Inside Hyperscale Data Centers: Energy, Water and Health Effects

Originally published by A Voice For Choice Advocacy on October 22, 2026.

EDITOR’S SUMMARY: Every chatbot answer, streamed movie and online bank transfer runs through a windowless building most people will never see. AI has made hyperscale data centers hungrier than ever for electricity and water, and the demand is growing faster than the evidence on what it means for people living nearby. Researchers have projected significant health costs from the air pollution tied to powering them, though they also say long-term data on host communities is scarce. Meanwhile, some counties are collecting tax windfalls, tech giants are pledging to cover their own power costs, and other places are restricting or banning new facilities.

Throughout the 21st century, technology has injected profound changes into everyday life. The internet and an ever-expanding universe of devices, from tablets to smartwatches, unlocked convenient access to digital work and leisure activities. Building on this foundation, artificial intelligence (AI) has emerged as the latest transformative development. In its simplest form, AI is a branch of computer science that attempts to make machines mimic human thinking.

This new technology has been touted for its ability to assist you with a plethora of tasks. It can help you craft emails, set up a budget or make a movie with computer-generated actors. It can tackle complex problems, perhaps too difficult for you to solve on your own. For example, AI accelerates the research and development of new drugs and medical treatments. It’s able to analyze situations and make decisions on the fly, allowing the software to perform feats such as driving a car. Governments and organizations around the world are rushing to adopt these almost magical abilities.

AI’s astounding achievements are not produced by magic, of course. They do require unprecedented levels of computational power. It takes thousands of computers connected and operating in unison to enable artificial intelligence. The number of machines is so great that they must be housed in enormous facilities called data centers. The largest, known as hyperscale data centers, are the focus of this story. These locations act like giant factories churning out AI systems, and like the traditional, coal-fueled plants of the Industrial Revolution, they are drawing scrutiny for their effects on the environment and human health.

What Is a Data Center?

Not all data centers are alike. Many are smaller, traditional facilities that serve a single company, university or local market. The hyperscale facilities behind today’s building boom are campuses run by the largest cloud and technology companies. There is no single official definition of the term. The research firm IDC, for example, counts a facility as hyperscale if it holds at least 5,000 servers and 10,000 square feet of space, and analysts’ power thresholds range from about 20 to 100 megawatts of electricity. The International Energy Agency says a typical AI-focused data center consumes as much electricity as 100,000 households. Not every hyperscale facility is built for AI, but AI is a major driver of their growth.

To grasp the environmental impact of data centers, it’s important to first understand how they operate. On the outside, a data center looks a lot like a warehouse. Imagine a large, windowless, rectangular building with steel framing. That’s where the similarities end. On the inside, data centers require complex environments so the equipment they house can function smoothly. Data centers contain specialized computers called servers. These machines are optimized for a specific purpose. While a laptop is operated by one person at a time, a server is built to be accessed by many people at once via the internet. That’s why servers are used to host websites, store electronic files and photos, and stream videos. Millions of people access the digital content on servers every day.

As society’s dependence on technology has grown, so has the demand for servers to be available at all times. Consequently, these machines run around the clock. The need for constant access means data centers must have backup servers that can take over if the primary ones go offline for maintenance or break down. All of this requires electricity, and as a failsafe against power outages, the facilities maintain backup generators. Data centers contain other equipment as well. Servers rely on networking components that connect them to each other. That connection is essential for AI systems, because networked servers can deliver more computational power than one server can produce on its own. Networking also includes internet connectivity, which makes servers accessible to the outside world.

How Data Centers Affect the Environment

Data centers have existed for decades, housing giant mainframes used by corporations, governments and research institutions before the personal computer era of the 1980s. Their use took off with the arrival of the internet. Even so, AI has drastically reshaped how data centers operate because the technology calls for different equipment that takes the environmental impact to new heights. At the center are specialized computer chips called graphics processing units (GPUs), which act as the brain of an AI system. These GPUs use up to five times as much electricity per square foot as the central processing units of non-AI data centers.

The greater energy consumption and constantly running equipment give off tremendous amounts of heat. To prevent heat damage to equipment, these locations require heavy-duty cooling systems. An estimated 75% to 90% of data centers use water for this purpose. Water is collected from the local supply and evaporated in a cooling tower. The evaporation efficiently draws heat away from the machines. However, the process consumes a lot of water. The International Energy Agency (IEA) estimates that an average 100-megawatt data center in the U.S. uses about 2 million liters (roughly 530,000 gallons) a day, and the largest facilities can use up to 5 million gallons, enough to meet the daily water needs of a 50,000-person town.

Data centers are adopting a new technology, liquid cooling, to reduce water use. It keeps hardware at acceptable temperatures more efficiently than older approaches, and several techniques are available. A common tactic uses a closed-loop network of pipes that constantly recirculate water or a chemical coolant to absorb heat and transfer it away. Another is immersion cooling, in which server components are submerged in a coolant.

Although liquid cooling reduces water use, some of the fluids used, mainly in two-phase cooling systems, are per- and polyfluoroalkyl substances, commonly known as PFAS. These are called “forever chemicals” because they do not break down easily. Direct releases from data centers are difficult to measure and likely limited because cooling systems generally operate as closed loops, according to the Environmental and Energy Study Institute. The U.S. Environmental Protection Agency (EPA) says exposure to certain PFAS can cause a range of health effects, including an increased risk of cancer, disruption of your body’s natural hormones and damage to your immune system.

Moreover, operators apply various chemicals, such as cleaning solvents and biocides, to prevent mold, algae and bacteria from growing in the cooling systems. These substances can build up over time and are periodically flushed into local waterways, where they can harm aquatic life and, if not properly treated, contaminate drinking water. Exposure to biocides has been linked to asthma and other respiratory illnesses. At the same time, the treatment serves a health purpose, because evaporative cooling towers can harbor Legionella, which causes Legionnaires’ disease.

Industry estimates suggest AI is driving about 33% annual growth in data center computing capacity, up from about 10% a year in pre-AI forecasts. That demand is fueling a surge in new construction, and if the pace holds, the challenges described above could intensify. In 2026, approximately 12,000 data centers were active worldwide, and the U.S. hosted the most, with around 5,000. Roughly 1,360 of the world’s data centers were hyperscale facilities at the end of 2025, according to Synergy Research Group, but they account for nearly half of global data center capacity, and almost 800 more are in the pipeline. Another estimated 3,000 U.S. data centers of all sizes were planned or being built as of the end of 2025. Some will be very large, contributing to habitat loss and environmental pollution. According to a 2025 Bloomberg News analysis, about two-thirds of new data centers built or in development since 2022 are in places already facing high levels of water stress. Many of these areas offer cheap land and a dry climate that lowers the risk of equipment damage from humidity, but their groundwater and aquifers are at risk of depletion.

The next frontier may be space. Elon Musk’s SpaceX, which merged with the AI company xAI, plans to put data centers in orbit. In August 2026, Musk said an AI computing system designed with Nvidia would launch to orbit in the fourth quarter of 2027, with significant scale in 2028, and he has proposed launching as many as a million satellites. Proponents say the sun would provide abundant solar energy and that space-based data centers would avoid the environmental and health concerns communities have raised. Experts doubt the timeline, though. The Government Accountability Office says cooling is a major hurdle, the costs are uncertain, and more satellites could raise collision risks and interfere with astronomy.

Others are looking to the ocean. To keep servers cool, China and South Korea are looking to place new data centers there. The natural cold of the underwater environment and the plentiful seawater benefit the computers, but the environmental impact is unknown. The localized heat discharge could affect fragile marine ecosystems or trigger algae growth.

The Energy Appetite

Most national estimates do not separate hyperscale facilities from other data centers, so the national figures cited here cover all of them. Data center machinery runs on electricity, and a large share of it goes to cooling. The enormous effort to expand AI infrastructure in the U.S. is leading to record energy use. In 2023, as artificial intelligence captured the public consciousness following the arrival of the ChatGPT chatbot, data centers already accounted for an estimated 4.4% of U.S. electricity consumption, according to Lawrence Berkeley National Laboratory. The laboratory projects that share could reach 6.7% to 12% by 2028. Globally, the IEA expects data centers to use about 3% of the world’s electricity by 2030, up from approximately 1.5% in 2024.

The risk of insufficient electricity has led to greater use of fossil fuels to produce the necessary power, adding to global carbon emissions. To boost energy availability, the EPA in 2026 repealed rules that set pollution limits on fossil fuel power plants and seeks to eliminate all remaining greenhouse gas emissions standards for them. Because power demand is so high, data centers are increasingly being expected to provide their own electricity, rather than tapping into local sources and driving up costs for residents. However, this approach raises its own environmental and health concerns. Companies building data centers are adding natural gas power plants. This is cost-effective, but according to a 2026 Environmental Integrity Project report, at least 74 such plants are proposed across the U.S. Together, they could release as much greenhouse gas pollution each year as all of Australia.

As data center power plants contribute to renewed reliance on fossil fuels, diesel backup generators add another source of pollution. Needed to ensure a steady electricity supply during power disruptions, these generators are large, and a hyperscale data center might run many of them for hours at a time. This creates noise pollution and degrades air quality. Julie Bolthouse, director of land policy for the Piedmont Environmental Council in Warrenton, VA, works in the state with the most data centers. She describes the problem with backup generators:

“That’s a lot of air pollution locally. That’s particulate matter and NOx [nitrogen oxides], which impacts growing lungs of children, can add cases of asthma, and can exacerbate heart disease and other underlying diseases in the elderly.”

Bolthouse’s concerns are echoed in a 2025 commentary in the academic journal Eco-Environment & Health. Its authors say the nitrogen oxides and fine particulate matter emitted by fossil-fueled power plants and diesel backup generators raise the risk of respiratory disease, cardiovascular conditions and cancer in nearby communities. They cite a recent model projecting that U.S. data centers could contribute to nearly 1,300 deaths a year by 2030, a public health cost of more than $20 billion. The Caltech and UC Riverside researchers behind the model also found that the burden falls disproportionately on some low-income communities, partly because of their proximity to power plants and backup generators.

Noise is among the complaints residents raise most often. Cooling fans, chillers and generators run around the clock, and neighbors describe a constant low hum more than occasional loud bursts. According to the Environmental and Energy Study Institute, residents near data centers have reported headaches, nausea, sleep disturbances and high blood pressure. In Prince William County, VA, residents say data center noise routinely exceeds 60 decibels. Much of the sound is low-frequency, which standard decibel meters can miss, making it hard to measure and enforce under typical noise ordinances.

Some local governments are responding. In 2023, Prince William County removed an exemption that let commercial cooling systems exceed its nighttime noise limit, a change aimed at data centers, though supervisors acknowledged they did not yet know how it would be enforced. The county has since extended the measure’s sunset clause while it studies data center noise. Weld County, CO, adopted a separate 65-decibel limit at the property line, measured in a way that captures low-frequency sound. In Vineland, NJ, officials have not established a link between a data center under construction and residents’ complaints about a persistent hum.

Wildlife may be affected too, though research specific to data centers is scarce. Utah’s Division of Wildlife Resources lists construction injuries and deaths, habitat loss and fragmentation, and habitat degradation from noise among the possible impacts, and notes that data centers can produce a constant sound of about 80 decibels. Broader research supports the concern. A systematic review of studies from 1990 to 2013 found that terrestrial wildlife responses to human-made noise begin at roughly 40 decibels, though the research has focused mostly on songbirds and marine mammals. Biocides and other chemicals flushed from cooling systems can also harm aquatic life, and some residents describe finding sick animals they can’t explain, though that evidence is so far anecdotal.

Batteries pose a separate safety risk. Many data centers use lithium-ion batteries for backup power, and these are more likely to catch fire than older lead-acid batteries. Once one ignites, it is hard to put out. Uptime Institute’s chief technical officer has said a burning battery will “burn until it expends all of its energy.” A 2022 fire at a South Korean data center that involved backup batteries knocked out major digital services for days, though the cause was still being investigated at the time.

The health evidence so far is limited. The Eco-Environment & Health commentary notes that noise from diesel generators and cooling systems can reach 96 decibels inside a data center, above the 85 decibels considered harmful to hearing, and that it also affects staff, nearby communities and local wildlife. It adds that in some areas data centers draw up to 57% of their cooling water from drinking-water supplies. But it stresses that little is known about long-term health outcomes in communities that host data centers, and it calls for studies that measure air, water and noise exposure and track residents’ health.

Who Benefits, Who Decides?

Despite concerns about the environment and human health, data center construction is taking off, helped by local governments that offer tax incentives to bring these facilities into their communities. The goal is to attract multimillion-dollar investments that promote economic growth and create jobs. For instance, a data center owner may contract with a local business to pour a building’s concrete and employ workers. But once the facility is built, the construction jobs disappear and residents are left with the environmental fallout. Illinois requires at least 20 new jobs for a data center to qualify for tax incentives.

Of the 27 facilities the state has certified, 22 reported creating exactly that number, while the national average is just 10 permanent jobs. Bolthouse notes that the exact environmental consequences of data centers are not easy to quantify, in part because companies often require non-disclosure agreements (NDAs), legal documents that bar local officials from publicly discussing details of their deals. She explains:

“Very often, data centers are coming in with non-disclosure agreements… They’re hiding a lot of information about water usage, energy usage, air quality impacts, emissions — none of that information is disclosed, and so communities don’t really know what they’re getting into.”

Erin Brockovich echoed the sentiment. She is famous for winning a legal battle against a California energy company that was contaminating Hinkley’s water supply, and she has taken up the fight to demand greater data center transparency and accountability. Referring to NDAs, she bluntly asks, “If data centers are so great, why are they being built in secret?” Brockovich says local governments are making deals and signing NDAs with the companies behind the data centers without input from residents. As a consequence, residents first learn of these facilities when construction starts and are stuck with the repercussions. Many have pleaded with Brockovich for help. She states:

“They’re watching their utility bills climb, finding sick animals they can’t explain, and worrying about the long-term impacts on their health and property values. These complaints are not small. They show a national pattern.”

Property values are another point of dispute. Two recent analyses, one by George Mason University of 2023 home sales in Northern Virginia and one by a real estate consultancy of four Indiana data centers, found no significant evidence that nearby homes lost value. Some research points the opposite way. A University of Calgary finance researcher’s preliminary findings show homes near newly built facilities selling for about 7% less than before construction. A September 2026 working paper by Garrison Schlauch and Simon Greenhill found that large data centers lower nearby housing values, with the effect concentrated close to the site and showing up when a project is announced. Some real estate agents say the picture is incomplete.

A Texas real estate agent told HousingWire that homeowners near a data center are trying to sell but can’t, and a researcher told Indiana Public Media that the Northern Virginia study covered a single year in a tight market with few homes for sale. Studies based on completed sales or estimated values would not capture homes that never sell. The National Association of Realtors’ 2026 report found agents split, with 25% seeing a positive effect on nearby home values and 22% a negative one.

When residents raise broader objections and officials try to put a hold on construction, data center companies have in some cases threatened multimillion-dollar lawsuits that communities may not be able to afford to fight. That’s why some municipalities proceed with caution. For example, Mount Shasta, CA, was approached about turning a vacant water bottling facility with valuable water rights and a high-producing well into a data center. City officials didn’t see how it helped the community and turned down the offer.

Data centers aren’t only a burden on the places that host them. They are the physical backbone of much of modern life, from streaming and online banking to AI tools, and their backers point to concrete economic benefits. In Loudoun County, VA, home to the world’s densest cluster of data centers, county data show the facilities generate 38% of general fund revenue and nearly half of property tax revenue while occupying about 4% of the county’s commercial land. The county’s real property tax rate has fallen from $1.145 per $100 of assessed value in 2016 to $0.805 in 2025, and officials say the county collects about $26 in tax revenue for every $1 it spends serving data centers. Because Loudoun’s concentration is unmatched, its results may not translate everywhere. In other places, tax breaks can cut into the benefit. Some states exempt large data centers from sales taxes on construction, equipment and electricity. In North Carolina, for example, state estimates put the annual cost of that exemption at $45 million to $57 million.

Energy estimates are also contested. The IEA expects data centers to account for roughly a tenth of global electricity demand growth through 2030, a smaller share than industrial motors, air conditioning or electric vehicles, and projects that about half of the growth in data center demand will be met by renewable sources. But the agency also notes that data centers’ local impacts are far more pronounced than their global share suggests. In the U.S., it says, they are on course to account for almost half of the growth in electricity demand through 2030. Who pays for the grid built to serve that demand is a related concern, because if a data center’s expected power use never materializes, other ratepayers could be left covering the cost.

Companies are responding to the criticism, though skeptics question how far the responses go. In March 2026, Amazon, Google, Meta, Microsoft, OpenAI, Oracle and xAI signed a White House pledge to build or buy the new power their data centers need and pay for the grid upgrades required to serve them. The pledge is voluntary, and a former Microsoft energy executive has called such announcements meaningless. Microsoft has separately said it will pay utility rates high enough to cover its electricity costs, replenish more water than its data centers consume and publish water-use data for each U.S. region. Amazon says it returned about three gallons of water for every four it used in its data centers in 2025. Regulators are also acting. Virginia approved a new rate class, effective in 2027, that requires large Dominion Energy customers to sign 14-year contracts and pay minimum shares of the grid costs of serving them. Whether those commitments and rules prove enough remains an open question, and as Bolthouse notes, much of the data needed to answer it is undisclosed.

What Communities Are Doing

Citing concerns about the effects of data centers, some neighborhoods are starting to push back. As Brockovich says:

“We have to have some courage to show up, and it’s difficult to do that when you’re up against forces that have all the money and all the intelligence and all the bandwidth in the world.”

Although they face giant corporations, some communities are winning victories. After residents of Monterey Park, CA, learned of plans to replace an office building with a data center, they put Measure NDC, which prohibits data centers within city limits, before voters. It was overwhelmingly approved in June 2026, with about 86% of voters in favor, creating what appears to be the first voter-enacted municipal ban on data centers in the country.

In July 2026, New York became the first state to impose a one-year moratorium on large new data centers. Under an executive order from Gov. Kathy Hochul, the state will not issue discretionary environmental permits to data centers that use 50 megawatts of power or more, enough for a small town, while it develops standards to ensure electricity and environmental impacts do not harm residents.

Other states are hitting the brakes as well. In September 2026, Texas Gov. Greg Abbott directed state regulators to halt all permits sought by data centers until state audits are complete. As of mid-October 2026, ElectricChoice, an independent electricity marketplace, counted 575 active data center restrictions, including moratoriums, bans and zoning rules, across 46 states.

You have a range of options. You can support transparency and environmental protection campaigns, several of which operate nationwide. You can also join local and regional groups that press municipal leaders for restrictions, such as keeping data centers away from residential neighborhoods or requiring renewable energy. Or you can start by learning who serves on your city council, attending public meetings and asking whether any data center proposals have been submitted, then build coalitions with neighbors before your town decides.

AI is here to stay, but a 2025 report from the Center for Biological Diversity, an environmental advocacy group, suggests a path forward. By transitioning data centers to renewable energy, the report says, they could use far less of the carbon budget the U.S. power sector would need to meet the federal government’s 2035 goal for pollution-free electricity. Their share of that budget could be cut to just 4%. That’s a significant drop from the 44% expected under current fossil fuel-driven electricity use. If all data centers switched to circular cooling systems that use recycled wastewater and harvested rainwater rather than evaporation, water use could fall by up to 70%.

The future of hyperscale data centers is evolving rapidly. The buildings that house GPU-powered servers today are likely to change within the next decade. Some approaches use former military bunkers or decommissioned mines, and some data centers are going into the Arctic, where freezing temperatures keep servers cool and hydropower supplies electricity. AI offers remarkable advancements, but the data centers behind it raise environmental and health questions that researchers are working to answer. Communities weighing a proposal are likely to find the benefits and costs depend on the project, the place and how much information the developer makes public.

~

If you’ve found value in this article, please cross-post and restack it!

To support the research and health education of AVFC editorial, please consider making a donation today. Thank you.

 

IPAK-EDU is grateful to A Voice for Choice Advocacy as this piece was originally published there and is included in this news feed with mutual agreement. Read More

Subscribe to SciPublHealth


Science-based knowledge, not narrative-dictated knowledge, is the goal of WSES, and we will work to make sure that only objective knowledge is used in the formation of medical standards of care and public health policies.

Comments


Join the conversation! We welcome your thoughts, feedback, and questions. Share your comments below.

Leave a Reply

De-Corrupting Medicine - a LIVE IPAK-EDU course with James Lyons-Weiler, PhD
  • Feds for Freedom

Discover more from Science, Public Health Policy and the Law

Subscribe now to keep reading and get access to the full archive.

Continue reading