Summary

  • AI developers locate hyperscale datacenters in drought-affected US regions, layering a projected 73-billion-gallon annual demand onto depleted regional water stocks.
  • Industry efforts to reduce direct withdrawal via closed-loop cooling transfer water consumption upstream to fossil-fuel power generation, preserving total demand rather than eliminating it.
  • Local opposition and state-level legislative proposals threaten to delay capital deployment, forcing relocation to water-abundant jurisdictions or escalating municipal and agricultural resource rationing.
  • A broader lifecycle analysis attributes only 4% of AI expansion’s mid-century water demand to datacenters, with power generation and semiconductor fabrication representing the primary unmet needs.

A Guardian analysis reveals that 517 of 809 planned US AI datacenters—roughly 64%—sit in areas experiencing drought conditions, a siting pattern driven by lower land costs and generous tax incentives that now collides with intensifying hydrological constraints and local resource rationing. Researchers project national datacenter water demand will scale from 17 billion gallons in 2023 to 73 billion gallons annually by 2028, as hyperscale facilities requiring up to 5 million gallons daily deploy in western and southeastern basins already facing agricultural and municipal shortages. The convergence of accelerated infrastructure rollout, technological cooling trade-offs, and localized political resistance establishes a structural conflict between national technological acceleration and regional water stability that legislative feedback loops may struggle to balance before physical depletion occurs.

Siting rationale and industrial expansion

Proponents of locating large datacenters in arid, sparsely populated regions construct a case based on total cost of ownership and deployment speed: land costs are lower, tax incentives are often more generous, and arid climates minimize equipment corrosion over time. This rationale drives capital investment by Google, Meta, Microsoft, and Amazon into physical infrastructure to secure market dominance, a pace characterized by experts as the industry “sprinting as fast as it can to gain market dominance.”

The Data Center Coalition frames this expansion as a mature operational shift rather than a net new burden. Vice-President of State Policy Dan Diorio states, “Datacenter operators work closely with local authorities to ensure compliance with all applicable rules and regulations and to ensure operations do not stress local water supplies,” adding that the industry is “actively prioritizing responsible water use through operational best practices and innovative development strategies.” Developers argue datacenter water consumption represents a reallocation of existing resources, noting that recreational irrigation for golf courses and lawns consumes more water than their facilities. Analysts at Xylem identify “an emerging consensus among the major hyper-scalers about the importance of water stewardship,” suggesting industry and environmental sectors share a view that long-term allocation management is required.

Water-energy constraints and infrastructure dynamics

A Guardian analysis indicates that 517 of 809 planned US AI datacenters—approximately 64%—are situated in areas that experienced drought conditions over the preceding year, layering new withdrawal demands onto already stressed regional hydrological stocks. National datacenter water demand is projected by researchers to scale from 17 billion gallons in 2023 to as much as 73 billion gallons annually by 2028, with individual hyperscale facilities requiring up to 5 million gallons of water per day for cooling.

The sector’s attempt to mitigate direct water withdrawal through closed-loop cooling introduces a secondary structural dynamic. While closed-loop systems recycle coolant and reduce on-site evaporative losses, they require higher electrical loads to reject heat without evaporation. In regions where local grids rely on fossil-fuelled thermal plants, the additional electricity demand transfers water consumption upstream to power generation sites via cooling towers and embedded fuel cycles, shifting the outflow rather than eliminating it. Meta’s Hyperion project in Louisiana exemplifies this trade-off: the facility will utilize closed-loop cooling and draw from an agricultural rather than community aquifer, but will require the energy equivalent of ten gas-fired power plants.

Experts assess that the compounding demand across populations, agriculture, and new industrial consumers creates a scenario where a “crunch point is inevitable.” However, a global analysis by Xylem places the datacenter footprint in perspective, calculating that datacenters account for just 4% of the additional water needed for AI expansion by mid-century, with power generation and semiconductor fabrication representing the larger shares. The immediate political and hydrological conflicts remain localized to the visible datacenter footprint, while the broader displaced water demand impacts generation and manufacturing watersheds outside municipal boundaries.

Political friction and regulatory response

As competing demands crystallize, opposition has intensified in rural and agricultural communities. Stakeholders highlight the perceived inequity of farmers being mandated to conserve water while industrial datacenters secure near-unlimited access. This friction has triggered legal challenges and localized coalitions, such as the coalition opposing the 9-gigawatt Stratos Project in Box Elder County, Utah, which has been in drought since summer 2025. Opponents argue the development will worsen water deficits threatening the Great Salt Lake, a situation Brigham Young University ecologist Ben Abbott attributes to climate-driven hydrological shifts and a “century of water overuse.”

Public resistance has translated into legislative action. Proposed policy responses include mandatory water-use reporting, requirements for closed-loop cooling technologies, and outright moratoriums on datacenter construction in states such as New York, Michigan, Iowa, South Carolina, and Kansas. Polling indicates 70% of Americans do not wish to reside adjacent to datacenters, and opposition has solidified in traditionally conservative, rural strongholds.

Institutional delays and forward discontinuities

If voter-driven referendums and litigation successfully delay or cancel developments in multiple jurisdictions, capital deployment may be forced to relocate to water-abundant regions or overseas markets, sacrificing the corrosion and land-cost advantages of current siting strategies. In the absence of relocation, localized scarcity events are projected to escalate into direct resource rationing for municipalities and agricultural sectors.

The structural tension presents divergent pathways based on the timeline of institutional feedback versus physical depletion. The balancing loops of regulation, litigation, and public opposition operate with delays potentially longer than the timeline on which regional water stocks tip into critical deficit. Neither the industry’s efficiency claims nor environmental depletion warnings can be definitively resolved before physical constraints are encountered, at which point available options narrow to forced curtailment, emergency water reallocation, or stranded capital.

Institutional delays extend to the technological adjustments required to resolve the “fixes that fail” dynamic inherent in closed-loop cooling requires technological discontinuities in thermal management—such as liquid-immersion or novel adiabatic systems—that bypass the massive fossil-fuel electricity penalties of current alternatives. Until such technological shifts are documented at scale, the competition for water in drought-hit basins presents an unresolved conflict between the acceleration of national technological infrastructure and the stability of local resource basins. The forward trajectory hinges on whether regulatory and market mechanisms can adjust allocation faster than the cumulative water-energy demand outpaces regional supply limits.

Analytical techniques used in this piece

This analysis applies the methods below. Each links to a short, plain-English explainer you can read and reuse.

Steelman Construction
Builds the strongest possible version of a position before judging it.
Systems Dynamics (Structural)
Maps a system’s structure — stocks, flows, and the architecture that shapes its behavior.
Wicked Futures
Explores a long-horizon, deeply entangled future with no clean resolution.
Tragedy of the Commons
A shared resource is depleted because each user’s incentive is to take more.