The servers hum nonstop. They draw power at scales once reserved for entire cities. And the environmental price keeps climbing. A single large AI-focused facility can consume as much electricity as 100,000 households, according to the International Energy Agency. Yet the true scale emerges only when those facilities multiply across continents.
Last year global data centers used 448 trillion watt-hours of electricity. That exceeds the annual consumption of all but 10 countries worldwide. The resulting carbon dioxide output reached about 208 million tons, matching Argentina’s total. Producing that energy also consumed roughly 1.2 trillion gallons of water, the Associated Press reported, citing a United Nations University study released in June 2026.
But those figures already feel dated. Projections show data-center electricity demand more than doubling to around 945 terawatt-hours by 2030. AI will drive nearly half the growth. The IEA updated its analysis just this week, confirming accelerated servers tied to AI workloads will grow 30 percent annually through the end of the decade. Conventional servers lag far behind at 9 percent.
Daily AI use, not just model training, accounts for the bulk of demand. One popular service processes 2.5 billion prompts each day. That activity alone burns through hundreds of gigawatt-hours yearly. The public fixates on dramatic training runs for frontier models. Reality shows inference and routine queries dominate the load, the UN University researchers found.
Water use tells an even starker story. By 2030 data centers could consume up to 9.3 trillion liters annually. Enough to meet basic drinking needs for 1.3 billion people. Or, put another way, more than the annual household requirements of everyone living in Sub-Saharan Africa. Cooling towers evaporate the majority. Power-plant generation for the electricity adds indirect thirst that often goes uncounted in company reports.
In the United States two-thirds of new data centers planned since 2022 sit in regions already facing high water stress. The Guardian analyzed permitting data and found facilities rising fastest where aquifers and rivers strain under drought. Local utilities in Virginia and Arizona have watched bills spike for residents as plants divert millions of gallons daily for evaporative cooling.
One 100-megawatt AI center can pull 1.5 to 3 million cubic meters of water per year just for cooling. Scale that across the hyperscalers building dozens of such sites and the numbers compound quickly. Microsoft, Google and Meta disclose broad sustainability goals. Their granular, facility-level water accounting remains patchy at best.
Emissions follow similar trajectories. The UN University study pegs current data-center CO2 at 189 million metric tons. That climbs to 399 million tons by 2030 under business-as-usual growth. Equivalent to the annual tailpipe output of nearly 87 million gasoline cars. And this excludes the backup diesel generators that fire up during grid outages or peak demand.
Texas offers the clearest window into those backup realities. The state has become the epicenter of U.S. AI infrastructure. Companies there exploit regulatory loopholes to install onsite natural-gas generation and fleets of diesel generators without full environmental permitting. A Floodlight investigation detailed how 38 data centers in Texas operate more than 2,100 diesel backup units. Those alone emit 2,500 tons of nitrogen oxides each year. The findings, covered by Futurism, highlight tactics of announcing modest projects then quietly expanding capacity after permits lock in.
Researchers at Cornell University modeled the cumulative effect. They project the AI-driven boom in Texas could produce 24 to 44 million metric tons of additional CO2 by 2030. That matches the yearly output of 5 to 10 million passenger vehicles. James Doty, a former staffer at the Texas Commission on Environmental Quality, told investigators the only realistic intervention window closes once permits issue. After that, enforcement proves nearly impossible.
Global Energy Monitor data shows Texas installing more natural-gas generation capacity for data centers than any nation except China. The plants often run as peaker units. They burn dirtier during the exact hours when renewable output dips. The pattern undercuts claims that data centers will accelerate clean-energy adoption. In practice many simply bid up wholesale power prices and lock in long-term fossil contracts.
Europe faces parallel pressures but with stricter rules. Ireland already derives more than 20 percent of its national electricity from data centers. That share could hit 32 percent by 2026. The European Union now requires new facilities to source 80 percent renewable power on average. Compliance varies. Hyperscalers sometimes rely on renewable-energy certificates that do little to add actual clean generation to the local grid.
Australia offers another cautionary case. Data-center demand there could triple both electricity and water consumption by 2030, according to the Climate Council. Much of that growth lands in states still heavily reliant on coal. Waste heat from the facilities offers limited reuse in a warming climate. Northern European examples of district heating from server exhaust remain exceptions, not the rule.
Policy responses have started to appear. The U.S. Senate introduced the Artificial Intelligence Environmental Impacts Act in June 2026. The bill would direct the Environmental Protection Agency to study full lifecycle effects, including upstream emissions, water quality, noise and grid impacts. It also calls for a standardized reporting system. Whether the measure gains traction in a divided Congress remains uncertain.
United Nations Secretary-General António Guterres has pressed the issue directly. In a July 2026 video statement he urged every major AI company to measure and publicly disclose the full footprint of its systems. He repeated an earlier call for all data centers to run on renewable energy by 2030. “AI may feel intangible,” Guterres said, “but its footprint is not.” Community opposition has grown in multiple countries as residents connect rising electric bills and dry wells to nearby construction cranes.
Tech executives counter that efficiency gains will blunt the worst outcomes. Newer chips from NVIDIA and others promise better performance per watt. Liquid cooling and advanced heat recovery could cut water use. Yet historical patterns suggest efficiency improvements often enable greater total deployment rather than absolute reductions. The IEA calls this the “rebound effect.” Accelerated AI adoption simply creates more demand that outpaces per-unit savings.
Some operators explore alternatives. Underwater data centers have moved beyond concept stage. One startup raised $140 million to develop surface-level ocean platforms. Proponents argue seawater provides free cooling and proximity to offshore wind. Critics point to maintenance challenges, environmental risks to marine life and the fact that most AI workloads still cluster near population centers with existing fiber.
Nuclear power has reentered the conversation. Hyperscalers have signed deals to restart retired reactors or fund small modular reactors tailored to data-center loads. Such projects take years to license and build. In the interim natural gas fills the gap. Goldman Sachs forecasts data-center power demand will rise 160 to 165 percent by 2030 compared with 2023 levels. That pace leaves little room for delay.
Local impacts often receive less attention than global totals. Noise from cooling fans travels miles in rural areas. Light pollution from 24-hour facilities disrupts wildlife. Construction disrupts aquifers. Once online, the constant low-level vibration and electromagnetic fields raise concerns among nearby residents. These effects compound in communities already hosting multiple campuses.
So the trade-offs sharpen. AI promises gains in drug discovery, climate modeling and energy optimization. Those advances require massive computation today. Tomorrow’s models will demand even more. The question facing policymakers, utilities and corporate boards is whether society can afford the resource bill without deliberate constraints on deployment or aggressive mandates for cleaner infrastructure.
Recent analysis from The Guardian argues the benefits must demonstrably outweigh the costs. Otherwise data centers risk slowing the broader clean-energy transition by consuming renewable supply that would otherwise displace coal and gas elsewhere. In Australia, for instance, state governments have clashed with federal guidance over whether fossil fuels should power new facilities at all.
Transparency remains the immediate lever. Most companies report Scope 1 and 2 emissions at the corporate level. Facility-specific water withdrawal, diesel runtime hours and actual renewable additionality stay opaque. The UN University’s report calls for standardized, audited disclosure of carbon, water and land footprints tied directly to AI workloads. Without that data, meaningful regulation or consumer pressure cannot take shape.
Engineers continue tweaking. Direct liquid cooling, higher-temperature chips and waste-heat recapture all show promise in pilots. Yet the aggregate curve still points upward. Electricity consumption for data centers grows four times faster than the rest of the global economy. AI accounts for the steepest segment of that line.
And the momentum shows no sign of easing. Venture funding, government AI strategies and consumer adoption all push in the same direction. Every new chatbot query, image generation and autonomous-driving inference adds an invisible draw on power plants and rivers. The infrastructure making modern AI possible now rivals entire nations in its resource demands.
Whether that bargain proves sustainable depends on choices made in the next few years. Faster permitting for renewables and nuclear. Strict water-performance standards for new builds. Mandatory real-time emissions reporting. Or continued reliance on loopholes, diesel backups and stressed grids. The servers will keep humming. The question is what else hums alongside them.
WebProNews is an iEntry Publication