report<\/a> makes this comparison explicit. US data centers directly consumed approximately 17.4 billion gallons in 2023. That same year, Lawrence Berkeley National Laboratory calculated indirect water usage, through electricity generation, at roughly 211 billion gallons. That is roughly twelve times larger than the direct figure everyone actually talks about.<\/p>\n\n\n\nCalculating a Single Facility’s Daily Draw<\/h4>\n\n\n\n
Here is where the math becomes genuinely useful. The International Energy Agency estimates that a 100-megawatt US data center can directly consume as much water each day as roughly 2,600 households. Using the EPA’s own figure of more than 300 gallons per household per day, this translates to a direct calculation.<\/p>\n\n\n\n
2,600 households \u00d7 300 gallons per day = 780,000 gallons per day for a 100 MW facility<\/p>\n\n\n\n
This lines up closely with independent industry benchmarking. Simple Mining Insights calculated that running the math on a 100 MW facility at industry-average water intensity produces over 1 million gallons per day using conventional evaporative cooling. The gap between these two estimates reflects genuine variation in cooling technology and regional climate, not measurement error.<\/p>\n\n\n\n
Scaling this up to a full 1-gigawatt AI campus, the kind now common in current hyperscaler buildouts, gives a rough estimate.<\/p>\n\n\n\n
1,000 MW \u00f7 100 MW \u00d7 780,000 gallons = 7.8 million gallons per day<\/p>\n\n\n\n
That single number, roughly 7.8 million gallons daily for one gigawatt-scale campus, exceeds the documented daily water use of many entire small American cities.<\/p>\n\n\n\n
Real Company Disclosures and What They Reveal<\/h4>\n\n\n\n
Actual hyperscaler disclosures let us check these estimates against reality directly, and the numbers are illuminating.<\/p>\n\n\n\n
Google reported that its data center water consumption grew from 4.3 billion gallons in 2021 to 6.1 billion gallons in 2024. In total withdrawal terms, Google pulled 7.8 billion gallons in 2024, consuming 78 percent of that through evaporation, with the remainder discharged back into local systems. Dividing the 6.1 billion gallon annual consumption figure across 365 days gives an average.<\/p>\n\n\n\n
6.1 billion gallons \u00f7 365 days \u2248 16.7 million gallons per day, company-wide across all facilities<\/p>\n\n\n\n
Individual facilities vary enormously around that average. Google’s largest single site, located in Council Bluffs, Iowa, withdrew an average of 3.9 million gallons per day on its own. A separate Google facility in Virginia consumed 173.2 million gallons across a full year, which works out to a per-day figure.<\/p>\n\n\n\n
173.2 million gallons \u00f7 365 days \u2248 474,500 gallons per day for that specific Virginia site<\/p>\n\n\n\n
Microsoft’s disclosed figures tell a similar growth story. The company reported total water consumption reaching nearly 6.4 million cubic meters in its most recent reporting year, converting to approximately 1.69 billion gallons, a 34 percent increase over the prior year. Equinix, operating 268 data centers worldwide, reported withdrawing 1.4 billion gallons while consuming 1.2 billion gallons, meaning roughly 85 percent of what it pulled from local systems evaporated rather than returning to the watershed.<\/p>\n\n\n\n
Why Cooling Technology Changes Everything<\/h4>\n\n\n\n
The single biggest variable in data center water usage is not facility size at all. It is cooling architecture, and the numbers here are genuinely stark.<\/p>\n\n\n\n
Traditional evaporative cooling consumes approximately 1.8 million gallons per megawatt annually, according to industry benchmarking from Kova Stack. Nvidia’s own disclosed figures for conventional cooling-tower systems land close to this, at roughly 2.6 million gallons per megawatt per year. Air-cooled and adiabatic cooling designs reduce this by a factor of six to thirty, depending on climate.<\/p>\n\n\n\n
The most dramatic recent shift comes from direct liquid cooling. Nvidia’s Rubin architecture, described by the company as its first entirely liquid-cooled AI infrastructure platform, uses closed liquid loops operating at temperatures up to 45 degrees Celsius. According to Nvidia’s own director of data center cooling, this design can reduce facility cooling water consumption from roughly 2.6 million gallons per megawatt annually down to near zero, since these closed systems rely on dry coolers rather than evaporation for approximately 99 percent of the year.<\/p>\n\n\n\n
Running the calculation on a 100 MW facility makes this concrete. Under conventional cooling, annual consumption reaches:<\/p>\n\n\n\n
100 MW \u00d7 2.6 million gallons per MW = 260 million gallons per year<\/p>\n\n\n\n
Under Rubin’s closed-loop liquid cooling architecture, that same 100 MW facility approaches zero gallons of cooling-related water consumption annually. This is the single largest lever available to the entire industry for reducing its water footprint, and it explains why hyperscalers are moving toward it despite higher upfront capital costs.<\/p>\n\n\n\n
The National Trajectory<\/h4>\n\n\n\n
Zooming out from individual facilities to the entire country reveals a trajectory worth stating plainly. Direct data center water usage in the United States grew from 5.6 billion gallons in 2014 to 17.4 billion gallons in 2023, more than tripling in under a decade. The MOST Policy Initiative, drawing on EPA and Lawrence Berkeley National Laboratory projections, forecasts this figure will reach between 38 and 73 billion gallons by 2028.<\/p>\n\n\n\n
Applying the lower end of that range gives a sense of the growth rate involved.<\/p>\n\n\n\n
38 billion gallons (2028 projection) \u00f7 17.4 billion gallons (2023 actual) \u2248 2.2 times growth in five years<\/p>\n\n\n\n
Even the conservative end of this projection represents a genuinely rapid expansion, driven directly by the computing demands of AI training and inference specifically, according to the Congressional Research Service’s own analysis.<\/p>\n\n\n\n
Putting the Numbers in Human Terms<\/h4>\n\n\n\n
A useful final calculation helps translate all of this into something genuinely relatable. The EPA states that an average American household uses more than 300 gallons of water per day. Using the earlier hyperscale campus estimate of 7.8 million gallons daily for a 1-gigawatt facility, we can calculate how many households that equates to.<\/p>\n\n\n\n
7.8 million gallons \u00f7 300 gallons per household = 26,000 households served by one facility’s daily water draw<\/p>\n\n\n\n
This figure sits squarely within the range multiple sources independently confirm. EESI states that large data centers can consume up to 5 million gallons per day, equivalent to a town of 10,000 to 50,000 people. Kova Stack’s independent estimate matches this range almost exactly. The consistency across multiple independent sources, using different methodologies, gives genuine confidence that these figures are not outliers or exaggerations.<\/p>\n\n\n\n
What This Means Going Forward<\/h4>\n\n\n\n
Data center water usage is not a fixed, unavoidable cost of computing. The gap between conventional evaporative cooling, at roughly 2.6 million gallons per megawatt annually, and next-generation liquid cooling, approaching near zero, demonstrates that the technology to dramatically reduce this footprint already exists and is being deployed today. The remaining question is one of pace and capital allocation, not technical feasibility.<\/p>\n\n\n\n
For communities weighing new data center proposals<\/a>, and for enterprises evaluating their own AI infrastructure choices, the calculations in this article offer a genuinely useful starting framework. Multiply a proposed facility’s megawatt capacity by 2.6 million gallons for a conventional cooling estimate, or by a fraction of that figure if liquid cooling is specified, and compare the result directly against local water availability. That single calculation, more than any headline statistic, is what actually determines whether a specific data center’s water usage is sustainable for the community hosting it.<\/p>\n","protected":false},"excerpt":{"rendered":"A Number That Rarely Gets Shown With Its Math Data center water usage gets cited constantly in news coverage. Millions of gallons per day. Billions per year. What rarely gets shown is the actual arithmetic behind those figures. This article walks through real numbers from the Congressional Research Service, Lawrence Berkeley National Laboratory, Google, Microsoft, and Equinix. Every figure comes with its calculation shown explicitly, so the scale becomes genuinely concrete rather than abstract.<\/p>\n","protected":false},"author":1,"featured_media":1345,"comment_status":"open","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[7,39],"tags":[],"class_list":["post-1344","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-ai-news-industry-updates","category-ai-ethics-and-governance"],"_links":{"self":[{"href":"https:\/\/learnerbox.net\/blog\/wp-json\/wp\/v2\/posts\/1344","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/learnerbox.net\/blog\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/learnerbox.net\/blog\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/learnerbox.net\/blog\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/learnerbox.net\/blog\/wp-json\/wp\/v2\/comments?post=1344"}],"version-history":[{"count":1,"href":"https:\/\/learnerbox.net\/blog\/wp-json\/wp\/v2\/posts\/1344\/revisions"}],"predecessor-version":[{"id":1346,"href":"https:\/\/learnerbox.net\/blog\/wp-json\/wp\/v2\/posts\/1344\/revisions\/1346"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/learnerbox.net\/blog\/wp-json\/wp\/v2\/media\/1345"}],"wp:attachment":[{"href":"https:\/\/learnerbox.net\/blog\/wp-json\/wp\/v2\/media?parent=1344"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/learnerbox.net\/blog\/wp-json\/wp\/v2\/categories?post=1344"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/learnerbox.net\/blog\/wp-json\/wp\/v2\/tags?post=1344"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}