AI Is Being Built as the Inevitable Future, but Who Pays for Its Environmental Costs?

Artificial intelligence can feel almost weightless. A question typed into a chatbot; an image generated in seconds; a paragraph produced by a large language model appears on a screen with little indication of what is happening behind it.

Who bears the environmental costs of AI locally? Who has a meaningful say in decisions that determine how a community’s electricity, water, and land are used?

But the infrastructure behind AI is anything but weightless. Across the United States, companies are racing to build enormous data centers packed with increasingly powerful computing equipment. Substations, transmission lines, cooling systems, new roads, and often new electricity generators soon follow. Depending on their cooling technology and location, enormous amounts of water are needed as well.

The International Energy Agency (IEA) projects that electricity consumption from data centers worldwide will more than double to around 945 terawatt-hours by 2030. AI is expected to be a major driver of that increase. In the United States, the IEA estimates that data centers could account for nearly half of electricity-demand growth through 2030. Lawrence Berkeley National Laboratory’s 2025 assessment estimates that US data centers could consume about 11.8 percent of the country’s electricity by 2030, with a range of 9.5 to 15.3 percent depending on how rapidly the industry expands. These are projections, not certainties, but they illustrate the scale of infrastructure that may be required to support the AI economy.

The environmental debate around AI has largely focused on energy consumption and carbon emissions. But there are other crucial questions that receive less attention: Who bears the environmental costs of AI locally? Who has a meaningful say in decisions that determine how a community’s electricity, water, and land are used?

The Infrastructure Behind the AI Boom

A data center does not exist in isolation. It has to get electricity through the grid. A large amount of heat is produced by high-density computing equipment, which needs to be cooled. Depending on the cooling system, this takes water. Large facilities may also need new substations, transmission infrastructure, roads, and backup power.

Virginia shows both sides of this expansion. The state has become one of the world’s largest data center markets, particularly in Northern Virginia. Virginia’s Joint Legislative Audit and Review Commission (JLARC) estimated that the industry contributes about 74,000 jobs, $5.5 billion in labor income, and $9.1 billion in annual GDP to the state. Much of that economic activity is associated with construction. JLARC found that a typical 250,000-square-foot data center may employ around 50 full-time workers, compared with roughly 1,500 workers during peak construction.

Those benefits matter. Data centers can bring investment, construction work, tax revenue, and infrastructure improvements. It would be misleading to portray every development simply as an environmental loss.

Yet the economic benefits do not answer every question or relieve every concern. JLARC found that roughly one-third of Virginia’s data centers are located near residential areas, and warned that inadequate local planning and zoning can result in adverse impacts on residents. The agency also found that water use by data centers is currently sustainable statewide, while noting that water availability varies among localities and that there is less oversight over how available water is allocated among competing uses.

That distinction is important. A resource can be available at the state level while becoming contested at the local level.

Who Pays for the Electricity?

The same question applies to the electricity system. When utilities anticipate enormous new loads from data centers, they may need new generation, transmission lines, substations, and other infrastructure. The challenge is determining who should pay for those investments.

The issue has become significant enough to attract federal attention. On June 18, 2026, the Federal Energy Regulatory Commission (FERC) ordered all six regional grid operators under its jurisdiction to justify or reform their tariffs for data centers and other very large electricity users. FERC said the proceedings were intended to speed the integration of large loads while protecting consumers.

The concern is not that every data center will automatically shift costs onto households. Properly designed tariffs can require large customers to pay the costs they create. But if infrastructure built primarily to serve exceptionally large customers is recovered too broadly, ordinary ratepayers could potentially bear part of the expense.

That matters for environmental justice because electricity costs are not experienced equally. A modest increase may be manageable for a high-income household but onerous for a family already struggling with energy costs. The question of who pays for AI is relevant long before anyone types a prompt into a chatbot.

When AI’s Power Demand Reaches the Neighborhood

The Memphis area provides a more direct example of what this infrastructure expansion can look like on the ground. Elon Musk’s xAI has built its Colossus data center in Memphis and expanded the project across the state line into Southaven, MS. For the Colossus 2 expansion, xAI has used natural-gas turbines to generate electricity for the facility.

The case is particularly important because it shows what can happen when the timetable for building AI infrastructure moves faster than conventional grid development. The demand for computing power can translate into demand for dedicated electricity generation, bringing questions about air pollution, permitting, and public health directly into nearby neighborhoods.

In April 2026, the NAACP sued xAI and its subsidiary MZX Tech, alleging that the companies were operating 27 natural-gas turbines in Southaven without the required air permit. Earthjustice and the Southern Environmental Law Center are representing the NAACP. The allegations remain part of ongoing litigation and have not been adjudicated.

The NAACP argues that the turbines effectively operate as a power plant for Colossus 2 and are located near homes, schools, and churches. The organization says the turbines could emit more than 1,700 tons of nitrogen oxides annually, along with particulate matter, carbon monoxide, and formaldehyde. Those are potential emissions estimates cited in the lawsuit, not measurements of actual annual emissions.

The controversy has also prompted residents to raise concerns about air quality and quality of life. Hundreds of people attended a Mississippi environmental hearing concerning the facility, according to the NAACP.

The significance of Southaven is not that every future data center will reproduce the same model. It is that the physical demands of AI can extend well beyond a server building. Electricity requirements can mean power plants, fuel infrastructure, transmission lines, and other facilities that change the environmental profile of a project.

What This Means for People and the Environment

The effects of this expansion will vary greatly by location. A data center using reclaimed water and clean electricity in a water-abundant area is not environmentally equivalent to one relying on freshwater and fossil-fuel generation in a water-stressed community.

But several potential consequences deserve attention.

1. People

For people, the most immediate concerns can involve air quality, noise, water availability, land use, and electricity costs. In Virginia, JLARC specifically identified residential proximity and noise as issues requiring stronger local planning and recommended that local governments have clearer authority to consider water use and sound impacts when approving data centers.

The Southaven dispute adds another potential human-health concern: air pollution from dedicated fossil-fuel generation. Nitrogen oxides and fine particulate matter are established air pollutants associated with respiratory and cardiovascular harms. The question in individual communities is therefore not simply whether a facility emits pollution, but how much, where, under what regulatory limits and who lives nearby.

2. Water

For water, the consequences can extend beyond the volume withdrawn by a data center. Water-intensive cooling can place additional demand on municipal systems, rivers, or groundwater. Electricity generation itself can also have a water footprint.

A 2025 study in Nature Sustainability estimated that AI servers deployed in the United States between 2024 and 2030 could have an annual water footprint of roughly 731 million to 1.125 billion cubic meters, depending on the scale and location of deployment. The researchers emphasized that the estimates carry substantial uncertainty because water and carbon impacts depend on server locations, electricity sources, and efficiency improvements.

3. Aquatic Ecosystems

For rivers, fish, and other aquatic life, the concern is not hypothetical in every location, but neither is it established that data centers are currently causing a nationwide biodiversity collapse. In a 2026 freshwater biodiversity study, researchers warned that greater water extraction, changes in hydropower operations, and increased nighttime electricity generation could affect aquatic organisms, but they emphasized that further research is needed.

Water withdrawals can reduce streamflows. Changes in water temperature and flow can affect aquatic habitats. Increased demand for hydropower can also alter how dams operate. These effects will depend on where data centers are built and how their electricity and cooling needs are met.

4. Birds

For birds, the clearest connection is through the infrastructure required to supply power rather than through the data-center buildings themselves. New transmission and distribution lines can create collision and electrocution risks. A widely cited US study estimated that power lines kill between 12 million and 64 million birds annually, including 8 million to 57 million through collisions and up to 11.6 million through electrocution.

That does not mean a particular new data center transmission line will kill a particular number of birds. The risk depends on location, line design, habitat, and bird movements. But as electricity infrastructure expands, wildlife impacts need to be considered alongside human infrastructure needs.

5. Wildlife

There is a similar concern for habitat and other wildlife. Large industrial developments can convert or fragment land, while new roads, substations, and transmission corridors can divide habitats and alter movement patterns. Freshwater biodiversity research from this year also points to a broader issue: The environmental footprint of data centers can be distributed across multiple parts of an energy and water system rather than concentrated at the facility itself.

The important point is that these are potential pathways of impact, not guaranteed outcomes. Better siting, water recycling, less water-intensive cooling, cleaner electricity, bird-safe transmission design, and stronger environmental review can reduce many of these risks.

Communities Need More than a Promise of Economic Growth

That is why community engagement is important.

Before approval of a major data center, residents should be able to see projected electricity and water demand, planned generation sources, expected infrastructure costs, tax incentives, permanent employment projections, noise assessments, and environmental studies.

But transparency only works if communities can make sense of the information with the resources they have. Community benefit agreements could provide residents a way to negotiate legally binding commitments on infrastructure, water systems, grid resilience, environmental monitoring, workforce development, or other local priorities.

Policymakers should also examine confidentiality and nondisclosure agreements involved in the negotiations of major public infrastructure. Although there may be legitimate needs to protect commercially sensitive information, information necessary to assess the use of public resources, environmental impacts, and ratepayer exposure should remain available to the public.

Philanthropy can help close another gap: know-how. When utilities propose new infrastructure or rates, grassroots groups often don’t have the funds to hire energy economists, utility attorneys, and technical consultants. Supporting intervenors and ratepayer-defense organizations at state public utility commissions can give communities a real opportunity to review the fairness of large-load cost allocation. That’s not necessarily anti-AI. It is an effort to ensure that technological progress does not outpace public accountability.

This is critical because the future of AI is not just being built on data center sites. It implicates neighborhoods, watersheds, farms, forests, rivers, and wildlife habitats connected to the infrastructure that keeps servers running.

The AI economy may be global, but its environmental costs are local. So before we ask only how efficiently AI can operate, perhaps we should ask two more fundamental questions: Efficient for whom? And sustainable for whom?