The number that dominates the data center water debate is wrong. Not wrong in the sense of inaccurate. Wrong in the sense of incomplete. When people talk about data center water consumption, they picture cooling towers on a rooftop evaporating thousands of gallons an hour. That is real. That happens. And it accounts for roughly 28% of the total water footprint.
The other 72% happens off-site. At the power plants generating the electricity that these facilities consume around the clock.
Bluefield Research published these figures in late February 2026 in a report titled "The Water-Power Nexus: How Data Centers are Reshaping the U.S. Water Landscape." The numbers reframe the entire conversation. Indirect water consumption from electricity generation is projected to nearly double from 54 billion gallons in 2025 to 91 billion gallons by 2030. On-site cooling water grows more modestly, from 22 billion to 34 billion gallons over the same period. Add them together and data centers are looking at 125 billion gallons of total water consumption by the end of the decade. The off-site portion dominates because 56% of data center electricity still comes from fossil fuels, and thermoelectric power generation is one of the thirstiest industrial processes on the planet. Coal plants consume approximately 19,185 gallons per megawatt-hour. Natural gas runs about 2,800. Over 75% of U.S. data center electricity comes from thermoelectric sources: natural gas at 40%, nuclear at 20%, coal at 15%.
Every kilowatt-hour a data center draws from the grid carries an invisible water cost that never shows up in the facility's WUE metric.
Days after the Bluefield report, a team from UC Riverside, Caltech, and Rochester Institute of Technology posted a paper on arXiv titled "Small Bottle, Big Pipe: Quantifying and Addressing the Impact of Data Centers on Public Water Systems." The authors, Yuelin Han, Pengfei Li, Adam Wierman, and Shaolei Ren, focused on a different dimension of the problem. Not annual averages. Peak demand.
Their central finding: data centers have dramatically higher peaking factors than typical water users. A peaking factor measures the ratio of maximum daily water use to average daily use. Residential and commercial buildings typically land between 1.5x and 2.5x. A data center running evaporative cooling towers peaks at 2.2x. Manageable. But a data center using dry cooling with evaporative assist, the hybrid approach that many operators adopt to reduce water consumption on moderate days, peaks at 6.5 to 10x. Individual facilities can spike as high as 30x their average daily draw on the hottest summer days.
Municipal water systems are engineered for peak capacity. The average daily draw is irrelevant if the system cannot handle the worst day of the year. A data center that looks modest on an annual water consumption report might need 10 times its average draw on a July afternoon in Virginia or Arizona. The treatment plants, the transmission mains, the storage reservoirs, the distribution pipes all have to be sized for that spike.
The "small bottle, big pipe" framing captures it precisely. The annual consumption (the bottle) looks manageable. The peak-day infrastructure requirement (the pipe) is enormous.
The UC Riverside team estimated the cost of building enough municipal water infrastructure to accommodate data center peak demand through 2030. The range depends on growth assumptions and efficiency improvements.
Baseline high-growth scenario: $15 to $58 billion. Baseline low-growth: $7 to $28 billion. Even the most optimistic scenario, assuming 10% annual efficiency gains across the industry, lands at $6 to $24 billion. The peak daily capacity gap ranges from 697 million to 1.45 billion gallons per day of new infrastructure. For context, New York City's entire daily water supply runs about 1 billion gallons. The high-growth scenario requires building more peak water capacity for data centers than NYC uses in a day.
The paper catalogued real-world examples of what this looks like on the ground. A Virginia data center requiring $300 million in infrastructure for 8 million gallons per day at full buildout. Amazon committing up to $400 million for water infrastructure in Louisiana to support a $12 billion campus across three sites. Meta pledging $75 million for public water upgrades in Lebanon, Indiana, where their campus will draw up to 3 million gallons daily. A Wisconsin facility triggering $100 million in infrastructure upgrades for 1.2 million gallons of peak capacity. In one Indiana innovation district, the combined water supply, infrastructure, and wastewater investments exceeded $1.1 billion.
These are not hypothetical numbers. They are commitments and expenditures already underway.
Three funding models exist for data center water infrastructure. Each one creates a different set of winners and losers.
In the corporate-funded model, the data center operator bears the full cost of water infrastructure upgrades. Amazon's $400 million commitment in Louisiana follows this approach. So does Google's arrangement in Leesburg, Virginia, where the company funded $25 million in improvements for 640,000 gallons per day.
Pro rata contributions distribute costs based on usage share, typically governed by state formulas. This model works when the data center is one of many large water users in a region. It falls apart when a single facility represents a disproportionate share of new demand.
The third model is the one communities fear most: ratepayer subsidy. Without corporate partnerships, the cost of water infrastructure upgrades falls on existing residential and commercial ratepayers. The UC Riverside study cited Newton County, Georgia, where water rates could increase 33% over two years to absorb data center infrastructure costs. For a household already paying $80 a month, that is an extra $26 per month to subsidize cooling infrastructure for a facility that generates relatively few local jobs.
Shaolei Ren, one of the study's authors, put it directly: "Even if you have money, the water source is another challenge." Reservoirs and snowpack are finite. Money can build bigger pipes. It cannot fill them.
In October 2025, California Governor Gavin Newsom vetoed AB 93, a bill authored by Assemblymember Diane Papan that would have required data centers to disclose projected water consumption before applying for a business license and certify actual annual usage at renewal. The bill passed the legislature. Newsom killed it.
His explanation: "I am reluctant to impose rigid reporting requirements about operational details on this sector without understanding the full impact on businesses and the consumers of their technology."
The veto happened against a backdrop of aggressive data center lobbying. CalMatters and U.S. News reported that Big Tech blocked California data center legislation, leaving only a study requirement. The state that produces more data center regulation proposals than almost any other ended 2025 with nothing binding on the books.
Meanwhile, a Roanoke Circuit Court judge ruled in February 2026 that a data center's water usage data cannot be considered proprietary information under Virginia's Freedom of Information Act. Google had been operating in Botetourt County behind a nondisclosure agreement with the Western Virginia Water Authority. The judge ordered the records released. Twenty-five of Virginia's 31 localities with data center operations have NDAs concealing water usage data.
Over 300 data center bills were filed across 30-plus states in the first six weeks of 2026 legislative sessions. Georgia's SB 421 would prohibit local governments from signing NDAs that hide energy and water data. Iowa's HF 2447 would require quarterly water usage reports. The legislative terrain is shifting from incentives to oversight. California is the exception, not the rule.
Microsoft, Meta, and Amazon have all pledged to become water-positive by 2030, meaning they will replenish more water than they consume. Microsoft is supporting 76 water replenishment projects across 25 global locations. Meta's Beaver Dam, Wisconsin facility includes 570 acres of wetland restoration. Amazon says its Louisiana facilities will use only "verified surplus water."
The pledges are real in the sense that money is being spent and projects are being built. They are limited in the sense that water offsetting does not work like carbon offsetting. Water is local. Replenishing a watershed 50 miles from a community whose pipes are at capacity does not solve the capacity problem. A data center in Loudoun County, Virginia consumed over 1 billion gallons in 2023. Restoring a wetland in another county does not add treatment plant capacity in Loudoun.
The distinction matters because the infrastructure problem is physical. Pipes have fixed diameters. Treatment plants have rated capacities. Storage reservoirs hold finite volumes. These constraints are not financial. They are engineering limitations that money can address only through construction, and construction takes years.
The data center industry's water problem is larger than most operators have acknowledged, because most operators only measure the water that flows through their own meters. The off-site consumption embedded in their electricity draw is invisible on their sustainability reports, absent from their WUE calculations, and outside their operational control.
Operators who source renewable electricity with low water intensity (solar, wind) carry a fundamentally different total water footprint than operators drawing from fossil-fuel-heavy grids. A data center in Iowa running on wind power and a data center in West Virginia running on coal power may report identical on-site WUE numbers while consuming radically different amounts of total water. The metric the industry uses to track water performance measures less than a third of the actual impact.
Bluefield Research projects data centers will account for 8.9% of total U.S. electricity demand by 2030. The water consequences of that electricity demand will be concentrated in the communities that host the power plants, not just the communities that host the data centers. The political dynamics of that distribution have barely begun to surface.
The cooling industry has spent the last three years focused on reducing on-site water consumption through liquid cooling, immersion, and dry cooler technology. That work matters. But the 72% figure suggests that the biggest lever most operators have for reducing their total water footprint is not a better cooling tower. It is a cleaner grid.
Premium subscribers receive weekly deep dives, Cooling Bytes, competitive landscapes, and full archive access.
Subscribe to Premium