What AI Data Centers Are Costing Their Neighbors, Their Grids, and the Climate
AI's most visible breakthroughs happen on screens, but its physical infrastructure is reshaping power grids, watersheds, and neighborhoods in ways most people only notice when the bill arrives. The data centers that power generative AI are now the fastest-growing industrial load in the United States, and the costs of that growth are landing disproportionately on the people who live near them and pay the same utility company they do. There are genuine benefits to the boom, but the harms are well-documented, accelerating, and increasingly hard to defend.
A Grid Built for People, Not For AI
We can start by looking at electricity. The current electrical grid is already oversaturated. Hundreds-of-thousands of Americans are currently working on switching to solar or other sustainable energy sources to help alleviate strain on our grid. Governments are even offering special credits and tax incentives for folks who make the switch. But a change is coming that may need to speed up the change even further.
Data centers consume roughly 4–5% of US electricity today; by 2030 the Electric Power Research Institute and other analysts expect them to hit 9–17%. A January 2026 report projects combined demand will nearly double between 2025 and 2028, from 80 to 150 gigawatts. Texas utility CenterPoint reported a 700% jump in large-load interconnection requests in just over a year alone.
That demand is arriving faster than clean supply. Planned natural gas capacity rose from 11% of new builds in 2024 to 18% in 2026, and non-renewable additions surged 71% year over year while renewable growth flattened to 2%. Pacifico Energy's GW Ranch in West Texas, approved in January, is now the largest gas project in the country at up to 7.7 GW—built to power a private grid for data centers. If 300 TWh of new demand is met with today's mix, US power-sector emissions could rise by 125–156 million metric tons a year, roughly 10% of current emissions.
The Water You Don't See
A single large data center can pull up to 5 million gallons of water a day for cooling. In Northern Virginia, the country's data center capital, those facilities draw from the same Occoquan watershed that supplies about 40% of the region's drinking water. By 2027 global data centers are projected to consume around 5 billion cubic meters of water annually; in the US alone, AI servers are expected to add 200–300 billion gallons of annual water use and 24–44 million metric tons of CO₂-equivalent emissions by 2030. Despite that scale, data centers remain among the least transparent industrial water users in the country.
Someone Has to Pay for All Those New Power Lines
US retail electricity prices are up nearly 40% since 2021, and 7% in 2025 alone. Data centers aren't the only reason, but they're a growing one. The mechanics are dull but important: when a utility builds new transmission and generation to serve a giant new customer, it traditionally spreads those costs across all ratepayers. One Virginia resident's bill jumped from about $100 to $281 in a single month early this year. A January 2026 Harvard Law analysis concluded that residential customers are subsidizing infrastructure that exists only because of hyperscale computing facilities.
State legislatures have noticed. Lawmakers in more than 30 states introduced over 300 data-center-related bills in 2026, ranging from outright moratoriums to rules requiring data centers to pay the full cost of their grid connections. Microsoft and Anthropic have publicly pledged to cover their own electricity-related cost increases — a useful precedent, but not yet an industry norm. The tax math isn't what was promised.
The economic case used to sell data centers locally has weakened as the numbers come in. Virginia gave up $1.6 billion in sales and use tax revenue from data centers last year, a 118% jump in a single year. Georgia is expected to lose at least $2.5 billion this year—664% more than the state's previous estimate. Three states (Virginia, Georgia, and Texas) are each losing more than $1 billion a year to data center incentives. Estimated forgone K–12 education funding in Virginia rose from $107.7 million in 2022 to $267.4 million in 2024.
The jobs case is also thinner than the marketing suggests. A typical hyperscale data center employs 50–150 people once built—vastly fewer than a comparable factory or warehouse. Construction jobs are real but temporary. As one Georgia state lawmaker put it bluntly this year, “these data centers aren't bringing jobs.”
And Then There's The Noise
Anyone who has stood next to a data center knows the sound: a continuous low-frequency hum from chillers and, increasingly, the whine of on-site gas turbines. Sterling, Virginia residents living near a Vantage facility describe a constant buzzing “in your head.” Measured noise levels at Virginia data centers run between 40 and 59 decibels — bumping against EPA's 55 dBA outdoor and 45 dBA indoor health thresholds. A January 2026 peer-reviewed paper in Frontiers in Climate concluded the noise, air, and water concerns represent “legitimate public health risks that demand policy attention.” A Fortune analysis in April put the annual hidden US health and environmental damage from data centers at $25 billion, and projected the air-pollution health burden could grow from $6 billion in 2023 to over $20 billion by 2028 — including an estimated 600,000 asthma cases and 1,300 premature deaths.
Public opinion has moved in step. In a Washington Post/Schar School poll, 59% of Virginia voters now say they'd be uncomfortable with a new data center in their community — up from 24% just two years earlier.
What Data Centers Do Bring
None of this means the facilities are valueless. AWS alone reports investing $108 billion in US cloud infrastructure and contributing nearly $38 billion to GDP, supporting an average of 30,000 jobs per year in the communities where it operates. Industry studies find that each direct data center job creates roughly seven ancillary jobs in construction, electrical trades, security, and maintenance. In Loudoun County, Virginia, data center tax revenue funded $1 billion in road improvements and helped build 36 new schools over 15 years. Data centers are also genuine critical infrastructure: the cloud services, AI tools, hospital systems, and financial networks the country runs on all live inside them.
The real question, as Brookings and the AAAS have argued, isn't whether data centers are good or bad — it's how the benefits and costs get allocated, and whether the communities absorbing the impacts have any meaningful say.
The Bottom Line
The AI boom is producing extraordinary technology and an extraordinary external bill: higher household electricity rates, stressed watersheds, rising emissions in a decade we can least afford them, billions in forgone tax revenue, and measurable health impacts on the people nearest the buildings. Some of this is fixable — closed-loop water cooling, dedicated rate structures so hyperscalers pay their own way, clean firm power commitments, stricter siting and noise rules. Some of it is going to require politicians and regulators to do their jobs faster than the build cycle. What's no longer reasonable is pretending the costs aren't real. They show up every month, on someone's bill, in someone's window, in someone's air.
