Why AI datacenters are being built on drought-hit land — and what business leaders must do

When hyperscalers and chip-hungry startups race to add capacity for AI agents and large language models, they’re not just chasing cheap floor space. A recent analysis of planned U.S. facilities found that of 809 proposed datacenters, roughly 517 — about 64% — are slated for counties that experienced drought over the past year (Guardian analysis; U.S. Drought Monitor data). That pairing of compute growth with shrinking water supplies is no longer a niche sustainability story. It’s an operational, regulatory and reputational risk that belongs on every CFO and CIO dashboard.

The scale of the problem

Datacenters are thirsty. Large facilities can use as much as ~5 million gallons of water per day for cooling under some designs — roughly the daily water needs of tens of thousands of people. U.S. datacenter water demand is estimated to climb from about 17 billion gallons in 2023 to as much as 73 billion gallons per year by 2028 under high-growth scenarios (industry estimates).

Small units of compute add up. One modeling estimate (arXiv) puts the water footprint of a single 100‑word prompt at roughly 500 ml when you account for the entire cooling and power footprint. Multiply that by millions or billions of queries across consumers, enterprises and AI automation workflows, and the numbers become material for local water managers and regional utilities.

At the same time, more than 60% of the contiguous U.S. was in some stage of drought at the time of reporting — the largest spring drought extent on record in modern monitoring (U.S. Drought Monitor). Choosing an arid county for scale economics can therefore look like short-term realism and long-term risk.

Why builders pick drought‑prone sites

Cheap land, favorable tax packages, low local permitting friction and lower corrosion risks for hardware are compelling incentives. Rural jurisdictions often trade tax incentives and promised jobs for new investment. Developers also factor in grid access, fiber routes and local political environments. Put together, these variables can make a dry county look like a fast lane to scale.

That calculus ignores the watershed. Water availability, aquifer resilience and competition with agriculture or municipal needs are starting to matter as permitting fights, lawsuits and moratoria show up in planning timelines.

The water–energy trade‑off, explained plainly

Cooling systems determine the water story. Here’s a quick primer for executives considering site strategy:

• Evaporative cooling: Uses water to lower intake air temperature; low electricity cost but high water consumption.
• Air (dry) cooling: Uses fans and heat exchangers with minimal water but higher energy use, especially in hot climates.
• Closed‑loop liquid cooling: Recirculates coolant and dramatically reduces evaporative losses; requires more pumps and often more electricity.
• Immersion cooling: Servers are submerged in dielectric fluids; can slash water use and footprint but raises capex and operational changes.
• Reclaimed or brackish water: Uses non‑potable sources to avoid freshwater withdrawals but faces permitting and treatment hurdles.

Those options reveal a core trade-off: cutting water often raises electricity demand. Electricity generation itself can be water‑intensive (thermal power plants require cooling), and if the additional power comes from fossil-fuel sources, you trade water for carbon. As Xylem’s analysis puts it, datacenters are the most visible part of AI’s water footprint, but power generation and semiconductor manufacturing will drive a larger share of total AI‑related water demand.

“The industry is racing to dominate markets and that rapid buildout is driving substantial new water demand in areas already short on water; a supply crunch is likely,” warns a water-resources law expert.

Community, legal and policy backlash

Local communities often see the optics clearly: farmers and ranchers are asked to conserve while large industrial users request new water access. That tension shows up in public opposition; Gallup polling suggests roughly 70% of Americans don’t want to live next to a datacenter — a blunt political reality for site planners.

Several states have moved to respond. California, Michigan, Iowa, South Carolina and Kansas are among those tightening reporting, permitting or withdrawal requirements for industrial water users. New York has proposed a moratorium on new datacenter approvals while it reconsiders policy. Those shifts signal that regulatory risk is no longer hypothetical.

“We are being asked to conserve while new projects gain access to water,” says a regional agriculture tech CEO, summing up why opposition often coalesces around shared local priorities like food and hygiene.

Case studies: Stratos and Hyperion

The Stratos project in Utah — backed by a high‑profile investor — and Meta’s Hyperion facility in Louisiana illustrate the stakes. Stratos became a focal point for coalition-building across environmental groups, ranchers and residents concerned about aquifer drawdown and impacts to nearby water bodies. Hyperion prompted scrutiny over its closed‑loop cooling and significant power needs in a state with existing water stress in places.

Those projects show how a single siting decision can cascade into lawsuits, permitting delays and reputational costs — all of which translate to higher lifetime operating expenses and political risk.

Quantifying the risk for business decisions

Water risk is now a material input for total cost of ownership (TCO). Key variables to stress-test when modeling a build include:

• Local watershed stress and recharge rates (short, medium and long-term climate projections)
• Type of cooling technology and its water vs. energy profile
• Availability and costs of alternative water sources (reclaimed water, brackish water)
• Projected local electricity mix and the water footprint of power generation
• Permitting timelines, legal exposure and community-opposition scenarios
• Contingency costs for mitigation measures and water‑stewardship investments

Simple scenario: a 100 MW datacenter using evaporative cooling in a hot, arid county will have drastically different TCO and reputational exposure than the same datacenter using immersion cooling and reclaimed water with renewable power contracts. The site that looks cheapest on rent per square foot often loses when you add regulatory delay, mitigation costs and social license erosion.

Executive checklist: 30/90/180 day actions

• Immediate (0–30 days): Add watershed stress as a hard site-selection filter. Require a water-stress score for every new site proposal.
• Short term (30–90 days): Run a water–energy tradeoff analysis on planned projects (compare evaporative vs. closed‑loop vs. immersion at local climate conditions).
• 90 days: Commit to transparency — publish planned water use, cooling technology, and the source of water (fresh, reclaimed, brackish) for all major projects.
• 3–6 months: Pilot low‑water technologies (immersion, closed‑loop with heat reuse) and monitor real performance publicly.
• 6 months and ongoing: Engage communities early, fund local water projects or water banking, and include binding water‑stewardship clauses in permits and vendor agreements.

Quick facts

How many planned U.S. datacenters were analyzed?

Analysts reviewed 809 planned facilities and found about 517 (roughly 64%) are in locations that experienced drought over the prior year (analysis based on Cleanview site data and U.S. Drought Monitor classification).

How much water can a large datacenter use?

One large datacenter can require up to ~5 million gallons per day for cooling under certain designs; sector demand could rise to roughly 73 billion gallons/year by 2028 versus about 17 billion in 2023 (industry projections).

Are datacenters the main driver of AI’s water footprint?

They are the most visible part, but studies (e.g., Xylem) suggest datacenters may represent only about 4% of the extra global water demand tied directly to AI through mid‑century — with power generation and semiconductor manufacturing accounting for far more.

What is the public and political response?

Roughly 70% of Americans report they would not want to live next to a datacenter (Gallup); multiple states are tightening reporting or considering limits and New York has proposed a moratorium on new approvals.

Longer-term policy and sector choices

Policymakers face a choice: treat water as a first‑order siting criterion, or allow market forces and local incentives to continue driving projects into the driest basins. Regulators are already moving; expect more reporting requirements, municipal-level conditional approvals, and even state moratoria in high‑stress watersheds.

For operators and enterprise buyers of AI for business, the strategic imperative is clear. Building compute capacity without watershed due diligence is a form of leverage that can snap back — through regulation, social conflict or resource depletion. Companies that treat water stewardship as integral to infrastructure strategy will avoid permit fights, reduce lifecycle costs and win goodwill with communities and regulators.

“Choosing cheap land without watershed due diligence is a false economy — and companies that plan for water, not just compute, will win the next decade of AI deployment.”

How the site count was measured

The 809 figure reflects a recent industry analysis of proposed datacenter projects across the contiguous U.S. Analysts cross-referenced planned site coordinates with federal drought classifications from the U.S. Drought Monitor and Cleanview site lists to identify projects in counties that experienced drought over the prior year. Methodological caveats: project lists include announced and permitted sites; not all plans move to construction, and local permitting changes can alter outcomes.

Final practical note for leaders

AI infrastructure growth is not optional; demand for compute to run AI agents and automation will rise. The shortcut is to stop treating water as an afterthought. Make watershed resilience a first‑order planning metric, stress-test cooling and power choices, and invest in community-aligned water stewardship. Doing so turns a rising risk into a competitive advantage — one that will matter as much as kilowatts and rack density when the next round of projects is evaluated.