump administration has systematically tried to erase socially useful government spending and programs—particularly climate and environmental justice—while denying negative externalities and real resource constraints. They have eliminated already thin spaces of public oversight, leaving unaffordability and devastating pollution for working people to endure. The Trump administration calls this an energy dominance agenda. 

The administration’s obsession with fossil fuels runs alongside attacks on renewable energy, with policies that incentivize investments in extremely expensive and polluting energy with cash flows derived from the AI boom. Life support to the coal-fired power plants only postpones important reinvestments and takes valuable time and space away from more deliberative processes around a just energy transition and affected communities; in other words, better alternatives are foreclosed. The Trump administration’s executive orders and declarations against offshore wind and solar on federal lands (which are often later deemed illegal) are the opposite of derisking, while also stymieing efforts to build the relevant supply chains and opportunities for workers to actively manage this transition.⁶⁴

Loss of federal tax credits and bonuses are leading to project delay and cancellations, which could be devastating for communities who stood to gain from affordable energy, cleaner air, and potentially greater public control of critical, socially useful infrastructure.⁶⁵ Instead of encouraging clean local energy, federal policy is pushing data centers into many of these communities instead.⁶⁶

Data centers’ impacts on community wellbeing can be wide-ranging, from eliminating natural spaces to increasing air, water, and noise pollution. A February 2026 study found that the increase in air pollution from a single data center in Loudoun County, Virginia, which operated gas turbines on site, caused between $53 million and $99 million in annual health damages to the nearby community.⁶⁷ One study places the total public health costs of US data centers by 2030 as comparable to those of California on-road car emissions, and emphasizes those costs would also disproportionately fall on marginalized populations, with one study estimating “the per-household health burden [nationwide] could be 200x more than that in less-impacted communities.”⁶⁸ Across the country, this ballooning footprint is replicated. A new Wired analysis found that natural gas-powered data centers in the United States could increase greenhouse gas pollution each year by 129 million metric tons, more than entire countries, and a Nature study found AI servers could come with an annual water footprint of 731 to 1,125 million cubic meters.⁶⁹ 

At the state and federal level, data center development increasingly benefits from regulatory exemptions and streamlined review processes, exemptions justified not only in the name of winning the AI “race,” but also US energy and geopolitical dominance. In recent years, a mix of Republicans, centrist Democrats, fossil fuel firms, and clean energy advocates and developers have engaged in a push for broad “permitting reform.” This effort began in earnest in 2022 with fossil-fuel-friendly senator Joe Manchin’s attempt to use deregulation of the permitting process as a legislative bargaining chip, culminating with the permitting provisions included in the Fiscal Responsibility Act of 2023. This campaign has reframed environmental oversight and public participation as obstacles to building infrastructure of all sorts.⁷⁰ Many of these policy ideas have outlived Biden’s term and were reflected in Trump’s One Big Beautiful Bill Act (OBBBA) and executive orders related to data center infrastructure and permitting reform. These new policies hollow out environmental protections and deregulate polluting energy infrastructure and other industrial projects.⁷¹ Deregulation derisks data center construction, signaling to markets that governance will prioritize Big Tech’s transgressions over working people’s health and cost of living.⁷²

Both the Biden and Trump administrations moved to accelerate the National Environmental Policy Act (NEPA) review timelines and offer incentives for data centers on federal lands, coinciding with the erosion of the United States’s bedrock environmental laws.⁷³ In the Fiscal Responsibility Act, Congress imposed arbitrary deadlines of environmental assessments and environmental impact statements to under a year and under two years respectively.⁷⁴ Doing so limits nuance in evaluating projects and cuts back on time to redesign projects in response to local conditions, without providing the necessary administrative capacity to accelerate review. The OBBBA further cut down the deadline for an environmental impact statement to under a year for developers willing to pay.⁷⁵ In July 2025, Trump issued an executive order that required the Council on Environmental Quality (CEQ) and federal agencies to find ways to expedite data center approval by modifying regulations promulgated under NEPA; the Clean Air Act; the Clean Water Act; the Comprehensive Environmental Response, Compensation, and Liability Act (CERCLA); and the Toxic Substances Control Act.⁷⁶

The Trump administration has followed through and paired down regulations that are already under-enforced by an underfunded and understaffed EPA whose political leadership is bent on easing regulations for industry, including by deliberately neglecting enforcement responsibilities.⁷⁷

Regulatory and administrative subsidies

Pushing past fossil fuel retirement and causing volatile and high energy prices

Alongside its attempts to quash renewable energy, the Trump administration has repeatedly invoked the need to power AI data centers as a reason to boost fossil fuels. Pursuant to an executive order in April 2025 to expand energy supply for AI data centers, the Trump administration has cited grid stability provisions in the Federal Power Act to mandate that six coal, oil, and gas plants that were set to retire continue to operate, even sometimes over the objection of operators who would endure losses.⁸⁴ If Trump expands his mandate across all 54 carbon-polluting plants slated for retirement, it could incur an additional $3 billion in ratepayer costs and cause tens of thousands of excess deaths, alongside enormous additional health damages from particulate and ozone pollution per year that they remain open.⁸⁵

Energy analysts at Cleanview calculated that each GW of power enables $10 to $12 billion in revenue for data centers, which means fossil fuel plant retirement pushback alone can enable hundreds of billions in data center revenue even if these fossil fuel plants do not run all the time.⁸⁶ This, in turn, enables future raising of capital to build additional data centers, and thus expands the potential bubble further. This is before accounting for other direct subsidies or federal land leasing.⁸⁷ 

The Trump administration has chosen selectively from the remains of the IRA to push for new nuclear supply chain investments and to restart old reactors. For example, they directed a $1 billion loan to Three Mile Island in Pennsylvania to open up 835 MW of power for Microsoft’s data centers, which could enable $8.35 to 10.02 billion of revenue. The Trump administration has also slashed nuclear safety regulations, including raising radiation exposure limits and environmental contamination rules, and proposed to use generative AI in nuclear licensing and commissioning, boosting several nuclear startup’s funding rounds in the process.⁸⁸

Energy abundance for AI, unaffordability for everyone else

All these energy investments do not always neatly translate into better or cheaper services for everyday electricity users that a publicly coordinated and planned one could have. While some are celebrating the possibility that supporting data center energy needs will result in long overdue grid improvements, it is important to consider how “improvements” are not all created equal and have the potential to actually lock in energy use patterns that warp the grid around Silicon Valley’s priorities as opposed to those of US residents.

The US electricity grid is plagued by a fractured paradigm of governance that makes coordination and connection of new resources difficult. It also suffers from outdated infrastructure and a lack of clear and consistent investments, which translates to inequitable access.⁸⁹ Insufficient and oligopolistic industrial capacity makes it difficult to build out necessary supply chains for components needed for clean and resilient energy. Liberalized power markets without deliberate coordination of supply and demand help create price spikes. Undemocratic governance of utilities means public input cannot determine grid access terms and resource planning questions. All of these issues, already exposed by the budding addition of renewable power during the IRA’s implementation, are now bringing the grid to the brink.

Starting in 2026, data center power demand began to outpace utilities and grid operators’ ability to grow supply at the same rate.⁹⁰ At the same time, the federal government eagerly commandeered grid governance to prioritize connecting data centers to the grid, while paying lip service to affordability and leaving grid reliability as afterthought.⁹¹ To do so, they weaponized the Federal Energy Regulatory Commission’s (FERC) authority and provisions from the Federal Power Act. The federal government is tying grid buildout to the growth of unpopular data centers and fossil capital. Data centers may “consume a majority of all power delivered by the utility,” which can mean utilities prioritize cheaper rates for data centers.⁹² A strained, marketized grid with these unplanned infrastructure additions means more price spikes, higher overall prices, dirtier air and water, locally unwanted data center and energy infrastructure, and less reliable energy.⁹³

Enabling private equity to rob ratepayers and hijack utility planning 

As energy infrastructures are seen as crucial to AI growth, asset managers and private equity have been buying up utilities from Minnesota to New Mexico as a way to profit, and possibly to ensure that data center deals and companies can get access to electricity as soon as possible in a bid to increase control over the AI supply chain where electricity is the major bottleneck.⁹⁴ As the Center for Biological Diversity uncovers, private equity firm Blackstone’s takeover of the New Mexico utility TXNM, for instance, is filled with “cross-subsidization” issues. Blackstone owns electricity component suppliers like MacLean Power Systems and Power Grid Components, co-owns and finances upstream energy producers that could provide reliable energy supply and AI data center deals downstream, who could access favorable rates from TXNM.⁹⁵ These deals often involve minimal rate protection sweeteners and clean energy investments, while ceding control over future investments, rate decisions, and allocation of energy sources to private equity.⁹⁶ Private equity’s troubling history of looting and exploitation in other industries is well-documented, and the implications for electricity users are troubling.⁹⁷

Nonetheless, the federal government has under-exercised its ability to intervene in a devolved grid governance paradigm. The approval of Blackrock’s Global Infrastructure Partner’s (GIP) takeover of ALLETE, a holding company that owns the Minnesota utility, Minnesota Power, for example, needed to undergo FERC approval, as well as DOJ and FTC anti-merger review, because the deal was larger than $100 million.⁹⁸ At the same time, there is a major Google hyperscaler data center proposal right next to a Minnesota Power substation.⁹⁹ Even if Blackrock and Blackstone are not specifically involved in these deals, they are major financiers and, in some cases, shareholders of data center developers where they benefit from data centers getting electricity access to commence operations (Blackrock for instance owns a little less than 7 percent of Google).¹⁰⁰ Yet all three institutions approved this takeover swiftly, routing power from public to private hands via deliberate regulatory and antitrust nonenforcement. One way a more robust regulatory state could have intervened would have been for each of these named authorities to scrutinize the potential conflicts of interests far more closely and to reject these attempts at consolidating energy entities for AI data centers.

Infrastructure and utility governance subsidies

Paying for the party: federal financial support for the AI boom

We have gone over the ways that AI companies first frame their industry as too strategic to fail, and how the Trump administration is removing barriers to speed up deployment. Now we will discuss how the US government is also providing direct financial support in a variety of ways to create a secure AI supply chain—from chips manufacturing to critical minerals.

The Trump administration is attempting to assert US control in supply chains for AI, especially in rare earth elements and other critical minerals.¹⁰⁴ There have been several examples where the Trump administration used a combination of equity stakes, warrants, debt service, guaranteed loans, and price floors to boost investments up and down the supply chain to support an AI-centered, even if incoherent, vision.¹⁰⁵ This should be read as an expansion of Biden administration policy rather than departure from it. For instance, the CHIPS Act of 2022 has resulted in the dispersal of more than $40 billion in grants and loans to 52 projects to establish a domestic semiconductor supply chain, with significant amounts of that capacity enabling the mass manufacturing of chips to fill data centers.¹⁰⁶

Grants and loans to develop domestic manufacturing capacity are important industrial policy tools. However, in these cases, the government did not direct the outputs in any specific socially productive directions, creating the space for private manufacturers to reroute these capacities toward what is most profitable. Right now, that is AI chips. Taiwan Semiconductor Manufacturing Company’s (TSMC) three new chip fabrication facilities in Arizona that now produce Nvidia’s newest AI chips were funded by $6.6 billion in such grants. Also, the $5.7 billion in these grants to Intel were converted by the Trump administration into an equity stake to boost AI production.¹⁰⁷

It is important to distinguish the major departure from the Biden administration in the policy tools used and their purposes. The Trump administration is issuing direct investments in firms and projects, while the Biden administration preferred more concessionary approaches. While the Biden administration’s goals for onshoring and supply chain resilience could be argued to have social benefits, the Trump administration’s intent seems to act as policy signals to protect extractive, highly polluting industries from legal scrutiny rather than pushing them for better practices. Additionally, the applications of these minerals will ostensibly be pushed toward a socially and ecologically destructive industry.

Direct expenditures

Tax “breaks” are actually just corporate handouts 

Since the late 20th century, the US approach to industrial policy has relied heavily on the tax code. Tax expenditures are seen to reduce the burdens on private investment, as opposed to direct “tax and spend” mechanisms of government investment. Tax code expenditures are viewed as less a political liability than direct government spending in the context of a fiscally conservative neoliberal consensus, even though they have similar fiscal effects.

Tax policies reflect political preferences and power imbalances. Scholars like Melinda Cooper have long argued that right-wing economic thought simultaneously pushes for balanced budgets while obscuring real “supply side extravagance” enacted through tax cuts and support for financial asset appreciation.¹⁰⁹ That economic thought has become mainstream policy, and the data center boom is just the latest iteration.

One reason why Big Tech has so much capital at its disposal is because the federal government has cut the top tax rate on corporate income by roughly 60 percent since the middle of the 20th century.¹¹⁰ Lower corporate tax rates function as a form of government spending. However, this type of expenditure allows corporations to retain more cash reserves and increase their collateral, which Big Tech has been using to finance AI development and data centers at a worldmaking scale exceeding that of the interstate highway or New Deal buildout as a percentage of GDP.¹¹¹

The next section offers examples of Big Tech’s tax avoidance practices and shows how they leveraged infrastructure-agnostic tax policies like accelerated depreciation for data centers and associated infrastructure, which can be seen as public spending that gives Big Tech the financial flexibility to build more, faster, and larger data centers.

Intensified corporate tax avoidance

Trump-era moves to further slash taxes on corporations and the wealthy have delivered a massive windfall to economic elites—Silicon Valley oligarchs chief among them. At the end of 2017, Republicans passed the Tax Cuts and Jobs Act (TCJA) and Donald Trump signed the bill into law. The TCJA lowered the top marginal corporate tax rate from 35 percent to a flat rate of 21 percent.¹¹² For context, the top marginal rate was 52 percent in the 1950s.¹¹³ A clear consequence of lowering the corporate tax rate is that revenue collected as a percentage of GDP fell from almost 6 percent in 1952 to less than 2 percent in 2024.¹¹⁴ Making matters worse, the OBBBA, a Trump-backed Republican budget reconciliation package enacted in July 2025, extended certain tax breaks for big businesses and established new ones.¹¹⁵

According to data compiled by the Institute on Taxation and Economic Policy from annual financial statements submitted to the US Securities and Exchange Commission, just three Big Tech companies driving the AI boom—Alphabet (Google), Amazon, and Meta (Facebook)—paid $49.7 billion less in federal corporate income taxes in 2025 than they would have if they paid the full statutory rate of 21 percent. They paid $93 billion less than the total amount compelled by the pre-TCJA rate of 35 percent.¹¹⁶ While these firms have many options at their disposal for reducing the amount they pay, TCJA lowered the ceiling of social responsibility.

Taken together, these three corporations paid an effective tax rate of just 4.9 percent in 2025—one quarter of the current statutory rate and one seventh of the pre-2018 rate. It is worth noting that if Alphabet, Amazon, and Meta paid the 52 percent rate that predominated in the middle of the 20th century, the federal government would have collected more than $145 billion in additional tax revenue. That is enough money to provide public housing to 16.6 million people for a year.¹¹⁸

Tax policy as a feedback loop for oligopolistic political power

It is not only the federal corporate tax rate that has decreased over time. Ultra-high-net-worth individuals, including the CEOs of the companies discussed above, are also paying lower total effective tax rates.¹¹⁹ A team of University of California, Berkeley economists estimated that “the total effective tax rate—all taxes paid relative to economic income—of the top 0.0002 percent (approximately the ‘top 400’) averaged 24 percent in 2018–2020 compared with 30 percent for the full population and 45 percent for top labor income earners.”¹²⁰ According to the study, “The top-400 effective tax rate fell from 30 percent in 2010–2017 to 24 percent in 2018–2020.”¹²¹ This is a long way from the 50 percent effective rate that the top 0.1 percent paid in 1945.¹²² Billionaires have used money gained from tax avoidance to bankroll reactionary or pro-status quo candidates, lobby for upwardly redistributive policies, buy media outlets, or otherwise undermine democracy. 

Examples abound of how corporations and billionaires are weaponizing their wealth gains to beget even more wealth and power. This was turbocharged by the 2010 Citizens United Supreme Court decision that allowed influence group spending to skyrocket, like the nearly $400 million spent by Silicon Valley on the 2024 presidential elections—including more than $240 million by Elon Musk in support of Trump.¹²³ In the 2026 midterm elections alone, Meta is projected to spend $65 million in Illinois and Texas.¹²⁴ Likewise, a new $100 million Leading the Future PAC funded by Open AI, Palantir, and AI venture capital firms is flooding cynical ads (often not even mentioning AI) against candidates who support AI regulations.¹²⁵ And beyond elections, the top nine AI-related tech companies spent more than $95 million on lobbying in 2025 alone.¹²⁶

Accelerated depreciation to accelerate capital—speed, scale, and speculation

These newfound political advantages have been turned into financial ones through the tax code. Industries in the AI technology stack face the same problems that other capital-intensive industries face, namely that investors eschew long periods of return on capital. Accelerated depreciation schedules in the US tax code allow companies to write off taxes in the early years of operations and, therefore, provide a structural subsidy for capital-intensive industries. While this could be a useful industrial policy tool for encouraging green and socially beneficial investments, with AI it allows firms to rapidly write off the costs of servers, cooling systems, power equipment, and specialized buildings precisely when capital expenditures are highest. Rather than incentivizing productive efficiency or long-term innovation, these provisions reward speed, scale, and speculative buildout. In other words, they are tax shelters that, when combined with the overall environment of asset price appreciation, result in mainly private capital gains in exchange for foregone tax revenue that could be spent on social goods and infrastructures. 

In practice, the tax code derisks investments by tech conglomerates with the scale and sophisticated financial capacity to monetize depreciation immediately without any reciprocal commitment or requirement that these investments work toward social benefit. The behavior of hyperscalers buying up GPUs as collateral that rapidly depreciates, however, means that the more that tax policies enable this behavior, the bigger the bubble becomes.¹²⁷ Thus, these tax expenditures accelerate the accumulation of risks to the entire financial system, which has severe potential consequences downstream for working people. 

Rather than steering investment toward the cleanest energy systems, depreciation policy rewards returns on whatever can be built fastest at scale with the greatest profit. Equipment that powers data centers including natural gas-fired generation, diesel and gas backup generators, batteries, and energy-intensive cooling systems all qualify for the same accelerated treatment. Given cheap gas in the United States, land and sizing requirements for renewables, operational experience with gas, the tax code effectively subsidizes carbon-intensive design choices. Indeed, 30 percent of planned data centers in the United States intend to build on-site power generation, with 75 percent of capacity proposed to be natural gas turbines or engines.¹²⁸ Even if demand for these data centers drops, gas plants run for decades and leave behind both a financial liability and a polluting legacy for years to come.

Tax code expenditures

People and planet first: Green Economic Populism

The ginned-up urgency of winning an AI race that most people do not even want to be running in the first place needs to be named for what it is: an industry-led hype cycle that aims to supplant democratic control with investor prerogatives. So long as industrial development is controlled by corporations singularly concerned with asset appreciation and profit-maximization, it is hard to imagine a viable path toward decarbonization and biospheric rationality. The all-out push for AI could not come at a worse time. Coupled with the destruction of the IRA, an intensifying climate crisis, and a dwindling carbon budget, a mostly fossil fuel-powered and resource-intensive data center boom is accelerating the breaching of planetary boundaries.¹³¹ After enduring more than three decades of industry-friendly bromides about the potential alignment of market forces with environmental goals, now AI’s off-the-chart energy needs are leading to one more round of industrial leaders walking back their voluntary pledges and vague commitments to “save the planet” or to achieve “net zero” while the planet continues to burn. 

In the face of an accelerating climate crisis, we reject the “AI race” as a productive and socially beneficial use of resources, particularly (but not exclusively) on the terms set out by the Trump administration. This “race” toward market, workplace and geopolitical dominance is only one of many possibilities for AI as a single component in our complex social and industrial ecology. Instead, we advocate for a “race” toward a liveable world (ecologically and economically) where abundance is not measured in asset prices alone, but in a more capacious, multi-dimensional sense of what working people actually need and want. 

Capital and state capacity are being mobilized behind an AI First agenda of rapid data center development because our collective resources are overwhelmingly controlled by a small group of fantastically wealthy people who prioritize maintaining and expanding their power over all other considerations. The current AI agenda is one in which the state collaborates with dominant technology firms to further erode our autonomy, democratic standing, personal privacy, and biosphere. The impact of this strategy on the working class is economic precarity, the destruction of knowledge and institutions, and mass surveillance. 

At the same time, across the United States, there is robust, well-organized, and successful local organizing against AI data centers. In 2025 alone, local opposition contributed to the blocking or stalling of 48 projects worth $156 billion, often on the grounds of opposing additional pollution and lavish public subsidies to feed corporate profits.¹³² Big Tech’s AI offensive is an unpopular, ruling class-driven project that people are increasingly and justifiably worried about. Rather than merely accommodating and making the best of Big Tech’s buildout, these democratic assertions of power and resistance open space to collectively imagine alternative visions.¹³³

Existing policy discourse, however, has failed to meet the moment. In addition to treating the data center buildout and AI deployment as a foregone conclusion, many policy advocates are focusing on different flavors of resigned policy visioning spanning from profit sharing, ameliorating the impacts of AI on the labor workforce, scavenging for scraps from the buildout to achieve siloed issue benefits like grid expansion. Many are even directly pushing back against data center resistance and calls for moratoria, which we see as one of the few levers of democratic intervention able to briefly pause AI data center developments. 

The alignment between public consciousness and the widely documented harms of unfettered AI and data center deployment means the moment is ripe for a more transformative vision that ensures the federal government’s obligation to democratically control, govern, and plan the scale of AI to instill confidence in the ability of the public sector to meet public desire. 

In other words, it is necessary to rebuild the state capacity to make choices for working people as opposed to working for AI and elite investors.

Financialization over the last several decades means private investments can dictate direct economic activity in ways that only public investments were once able to. Corporate power is at an all-time high, while the state has either seen its regulatory capacity rolled back, or in the cases of military applications and energy, is actively intervening against working people’s interests. This state-capital collusion means windfalls for Big Tech, finance capital, and oligarchic shareholders, while working people’s jobs and lives are degraded and expectations are warped around a newly impoverished normalcy in which people forget that things were not always this bad.

There is an alternative. As the data center boom shows, a state-capital partnership can rapidly roll out massive, world-shaping infrastructure when interests and motivations align. The government can pick winners and direct markets, innovation, and production according to its interests. We want to see a government that directs our collective resources and energy toward an agenda that serves people and the planet. This will require governing power and authority wielded for the public interest, with a focus on jobs with dignity, environmental stewardship and green transition, and robust social welfare programs. Accomplishing this is not possible without a means of democratizing investment decisions and developing a decision-making calculus that is insulated from corporate influence and can balance multiple, sometimes conflicting, objectives. 

Given the urgent ecological and affordability crises, there is a political opening for Green Economic Populism: to push for a more equitable and rational vision that is carefully planned and intentionally democratic, focused on meeting people’s immediate needs, directing and regulating the private sector, and building requisite public capacity and institutions. CCI’s “Stop Greed, Build Green” framework and agenda can realize such a vision, using a Green Economic Populism framework to align policies that bring immediate relief to working people with building a dynamic public sector that can enact the robust regulations and enormous green investments needed to realize a better future for all.¹³⁴ To that end, we offer several points of alignment upon which we invite collaboration:

1. Contest Big Tech’s desired expansion and scale with democratic planning

The speed and urgency with which this buildout is occurring is without any historic precedent. Democratic planning and deliberation take time, and the only way that time will happen is by pulling the breaks. The industry’s manufactured urgency has nothing to do with meeting a present need, or even any reasonable fear of losing some metaphorical race, but solely to do with the massive financing deals that have poured hundreds of billions of dollars down the AI and data center drain. In fact, it is more that they have an urgent need to make their investors whole. This is not social, political, or cultural urgency—it is financial urgency. This is a race toward trillion-dollar valuations, not some mythical “artificial general intelligence.”

While data center site fights are succeeding, their victories are largely limited to constraining where—not whether—data centers get built. A federal moratorium on new data centers would be a welcome start to allow for more deliberate governance to be put into place, and for public investments in critical social infrastructure to be the main driver of economic activity. 

Part of that governance should include ending support like subsidies, tax incentives, and federal approvals that are driving data center development and the fossil fuel expansion that powers it until adequate standards and safeguards are in place. The federal government should be setting the floor for any data center developments, and empowering local and state authority that might emerge from local site fights against data centers.¹³⁵ It is also necessary to develop a means of slowing and right-sizing the overbuild and overdeployment of AI and AI data centers, including decommissioning excessive private data centers and developing publicly owned computing infrastructure. That means creating a new economic and resource planning authority capable of democratically deciding what a sufficient and sustainable number of data centers might be—as determined by social and ecological welfare, not profit motives. 

2. Build and invest in a grid for social priorities

The data center buildout is hampered by a severely maintained energy infrastructure; this should trigger a rethink around who we build the grid for. The geography and operating requirements of AI data centers are forcing poor choices for new energy infrastructure without evidence of substantive service or affordability improvements for everyday electricity users. This data center-centric grid build out is crowding out grid equipment and taking labor away from more socially useful electrification needs for homes, transportation, and heavy industries. As previous CCI research has shown, a federal public power authority with a strong, democratic structure could ensure that the grid is built for working people, not Big Tech oligarchs.¹³⁶ 

Public power and municipalization movements, with allied policymakers, can also bring about or reinforce public power authorities that reject power access agreements for data centers, which, if done at scale, could limit speculation and its huge wake of wasted resources. They can also use public finance institutions to pay for socially useful electrification and decarbonization and merge the goals of more widespread economic development with popular control. This can be supported by alternative green industrial policy including co-locating green industries and providing end-to-end supply chain support for them.

3. Expand public investments that put people and the planet first

The root cause of why so many unions and local governments across the United States find data center deals irresistible despite their (failed) promises of significant tax revenue and jobs. is systematic under- and mal-investment.¹³⁷ The antidote is public investment in economically robust and environmentally sustainable regional economies and redistribution. Rather than having cyclical industry booms drive new investments, a National Investment Authority can help deliver more equitably distributed and durable economic investments, such as those that meet local needs for social housing, broadband, accessible public transportation, water and energy infrastructure.¹³⁸ That will also require bolstering local governance and civic capacity to conduct spatial planning of infrastructure and develop the alternative industries that communities desire. 

4. Center the need for well-paid jobs with dignity across the labor market

The AI First agenda offers an opportunity for a climate-labor coalition as the agenda is so clearly both a job- and planet-destroying enterprise. Resistance to AI and data centers should always be coupled with affirmation of working peoples’ dignity and their irreplaceable skills and creativity. 

A federal jobs guarantee that provides much-needed green public sector work—such as education, healthcare, child and elder care, green infrastructure construction, legacy infrastructure repair and ecosystem restoration—would backstop workers from disruptive effects of AI. At its best, this renewed and emboldened state capacity should provide a floor—both in compensation and quality of work—that raises the level of dignity and security throughout the labor force. 

As bastions of organized working class power, labor unions are, in theory, ideally positioned to push back against AI deployment (or at least for a more sensible form of it). However, they are at a historically weak moment, and some unions that are benefitting from the short-term job boom of the data center buildout are even supporting it.¹³⁹ The only way to unite labor against the AI First agenda is to offer a material alternative: a people- and planet-first agenda that offers working people a meaningful and secure alternative ready to hire (and train) skilled workers today, not in some hypothetical future. 

5. Fight Big Tech’s dystopian visions and increase public leverage over AI firms and technologies

As the tech industry seeks to erode and replace human knowledge and capabilities, a renewed commitment to the value of human effort is necessary and buoying. AI—and the broader incursions by the Trump administration against research, education, and freedom of the press—highlight the desperate need to fund, repair, and rapidly expand public educational systems and public research alongside robust public funding for arts, culture, and independent journalism. It is important to be clear-eyed about the costs that AI’s ostensible benefits are coming with, particularly when provided through revenue- and attention-maximizing corporate strategies. Research is being replaced with sycophantic validation, knowledge muddied by hallucinated “facts,” and the art and practice of critical thinking is being eroded by easy shortcuts.

The largest AI firms have eschewed their ethical guardrails as investor demands require more singular focus on rapidly demonstrating robust revenue streams that can justify the billions being spent on data center infrastructure. As exhibited by Anthropic’s recent partial release of its Mythos model, AI firms are racing to release powerful models with socially disruptive potential and the complete absence of meaningful oversight. This makes it clear that a layer of regulatory security needs to be established that can slow these releases down; subject them to rigorous third-party testing; and establish ethical, social, and environmental benchmarks for any approved model to adhere to. Just as the FDA is meant to protect consumers by demanding new food and drugs meet rigorous standards, an equivalent infrastructure needs to be developed for AI. 

It may even be the case that some version of the technologies currently referred to as AI will play a role in positively shaping this future, perhaps under control of different people, researched and advanced with different goals, and deployed in different forms and with different guardrails built into their systems. 

The United States needs creative planning practices that experiment with new public sector capacities as well as new incentive structures to mobilize private capital behind public agendas, and to rein in the power of corporate and financial interests. In addition to pushing back on the physical infrastructure that enables AI deployment, fighting Big Tech’s attempt to insert AI in every part of our society will require institutions powerful enough to constrain the financial and political power of monopolistic firms and their owning class. This includes dramatically increased taxes on the rich and strict regulations on political spending, alongside reimagining and reinvigorating antitrust law to more strongly consider market architecture and a diversity of objectives beyond geopolitical supremacy and consumer access. 

Strong rules governing AI deployment and shared digital commons, combining stringent industry regulation around what, how, and when AI can be used are also needed. Experiments in publicly owned software and related technologies could follow. The AI Now Institute has recommendations around whole-lifecycle regulation of AI, bright-line bans, independent oversight of AI development, anti-monopoly moves against corporate concentration in the tech sector, and public conditions for data center development.¹⁴⁰ This must include wholesale prohibitions on the most societally harmful uses of AI technologies. “No” has to be a viable answer.