Who pays for the cloud's thirst?

Data-center water use is real, uneven, and often opaque. Ask for disclosure before you treat the cloud as weightless.

The trap

Chat interfaces feel weightless. The water used to cool servers and to generate their electricity does not. Privette et al. (2026) argued that AI-driven data-center growth raises water concerns that are especially sharp in already stressed regions, and that weak public disclosure blocks regulation, innovation, and community planning.

What the evidence shows

Privette et al. (2026) described a multi-part water footprint (on-site cooling, electricity generation, and supply chains) and stressed localized impacts even when national totals look modest. The International Energy Agency (2025) estimated that data centers used about 415 TWh of electricity in 2024 (about 1.5% of global electricity) and projected substantial growth through 2030 in its base case. The Congressional Research Service (2025) summarized U.S. data-center electricity estimates and, citing International Energy Agency analysis, noted an illustrative figure that a 100 MW U.S. data center may consume on the order of 2 million liters of water per day across cooling strategies, with a large share sometimes on site.

Chen and Wemhoff (n.d.) showed that on-site and off-site water burdens differ by cooling design and location, and proposed scarcity-aware metrics so a water-saving choice in one place is not assumed everywhere.

What this means for people

Someone always pays: ratepayers, neighbors in drought-prone basins, or future capacity. Enthusiasm for models does not erase the physical plant.

Practice (15 minutes)

Pick one AI or cloud vendor you use. Find their latest water or sustainability disclosure (or note that you cannot).
Write three questions: Where are the facilities? What is reported for water? Is the region water-stressed?
For one workflow, ask whether a smaller model, fewer idle jobs, or batching would cut needless load.
Share the unanswered questions with one colleague who buys or configures tools.

Reflection

Where did “the cloud” feel free until you tried to open a number?

Skeptic check

Facility-level water data remain incomplete; averages hide local stress (Privette et al., 2026).
Illustrative liters-per-day figures are not your vendor’s meter reading (Congressional Research Service, 2025).
Growth scenarios in energy outlooks are projections, not destiny (International Energy Agency, 2025).

References

Chen, L., & Wemhoff, A. P. (n.d.). Characterizing data center cooling system water stress in the United States. NSF Public Access Repository. https://par.nsf.gov/servlets/purl/10341797

Congressional Research Service. (2025). Data centers and their energy consumption: Frequently asked questions (CRS Report R48646). https://www.congress.gov/crs-product/R48646

International Energy Agency. (2025). Energy and AI. https://www.iea.org/reports/energy-and-ai/energy-demand-from-ai

Privette, A. P., Barros, A., & Cai, X. (2026). Data centers water footprint: The need for more transparency. AGU Advances, 7(2), Article e2025AV002140. https://doi.org/10.1029/2025AV002140