By Jay Lund

. . .

Artificial intelligence (AI) will affect many economic and natural resource sectors as these new technologies develop and mature. We are in the early years of this process. Like most new things, AI has become an object of small and great hopes and fears – from hopes for saving and helping humans to fears for destroying human minds and civilizations. A common concern in the media is AI’s water use and its larger implications. While most AI concerns are speculative in these early days, AI water use is an example of our fears and hopes, as well as how some advocates (and researchers) can seize on public attention as an opportunity for advocacy (and funding).  

Fears and Water

Early days of new technology bring wild fears and hopes as seen in media and public discourse. Americans, as historical leaders of new technologies, have seen these many times, from flying cars of the Jetsons and Star Wars, to vaccines, surveillance technologies and databases, sewers, drinking water chlorination, etc. Some hopes and fears prove illusory (e.g., flying cars), some mostly positive (e.g., vaccines, water chlorination and fluoridation), while others prove to be more mixed (e.g., surveillance technologies and databases, the internet, and automobiles).

The rise of artificial intelligence is built on factories of data and computation, so-called data centers. These large warehouses of networked computers on racks require substantial energy to operate and water for cooling, in addition to physical square footage on the landscape. These computation “factories” have large energy demands that can influence local electricity prices. Their water use is mostly for cooling needs from the heat produced from their electricity use. 

California water discussions are sometimes driven by fears, at times with little scientific basis.  Data center water use has become a subject of fear and concern.  As shown below, California data center water use is mostly modest, but will be larger in some other states having more data center activity and less well developed water infrastructure. 

Estimates of Data Center Water Use in California

Many popular discussions, articles, and media reports reflect concerns for water use from the artificial intelligence industry. Some complain that AI companies and facilities are not “transparent” about their use of energy, water, and other resources, and this is certainly true, likely due to the field’s competitiveness. But too many journalists, academics, and advocates wallow in speculation arising from this lack of explicit water use information.

Here are a range of estimates of AI data center water use for California, based mostly on simple fundamental physics of converting energy use to water use for cooling. I did these calculations and then, perhaps appropriately, checked and explored these estimates using four AI models. 

Here are the results:

1. California has about 15 million square feet (sq ft) of floor space for data centers (about 340 acres). Total data center facility area would be larger, including parking, landscaping, and support buildings. Source: https://www.aterio.io/insights/us-data-centers

2. The energy dissipation needed for data center racks is about 2-12 kw/square meter.

3. At 100% efficiency, this rate of heat dissipation would evaporate 70–420 mm/day of water per square meter of floor space.

4. Major industrial cooling systems seem to have efficiencies of 60-90%, so this expands the range to 80 – 700 mm/day per cubic meter of floor space. This would be 29–255 meters of evaporation annually per square meter of data center floor space, roughly 25–150 times more annual evaporation  than irrigated agriculture, per unit area.

5. So 15 million sq ft (1.4 million square meters) of data center, all operating continuously and using industrial evaporative cooling only, would have a total evaporation of 40 million to 357 million cubic meters of water for California annually, or 32,000 – 290,000 acre-ft per year. 

6. Using the prompt, “How much water is likely to evaporate from data centers in California per year, assuming they are all using mostly evaporative cooling?” several free AI websites provided ranges of estimates, below.  These AI also can provide ranges and sources for calculation assumptions.

Table 1: AI estimates of annual water evaporative losses from California data centers 

The overall range of estimates is broad, 2,300 acre-ft/year to 400,000 acre-ft/year. The still broad 32 – 290 thousand acre-ft (taf) per year water use estimate seems reasonable. A narrower estimate supported by all four estimations would be about 20,000 acre-ft/year. This is a lot of water for you and me, but pales (pails?) compared to total human water use in California, which is about 40 million acre-feet per year.  So AI use is about 0.055 percent of annual human water use in California, and is probably among the more economically effective uses of water.

Using the broader initial AI water use estimate of 32,000 acre-ft/year to 290,000 acre-ft/year, this would be 0.08% to 0.7% of annual human water use in California. This would be enough to supply 10,000–100,000 acres of California’s 7 million acres of irrigated agriculture. 

For some areas outside of the arid West, this new industrial water use comes at a time when many large urban areas face declining use from conservation, and might provide desirable revenues for cities with excess water supply capacity.  All water problems are local.

By the way, my breathing in making the blog post above might well have evaporated more water than occurred (incrementally) from all four AI estimates. 

Lessons

I see some lessons here:

1. Don’t panic over AI data center water use in California. A recent study for Central Arizona found that beer production consumed more water than data centers in that region. (But AI will bring more important concerns, such as the end of human civilization.) 
2. The AI estimates spanned reasonable (and appropriately broad) ranges. AI is useful for quick preliminary estimation.  AI also shows most of its work, especially if well-queried.  AI can help expedite and formalize preliminary estimations for a variety of public and policy assessments, where quantitative estimation is sometimes conveniently omitted from discourse.
3. Beware of shallow discussions, articles, and “technical” reports that lack honest and reasoned estimates, even preliminary estimates. Expect better, with more technically supported policy reports.
4. “Facts are facts, but perception is reality.” So much of our public discourse on water and other subjects is choked by chatter, untamed by reasoned evidence, data, and quantification. Today, with AI, we have little excuse for not attempting and using honest estimates to inform our discussions and tame our fears and hopes.

Alas, despite modern technologies and institutions, our human societies, technology, and understanding ultimately rely on 50,000-year old hardware (our brains!), which evolves slowly and mysteriously.  Unavoidably, we work with individual and collective neural hardware limits.

About the Author

Jay Lund is an Emeritus Distinguished Professor of Civil and Environmental Engineering and Geography at the University of California – Davis.  He is also a Vice Director of the Center for Watershed Sciences.  His 68-year-old hardware with 50,000-year-old architecture is enjoying and struggling with the promise, threats, and turbulence of the AI revolution. 

Further Reading

Kyl Center for Water Policy (2026), Large Non-Agricultural Water Uses in Central Arizona, Arizona State University.

McGuire, M. (2013), The Chlorine Revolution: Water Disinfection and the Fight to Save Lives, American Water Works Association.

Tarr, J. (1984), “A Retrospective Assessment of Wastewater Technology in the United States, 1800-1932,” Technology and Culture, 25 (2), 226-263.Han, et al., (2026) Small Bottle, Big Pipe: Quantifying and Addressing the Impact of Data Centers on Public Water Systems,