Regulating AI Data Centers

The Hidden Cost of Artificial Intelligence: An Energy Crisis in the Making

Artificial intelligence promises transformative breakthroughs across medicine, science, and industry, but this technological revolution carries an enormous hidden cost: an unprecedented surge in electricity consumption that threatens to overwhelm our power grid, drive up residential utility bills, and undermine climate goals. Data centers consumed over four percent of total United States electricity in 2023, a figure projected to surge to nine percent by 2030.¹ This explosive growth—driven primarily by AI training and inference—represents what MIT Energy Initiative Director William H. Green calls “a gigantic new demand that no one anticipated,” dwarfing traditional annual growth rates of approximately half a percent.¹

The stakes extend far beyond abstract energy statistics. Across the United States, investor-owned utilities have received requests from proposed data centers for massive new electricity capacity. In Massachusetts alone, utilities received requests for an additional two gigawatts—equivalent to eighty-five percent of all Commonwealth households.² We face a fundamental question: will working families subsidize the energy costs of corporate AI infrastructure while watching their own electricity bills climb?

Congress must act decisively to regulate data center energy consumption, require emissions accountability, protect residential ratepayers, and ensure the AI revolution serves the public interest rather than merely enriching technology corporations at everyone else’s expense. Without federal intervention, unchecked data center growth will saddle communities with infrastructure costs, accelerate climate change, strain water resources, and drive energy poverty while concentrating AI’s benefits among wealthy corporations and their shareholders.

A Surge in Electricity Demand

The scale of AI-driven energy consumption defies comprehension. Global electricity demand from data centers is projected to more than double by 2030, reaching approximately 945 terawatt-hours—surpassing Japan’s entire national energy consumption.³ In the United States, data center annual energy use in 2023 reached approximately 176 terawatt-hours, about 4.4 percent of United States annual electricity consumption, with projections showing that data center energy consumption could double or triple by 2028, potentially accounting for up to twelve percent of United States electricity.⁴

A single large data center consumes electricity equivalent to powering approximately fifty thousand homes.¹ The United States currently operates over five thousand data center facilities, with more than 160 new AI-focused data centers constructed across the country in just the past three years.⁵ The Federal Energy Regulatory Commission reports an increased demand projection of between sixty and 130 gigawatts more by 2030 than they forecast just two years prior.⁶ House Energy and Commerce Committee witnesses in March 2025 forecast that data centers are expected to account for the single largest growth segment in electricity demand, adding up to forty-four percent of United States electricity load growth through 2028.⁶

Training a single large AI model can require a total power draw of 25.3 megawatts, and the power required to train these models could double annually.⁶ Each one hundred-word AI prompt consumes roughly 519 milliliters of water—equivalent to one bottle—for cooling purposes alone.⁵ When millions of users generate billions of prompts daily, the cumulative impact becomes staggering. The National Academies convened a workshop in November 2024 gathering more than 95 in-person and 350 virtual participants from academia, technology industry, electric utilities, community advocacy groups, and government agencies to explore how to map, measure, and mitigate the impacts of AI data center electricity usage.⁷

These numbers translate into tangible infrastructure crises. Approximately five-year waits now characterize interconnection queues for grid access, as utilities struggle to accommodate new data center demands.¹ Transmission infrastructure presents additional bottlenecks: sufficient power generation capacity means little if transmission wires lack capacity to transport electricity to needed locations.¹ Renewable energy sources prove insufficient for both hyperscalers and residential consumers, forcing continued reliance on coal-fired plants and delaying facility closures, directly contradicting climate commitments.¹

Residential Ratepayers Subsidizing Corporate Infrastructure

The explosion of data center construction creates a perverse dynamic: working families bear infrastructure costs while corporations capture profits. In Maryland and Virginia, ratepayers are funding a five billion dollar upgrade to transmission lines to accommodate data centers.² Across the country, current regulatory frameworks often permit utilities to pass infrastructure costs directly to residential customers through rate increases, forcing working families to subsidize corporate AI infrastructure.

Data centers provide minimal local employment—typically automated operations requiring few workers—while straining local electricity rates and infrastructure.¹ Communities bear costs through higher utility bills, grid reliability concerns, and environmental degradation, receiving little compensating benefit. This represents wealth extraction: technology corporations externalize their infrastructure costs onto residential ratepayers while privatizing AI’s economic returns. Goldman Sachs Research estimates that approximately sixty percent of increased data center electricity demand will be met through fossil fuels, contributing roughly 220 million tons of additional global carbon emissions—equivalent to the carbon impact of driving a gas-powered vehicle five thousand miles per ton of carbon dioxide.³ Residential customers thus subsidize not merely electricity infrastructure but the climate damage that infrastructure causes.

Lawmakers across multiple states fear AI data centers will drive up residents’ power bills, with particular concerns in states pursuing clean energy targets that data center demand could derail climate goals.⁶ The fundamental injustice requires correction: if technology corporations profit from AI, those corporations—not working families already struggling with housing costs, healthcare expenses, and wage stagnation—should finance the infrastructure AI requires.

Climate Consequences and Environmental Degradation

The climate implications of unchecked AI data center growth threaten to undermine decades of decarbonization progress. The International Energy Agency warns that climate pollution from power plants running data centers could more than double by 2035.⁶ This acceleration arrives precisely as climate science demands rapid emissions reductions to avoid catastrophic warming. Goldman Sachs Research projects that sixty percent of increased data center electricity demand will rely on fossil fuels, generating approximately 220 million tons of additional global carbon emissions.³ For context, driving a gas-powered vehicle five thousand miles generates approximately one ton of carbon dioxide, meaning data center emissions equivalent to more than one billion such trips.³

The renewable energy challenge compounds climate concerns. Data centers require reliable baseload power operating continuously, a demand poorly matched to intermittent renewable sources like solar and wind without massive energy storage infrastructure that remains prohibitively expensive and insufficiently deployed. Consequently, data center growth often relies on continued fossil fuel generation or delays retirement of existing coal and natural gas plants, directly contradicting state and federal clean energy targets. Researchers explore scheduling flexibility to shift AI workloads to periods when renewable sources predominate, but current economic incentives favor constant maximum utilization rather than grid-responsive operation.³

Water Scarcity and Resource Competition

Beyond electricity, AI data centers impose enormous water demands for cooling, creating competition with residential users, agriculture, and ecosystems in increasingly water-stressed regions. Large data centers can consume up to five million gallons per day, equivalent to a town of ten thousand to fifty thousand people.⁵ AI could require as much as 720 billion gallons of water annually, enough to meet indoor needs of 18.5 million households.⁵

More than 160 new AI data centers built across the United States in the past three years emerged in places with high competition for scarce water resources.⁵ In Arizona, a data center’s monthly water usage during summer can nearly double its average level, spiking precisely when residents and businesses need water most.⁵ This competition for scarce water during heat waves and droughts creates direct conflicts between corporate AI infrastructure and community survival needs.

Congressional Response: The Clean Cloud Act

Federal legislation addressing data center energy consumption and emissions remains nascent but represents crucial first steps toward accountability. Senator Sheldon Whitehouse of Rhode Island and Senator John Fetterman of Pennsylvania introduced the Clean Cloud Act of 2025, Senate Bill 1475, establishing an emissions performance standard for electricity used by cryptomining facilities and data centers while directing revenues toward consumer utility relief and clean energy investment.⁴⁸

The Clean Cloud Act grants the Environmental Protection Agency and Energy Information Administration authority to collect data on annual energy consumption of data centers and cryptocurrency mining facilities, electricity providers, power purchase agreements, and related information.⁴ Transparency represents a necessary precondition for effective regulation: without comprehensive data on which facilities consume what quantities of energy from which sources, regulators cannot design targeted interventions. The technology industry has resisted disclosure requirements, making federal data collection authority essential.

The bill establishes a carbon fee starting at twenty dollars per ton of carbon dioxide equivalent and rising at least ten dollars annually, based on how facilities’ emissions profiles compare to regional power grid averages.⁴ This escalating fee structure creates growing economic incentives for data centers to pursue clean energy sources, efficiency improvements, and emissions reductions. Crucially, resulting revenue funds two purposes: easing residential electricity rates and financing clean firm generation including battery storage and nuclear power.⁴ This revenue structure ensures that emissions penalties benefit the residential ratepayers who otherwise bear data center costs while simultaneously accelerating clean energy infrastructure deployment.

The Clean Cloud Act faces formidable political obstacles, with observers noting it carries near-zero chance of passage in the current Congress.⁴ However, the legislation establishes important policy markers that will inform future debates as data center impacts become increasingly visible through rising electricity bills and grid reliability challenges. Lawmakers addressing this issue in coming years will reference the Clean Cloud Act’s framework as a starting point for regulatory design.

House Resolution 5227 proposes conducting a comprehensive study on the impact of artificial intelligence and data center site growth on energy supply resources in the United States.⁴ While merely a study rather than binding regulation, H.R. 5227 acknowledges congressional recognition that AI data center growth demands federal attention and systematic assessment. The House Energy and Commerce Committee held hearings in March 2025 examining data center energy consumption, with witnesses forecasting that data centers will account for the single largest growth segment in electricity demand.⁶ This growing congressional attention signals potential for future regulatory action as constituent pressure intensifies.

Policy Solutions: What Congress Must Do

Congress must act to regulate AI data center energy consumption through comprehensive federal legislation addressing cost allocation, emissions accountability, efficiency standards, and transparency requirements. The Clean Cloud Act provides a strong foundation, but Congress should go further to protect working families and ensure the AI revolution serves the public interest.

Emissions Performance Standards and Carbon Pricing: Congress should pass the Clean Cloud Act or similar legislation establishing escalating carbon fees for data centers based on their emissions profiles. These standards must account for both direct operational emissions and indirect emissions from electricity generation, preventing facilities from claiming clean energy credentials while consuming fossil fuel power. Revenue from emissions penalties must fund residential utility relief and clean energy infrastructure, ensuring that those harmed by data center externalities receive compensation.

Cost Allocation Protection: Federal legislation should prevent utilities from socializing data center infrastructure expenses across residential ratepayers. When data centers require grid upgrades, transmission infrastructure, or generation capacity, those costs should be borne by facility operators through direct charges and capacity fees rather than embedded in general utility rates passed on to families. The Federal Energy Regulatory Commission should establish national cost allocation frameworks protecting residential customers from subsidizing corporate AI infrastructure.

Energy Efficiency Mandates: Congress should require data centers to deploy best available technologies for reducing consumption. Research demonstrates that reducing GPU energy consumption to approximately thirty percent of typical levels “has minimal impacts on the performance of AI models” while improving cooling efficiency.³ However, economic incentives favor maximizing utilization over maximizing efficiency absent federal requirements. Efficiency standards would compel technology corporations to prioritize environmental performance alongside computational performance.

Water Conservation Requirements: Federal legislation should address cooling demands, particularly in water-stressed regions. Requirements for closed-loop cooling systems, prohibitions on potable water use for cooling, and mandates for water-efficient technologies would reduce competition between data centers and residential users. Arizona’s experience with data center water consumption spiking during heat waves when residents face shortages illustrates the national urgency of federal water regulation.⁵

Transparency and Disclosure: Congress must mandate comprehensive reporting of energy consumption, emissions, water usage, and environmental impacts, following the Clean Cloud Act’s data collection framework.⁴ Technology corporations resist disclosure, claiming proprietary concerns, but public interest in energy and environmental impacts outweighs corporate preferences for opacity. The EPA and Energy Information Administration need federal authority to collect this data and make it publicly available.

Grid Planning and Residential Priority: Federal energy policy must prioritize residential reliability over commercial data center expansion. When transmission capacity proves insufficient for both data centers and residential demand, residential users must receive priority. Similarly, renewable energy capacity should preferentially serve residential decarbonization and essential public services rather than enabling unconstrained AI expansion.

---

References