AI’s growth is typically framed as a story about compute and algorithms, but its physical build-out depends on land, water, energy, and labor supplied by real communities. Historically, the digital economy has been marketed as a virtual cloud, disconnected from physical constraints. However, as the demands of generative artificial intelligence scale exponentially, this illusion is fracturing. The physical footprint of AI is expanding at a rate that local municipalities and regional grids were never designed to accommodate, triggering a profound collision between hyperscale technology companies and the communities that house their infrastructure.
That dependency is no longer going unquestioned. Across the United States, a pattern of community, state, and regulatory pushback is rapidly emerging. Local governments are realizing that the economic promises of data centers often fail to offset the severe localized extraction of resources. Virginia has introduced a new data center electricity tax, Monterey Park, California became the first U.S. city to ban data centers outright, and New York lawmakers have proposed a statewide moratorium on permits for certain large-scale data centers. These escalating responses signal a critical shift: the limits of treating AI infrastructure as exempt from local resource and community constraints have been reached.
This article examines what these regulatory actions and grassroots interventions reveal about the future of AI expansion. It argues that AI is not just a technology story; it is a story about place, power, and people. Governments, companies, and investors who fail to account for this physical dependency are carrying an unpriced systemic risk. By reframing AI growth as a fundamentally extractive infrastructure challenge, we can better understand the federal-state tensions and corporate blind spots that will define the next decade of digital expansion.
The physical requirements of large-scale data center build-outs are staggering, yet they have historically been treated as frictionless externalities by the technology sector. To understand the scale of the impending bottleneck, one must quantify the land, water, and energy dependencies that underpin artificial intelligence. Energy consumption is the most immediate constraint. According to International Energy Agency (IEA) projections, global data center electricity consumption could more than double from 460 TWh in 2022 to over 1,000 TWh by 2026.1 This surge is driven heavily by the operational phase of large language models; inference alone can account for up to 90% of a model’s total lifecycle energy use.2
Water consumption presents an equally critical vulnerability. AI models require massive volumes of freshwater for evaporative cooling systems to prevent server overheating. Estimates indicate that training a model like GPT-3 consumed roughly 700,000 liters of freshwater.3 When scaled to the deployment of newer, vastly more complex architectures, this water footprint threatens to exacerbate regional droughts and strain municipal water supplies. Labor and land requirements further complicate the build-out. While hyperscale projects are often touted as job creators, the labor benefits are highly skewed. In Virginia, the data center industry contributes an estimated 74,000 jobs and $5.5 billion in labor income, but these economic benefits are overwhelmingly driven by the initial construction phase rather than long-term operational employment.4 Indeed, state legislative evaluations reveal that while a typical facility can bring up to 1,500 workers on-site at the height of construction, it operates with approximately 50 full-time employees once live.4 Once built, these capital-intensive facilities require vast tracts of land while delivering minimal ongoing employment, disrupting local socio-economic dynamics and displacing alternative community developments.
Crucially, this resource extraction is not an isolated domestic anomaly but part of a globally escalating recognition of physical limits. International regulatory bodies are already moving to price these externalities. For example, the European Union’s revised Energy Efficiency Directive (Directive (EU) 2023/1791) mandates stringent, standardized reporting on data center energy and water consumption, signaling an end to the era of opaque resource usage.5 Furthermore, the upstream impacts of this infrastructure extend deep into the Global South. Comprehensive cradle-to-grave lifecycle assessments of AI hardware, such as the Nvidia A100 GPU, reveal that the manufacturing stage accounts for 85% of abiotic mineral and metal depletion, heavily relying on resource extraction in developing nations.6 A 2025 analysis by Hines Research estimates that another 40,000 acres of “powered land”—nearly 2 billion square feet, equivalent to roughly three times the land area of Manhattan—will be needed globally over the next five years to keep up with projected data center growth.7 This global footprint underscores that AI’s dependency on physical resources is a systemic vulnerability, tying advanced algorithms directly to the earth’s finite materials.
As the sheer scale of AI’s resource footprint becomes apparent, communities and local governments are transitioning from passive hosts to active regulators. The historical deference granted to technology companies is evaporating, replaced by escalating public and regulatory pushback. This friction is most visible in the United States, where a series of unprecedented legislative and municipal actions have fundamentally altered the landscape for data center expansion. These interventions demonstrate that local tolerance for unpriced externalities has reached a breaking point.
In Virginia, long considered the epicenter of global data center capacity, the political calculus is shifting. Faced with grid strain and the realization that nearly one-third of the state’s data centers have been built within 200 feet of residential areas due to inadequate local planning, lawmakers have taken decisive action.4 The introduction of a new data center electricity tax represents a paradigm shift, effectively forcing hyperscalers to internalize the costs of their massive grid impact. This financial penalty is compounded by fierce grassroots opposition. In Prince William County, community groups successfully launched lawsuits to overturn decisions that would have rezoned 1,700 acres of homes and historic farmland for 37 new data centers.8
The pushback has escalated from taxation to outright prohibition in other jurisdictions. In California, Monterey Park became the first U.S. city to permanently ban the construction of new data centers. Driven by concerns over water usage, noise pollution, and the strain on local utilities, residents overwhelmingly rejected the premise that digital infrastructure should supersede community well-being. This localized rejection has also been echoed at the state level in New York, where lawmakers have proposed a statewide moratorium on permits for certain large-scale data centers. The proposed legislation would pause permitting for covered data centers while allowing time for environmental and utility impact reviews.
These examples are not isolated incidents. They represent a broader awakening among municipalities and state policymakers that the physical presence of data centers carries severe, tangible costs. The era of rubber-stamping zoning approvals for server farms is ending, replaced by a demanding public that requires transparency, environmental mitigation, and equitable economic benefits before surrendering their land and resources.
The localized pushback against AI infrastructure has ignited a complex jurisdictional battle, exposing a deep federal-state tension over who ultimately governs the physical expansion of the digital economy. As states and municipalities pursue moratoriums and restrictive zoning measures, these actions increasingly intersect with national priorities surrounding technological competitiveness, grid reliability, and interstate commerce. This live conflict serves as a proxy for the broader contest over AI infrastructure governance in the United States, highlighting the friction between national strategic interests and local physical constraints.
At the heart of this tension is the impact of hyperscale data centers on regional electricity markets. The massive, concentrated loads required by AI facilities are destabilizing established grid planning. In the PJM Interconnection, the largest wholesale electricity market in the U.S., covering much of the Mid-Atlantic, the influx of data center demand has driven substantial capacity market price inflation. Wholesale capacity prices in the PJM market skyrocketed more than elevenfold in just two years, jumping from $\(28.92/MW-day for the 2024–2025 delivery year to\) $329.17/MW-day for 2026–2027.9 These pressures create a complex governance challenge, as federal priorities for grid reliability and market efficiency increasingly intersect with state and local authority over permitting, land use, and infrastructure development.
This dynamic echoes historical infrastructure battles, where the assertion of local citizen voice significantly altered national development trajectories. Empirical data from the construction of the U.S. Interstate highway system in the 1970s demonstrates that local community opposition was a primary driver of skyrocketing infrastructure costs, causing spending per mile to triple as projects were delayed, rerouted, or heavily mitigated to appease local resistance.10 These historical dynamics offer a useful precedent for understanding how community opposition could similarly affect the cost and timing of AI data center development.
This unresolved federal-state dynamic creates a complex regulatory environment for AI infrastructure development. National priorities surrounding technological competitiveness, grid reliability, and expanding computing capacity increasingly intersect with state and local authority over land use, permitting, resource management, and infrastructure development. As data center demand continues to grow, companies and investors must therefore account not only for technical and financial feasibility but also for the possibility that local resource constraints and regulatory requirements may delay, reshape, or limit proposed projects.
The escalating conflict over land, water, and energy fundamentally alters the risk profile of AI infrastructure investments. Currently, a dangerous asymmetry exists in corporate sustainability reporting: companies meticulously measure their impact on communities and resources, but they rarely measure their operational dependency on them. This failure to account for physical resource constraints creates a volatile blind spot for risk management, leaving investors exposed to sudden regulatory shifts, stranded assets, and spiralling operational costs.
To circumvent these immediate physical constraints, the technology sector has increasingly pivoted toward contrarian technological mitigations, heavily promoting solutions like advanced closed-loop cooling and Small Modular Reactors (SMRs) as panaceas for resource scarcity. The premise is that advanced engineering will decouple AI from its environmental footprint. However, rigorous techno-economic evaluations suggest these solutions are highly speculative and financially perilous. Simulations of fully nuclear-powered data center configurations reveal that such setups can cost between $\(1 billion and\) $10 billion, introducing massive capital expenditure risks that far exceed traditional grid-tied facilities.11
The uncertainty surrounding these longer-term technological solutions was illustrated by the late-2023 cancellation of NuScale Power’s small modular reactor project with Utah Associated Municipal Power Systems. The project was terminated after the parties determined that it was unlikely to secure sufficient subscription to continue toward deployment.12 The collapse of the NuScale project illustrates the core premise of unpriced risk: relying on speculative, decade-away technological breakthroughs rather than acknowledging and adapting to immediate physical limits creates an unstable foundation for growth.
Investors must recognize that the cloud is tethered to the ground. When a hyperscale facility is denied a water permit, or when a local grid operator caps capacity allocations, the underlying algorithms cannot function. Risk management frameworks must evolve to treat community tolerance and resource availability as hard operational limits, not just public relations hurdles. Companies that continue to project infinite compute scaling without securing the corresponding physical resources are fundamentally mispricing their future liabilities. The failure to integrate these dependencies into core strategic planning will inevitably lead to severe financial and operational disruptions.
The physical build-out of artificial intelligence is rapidly colliding with the finite realities of the natural world and the sentiment and political will of local communities. As demonstrated by the skyrocketing projections of energy and water consumption, the infrastructure required to sustain the AI boom is unprecedented in its intensity. The era in which data centers could expand invisibly, treating land, water, energy, and labor as cheap, inexhaustible externalities, is rapidly coming to an end. This transition marks a critical inflection point for the global digital economy, forcing a reevaluation of how compute power is generated, distributed, and sustained.
The escalating regulatory pushback, from Virginia’s targeted electricity tax to Monterey Park’s outright ban and New York’s proposed statewide moratorium, signals a significant shift in how digital infrastructure is governed. These developments, coupled with international frameworks like the European Union’s Energy Efficiency Directive, represent a global awakening to the unpriced costs of compute power. With capacity market prices increasing more than elevenfold in the PJM market and projected global energy demands exceeding 1,000 TWh by 2026, the tension between national priorities for technological competitiveness and state-level protections of local resources guarantees a highly volatile landscape for future development. The historical precedent of infrastructure cost inflation driven by community opposition suggests that the financial models underpinning current AI expansion are dangerously optimistic.
For corporate leaders, policymakers, and investors, the mandate is clear: community and resource dependency must be elevated to a core strategic consideration in AI infrastructure planning, not treated as an afterthought or a public relations exercise. Relying on speculative techno-fixes like Small Modular Reactors to bypass immediate physical constraints can deepen exposure to systemic risk and divert attention from necessary efficiency improvements. Ultimately, the future of artificial intelligence will not be dictated solely by the sophistication of its algorithms or the speed of its processors. Instead, a critical constraint on AI’s growth will be the availability of the physical resources required to run these systems and the willingness of local communities to supply them without sacrificing their own socio-economic and environmental stability.
