Hannah Wiseman (Sept. 4, 2026)

Staff at the Center for Energy Law and Policy are researching policy tools to address the impacts of data centers. We are summarizing our findings in a series of blog posts, also to be published as pdf documents. Our intent is to share these tools as options that governments might consider when addressing data center construction and operation. This blog post introduces general data center concepts and the categories of impacts that policymakers might consider addressing. It also provides a broad overview of the types of governments that have authority over different data center impacts.

What is a data center?

Data centers are warehouses full of computer servers. The main components of data centers include a building or buildings to house the servers, racks of servers, cooling systems for the servers, “power distribution units” that control the flow of power to cooling equipment and servers, and back-up generation in the event that electricity voltage changes or electricity from the grid is suddenly unavailable. Back-up generation typically consists of on-site diesel generators.  Some data centers generate their own electricity. They rely on the electric grid only in a minor way or are disconnected from the grid entirely. These data centers have on-site electricity generation technology, such as natural gas (“gas”) combustion turbines.

How many data centers are there, and where are they? 

The Pew Research Center, relying on Data Center Map, estimates that there are more than 3,000 operational data centers in the United States and that, as of April 2026, “more than 1,500 new data centers are in various stages of development nationwide.” Estimates of existing and planned data center numbers vary substantially, however, because the definition of “data center” can range from several servers in a space the size of a closet to a larger hyperscale data facility.

Also according to Pew, 87% of existing data centers are in urban areas, whereas “67% of planned data centers are in rural areas.” Most new planned data centers are slated for locations in the U.S. South and Midwest, according to Pew.

Data Center Map, which estimates 4,767 U.S. data centers as of September 4, 2026, showed the majority of data centers as located in Virginia (674 data centers), Texas (537), California (296), Georgia (276), Illinois (242), and Ohio (240).

How much land do data centers occupy? 

Data centers occupy varied amounts of land depending on their function. “Hyperscale” data centers are “data centers built by companies that deploy internet services and platforms at massive scale,” and experts at Lawrence Berkeley National Laboratory estimate that hyperscale data centers have an average building footprint of 30,000 square feet. IBM reports that the average data center building is 20,000 to 100,000 square feet in size. Data center campuses are larger than the building that houses the servers because they also house cooling towers or other cooling equipment, exterior lighting, back-up electricity generation and sometimes onsite generation, on-site water storage (depending on the water source), and parking for employees. Fortune reports that one of the largest U.S. hyperscale data center campuses–Meta’s “Hyperion” campus in Louisiana–will occupy approximately 3,650 acres. This is similar to the size of a large city’s airport, including the airport building and all surrounding grounds and runways. The Homer City campus in Pennsylvania, which will house data centers and natural gas-fired turbines, is more than 3,200 acres. Contrast these acreages with data center campuses in a state such as Missouri, where the largest data center projects in progress range from 400 to 900 acres.

Why are companies building more data centers? 

Data centers have existed for a long time. They support anything digital–storing data such as photos and text files in the cloud, streaming music and videos, online sports betting, online purchases, and similar digital endeavors. But data centers are growing rapidly in number, in large part because of their use for artificial intelligence (AI). Companies use data centers to train AI models that respond to queries, and to run these models in response to queries–an activity called inference.

How much electricity and water do data centers use? 

The electricity usage of a data center can be measured in many ways. One is “nameplate capacity,” meaning the absolute maximum amount of electricity a data center facility could draw at any given time, including server (“IT”) use and non-server use for operations such as cooling and lighting. With respect to electricity use of just the computer servers, Lawrence Berkeley National Lab explains that “rated power” is the “absolute maximum power draw” of a server; the “maximum power” is the observed power draw if you run the server at its maximum workload, and the “operational power” is the power use of servers as observed, when they are operating in their “typical workload mode.”

Types of infrastructure that affect data center energy use 

Approximately half of the energy used by data centers is for running the computer servers, and the remaining half is primarily for cooling. This is changing, however, as data center companies install more efficient cooling technologies such as liquid cooling.

The type of servers and the type of cooling system at a data center both impact the amount of electricity the center uses.  Lawrence Berkeley National Lab provides a helpful breakdown of server types: “Conventional” servers that serve “legacy” (non-AI) data center workloads use less electricity than “AI specialized” servers. And even each of these main server categories has variations that impact electricity use–conventional servers have different numbers of “central processing units.” Multiple processer conventional servers are used for activities such as “scientific computing” or “big data processing.” AI specialized servers are either “AI accelerated”–meaning they have graphics processing units (GPUs) that quickly respond to queries or conduct training–or AI non-accelerated.

The type of cooling system at a data center also substantially impacts its energy use. Computer room air conditioners (CRAC) are a common method for cooling small and mid-size data centers and are energy intensive, using pumps to move chilled water and air and sometimes inefficiently sucking hot air into cooled aisles between server racks. Water-cooled units are more energy efficient but have higher water use. For these systems, water condensers absorb heat from the system and then release the heat through cooling towers; this evaporative cooling approach is water intensive.

Liquid cooling involves using water or another liquid at each server rack rather than pumping air or water through pipes in the rooms that house servers. Some liquid cooling technologies replace the door of the server rack with a heat exchanger; others immerse the actual computing equipment, such as servers, within liquids. Some cooling even occurs at the level of the computer chip, with “cold plate” or “direct-to-chip” cooling using “small heat exchange modules” on top of chips and other components inside servers; these heat exchange modules have tiny channels that “enhance heat transfer.”

All types of data center cooling have trade-offs. For example, liquid cooling that immerses servers directly within liquids is far more energy efficient than air conditioners or water-cooled units. But some forms of immersion cooling use PFAS (polyfluoroalkyl substances) fluids, which are known as “forever chemicals” that can persist in biological systems and the environment.

Types of operations that affect data center energy use 

Data centers consume different amounts of power depending on how much work (if any) the servers are completing; servers use energy even when they are wholly idle, but they use less energy. Training an AI model usually uses more electricity than inference because servers training a model typically run approximately 80 percent of the time, whereas servers doing inference are assumed to run approximately 40 percent of the time. Energy use for inference varies substantially, though. The more queries sent to an AI model, the more the servers run.

Power usage also varies depending on the weather; cooling systems have to run more when it is hotter.

Overall energy use

At any given moment, a hyperscale data center’s servers could use multiple megawatts (MW) or even gigawatts (GW–1,000 megawatts) of electricity. To put one megawatt in perspective, the Solar Energy Industries Association estimates that one megawatt of solar photovoltaic electricity powers, on average, 174 homes. For generation sources that are continuously available, one megawatt can power around 400 to 600 homes.  All of these numbers are rough because one megawatt of electricity, whether from a “baseload” runs all-of-the-time source, or an intermittent source such as solar not paired with batteries, goes farther in some parts of the country than in others. This is due to varied efficiencies of buildings, among other regional differences. So a hyperscale facility like Meta’s Hyperion, which is designed to have up to 5 gigawatts of computing power, could use at any given moment the same amount of electricity used by 870,000 homes if using the solar MW estimate, or 3 million homes if using the 1 MW per 600 homes estimate. Meta’s proposed 5 gigawatts of computing capacity represents nearly one-fifth of Louisiana’s total net generating fleet for peak (maximum) demand throughout the state (total net summer generation capacity of 24,707 MW).

What categories of data center impacts might governments consider?

Electricity rates: Electric utilities’ investments in new transmission and distribution lines and generation to serve data centers can cause all ratepayers’ (including residential ratepayers’) rates to increase if data centers are not required to directly pay for the infrastructure. Even if rates are designed to make data center companies internalize these costs, rates can increase simply due to overall rising demand for electricity, as determined by Virginia’s Joint Legislative Audit and Review Commission. The Commission estimates a possible $14 to $37 increase in the monthly average residential electricity bill for Dominion Energy customers by 2040, independent of inflation.

Noise: Near the servers at a data center, noise levels can reach approximately 90 decibels–“above the 85” decibel threshold “considered harmful to hearing.” 90-decibel noise falls between the noise levels of heavy city traffic (80 decibels) and a jack hammer or power tools (100 decibels). Noise levels off the site are lower but can still affect wildlife. These impacts arise due to low frequency noise and many other frequencies emitted from data centers. Offsite noise can also affect neighboring residents. There are few, if any, peer-reviewed studies of noise from data centers.

Light: Data centers do not require large amounts of lighting, but according to some studies they use potentially more lighting than is needed for security and similar purposes.  This can impact wildlife, the “dark sky” in rural areas, and nearby residents.

Air pollution: Back-up diesel generators at data center sites emit nitrogen dioxide, particulate matter and hydro carbons, and carbon monoxide. Some data centers have hundreds of diesel generators on site. For example, one data center in Loudon County, Virginia houses 245 generators according to the state environmental agency’s permit for the facility.  The Virginia Department of Environmental Quality estimates that Loudon County data centers in total have “approximately 4,021 Tier II diesel generators (those that are supposed to run for standby purposes and therefore have fewer pollution controls) and 130 Tier IV generators (those with stricter air pollution controls). If data centers switched to natural gas generators for backup, this would reduce levels of particulate matter but would still emit nitrogen oxides, which can contribute to haze and acid rain.

Potential ground and water pollution: Data center cooling systems and their discharges, depending on the cooling system used, can contain PFAS.

Heat island effects: Studies have shown that data centers can measurably increase local air temperature. One study showed that a data center raised downwind air temperature by an average of 0.7-0.9 degrees Celsius, with warming as high as 2.2 °C, compared to corresponding upwind areas. Heat impacts could be measured as far as 500 meters from the facility.

E-waste and decommissioning: Data center companies replace servers every three to five years, meaning that the replaced servers must be repurposed or disposed of.

Which governments regulate which impacts? 

The federal government, through agencies such as the Environmental Protection Agency and Fish and Wildlife Service, regulates impacts such as: 1) water pollution (pretreatment of liquid waste before it is sent to wastewater treatment plants and controls for stormwater runoff from sites), 2) air pollution (diesel generators and on-site generation), 3) impacts on endangered and threatened species and their habitat, and 4) interconnection of data centers to the transmission grid. Many states have received delegated federal authority to regulate air and water pollution and have additional laws that can be stricter than federal laws. States therefore also regulate air, water, and wildlife impacts of data centers, among other impacts.

Regional organizations called regional transmission organizations or independent system operators control portions of the transmission grid, plan for new transmission lines and generation needed to support data centers, run auctions for new generating capacity, and determine whether or not new generators and data centers may connect to the transmission grid. These organizations are overseen by the Federal Energy Regulatory Commission. In areas without regional transmission organizations that run the grid, individual electric utilities do transmission and capacity planning. Grid operators are actively planning for new infrastructure, including generation capacity and transmission, needed to support data centers.

States regulate electric utilities’ provision of electricity to data centers and the rates that electric utilities charge data centers and other customers. States also regulate the siting (location) of new transmission and distribution lines for data centers and eminent domain for these lines, and most regulate the siting of new electric generation. States also regulate the extent to which local governments may regulate land development and use to ensure the health, safety, and welfare of the public through zoning and other land use regulations. Very generally speaking, local governments in zoning “home rule” states have more authority to regulate land use for health, safety, and welfare than do local governments in “Dillon’s Rule” states, but even local governments in many Dillon’s Rule states implement relatively detailed land use ordinances.

Local governments control whether and where a data center company may build a data center. Depending on the extent of their powers delegated from the state, through zoning and other land use-related regulations, they may regulate any aspect of a data center that impacts human health, safety, and welfare, including aesthetics and impacts on the environment. One exception is West Virginia, which has preempted (blocked) local government control over the construction and location of certified “high impact” data centers.

For Pennsylvania local government ordinances addressing data centers, see, for example, Archbald Borough, Susquehanna County, and Tioga County. For a broader review of local government data center ordinances around the United States, see the Georgia Data Center Ordinance Hub.