• Hyperscale AI infrastructure expansion in Western desert states is increasingly constrained by water access rather than land, with projects facing redesigns, delays, and political challenges due to cooling-water demands and wastewater capacity limits.
• Major tech companies' data centers consumed massive water volumes in 2024—Meta withdrew 35M+ gallons in Utah, Google used 352M+ gallons in Nevada—triggering new public reporting requirements and permit restrictions across water-stressed states.
• Explosive AI infrastructure capex ($52.5B Alphabet, $77.7B Amazon) risks becoming stranded assets as water scarcity, permitting delays, and mandated cooling-system redesigns erode margin assumptions and create potential bubble conditions.

Hyperscale AI infrastructure is expanding rapidly across Western desert states, where land, tax incentives, and grid interconnects are often easier to secure. But water — not land — is emerging as the binding constraint. Permits, ordinances, and public records now show projects being redesigned, delayed, or politically challenged because cooling-water access and wastewater capacity can no longer be assumed.

This is no longer a sustainability footnote. Has it emerged as a capital allocation risk?

Source: NASA

Desert-State Flashpoints

Utah. Reporting from the Salt Lake Tribune indicates (opens in a new tab) Meta’s Eagle Mountain campus withdrew more than 35 million gallons in 2024. A prior agreement reportedly treated monthly water-use data as confidential. Meanwhile, Utah lawmakers have advanced proposals requiring public reporting of large data-center water consumption — signaling growing political scrutiny.

Arizona. Business reporting shows Microsoft’s original design for its Goodyear campus (opens in a new tab) https://www.goodyearaz.gov/Home/Components/News/News/9847/1549?arch=1 (opens in a new tab) contemplated roughly one million gallons per day per planned building before redesigns shifted later phases toward mechanical air cooling and expanded wastewater infrastructure. In Tucson, the controversial “Project Blue” proposal helped trigger a new ordinance (opens in a new tab) requiring very large water users to submit conservation plans prior to accessing municipal supply. Public records reporting has linked the project to Amazon Web Services.

Nevada. The Las Vegas Review-Journal reported Google’s Henderson facility (opens in a new tab) withdrew more than 352 million gallons in 2024. Across Southern Nevada, 23 facilities reportedly consumed more than 716 million gallons combined, much of it tied to Colorado River allocations via Lake Mead. The region has since adopted permit restrictions limiting new developments that rely on evaporative cooling systems.

The pattern is clear: water access is now a political variable in AI siting decisions.

The Mechanics of Thirst

Cooling drives water consumption.

Evaporative and cooling-tower systems shed heat by evaporating water. Dry-air or refrigerant-based systems reduce on-site water usage but generally increase electricity demand — shifting the burden upstream.

The industry metric, Water Usage Effectiveness (WUE), measures liters of water per kilowatt-hour (kWh) of IT load.

 Of course, rare earth elements (REEs) are used extensively in the cooling systems of modern data centers, particularly to enhance the efficiency of AI-driven, high-density infrastructure. They are primarily used to create high-performance permanent magnets in motors, fans, and pumps that move air and liquid coolant. This is the topic of another article.

The 2024 Lawrence Berkeley National Laboratory U.S. Data Center Energy Usage Report (opens in a new tab) estimates:

• Average direct on-site WUE across U.S. data centers: ~0.36 L/kWh through 2023
• Risingtoward ~0.45–0.48 L/kWh as liquid-cooled AI servers expand
• Roughly 0.8–1.1 million gallons per MW of IT load per year if run continuously
• Indirect water consumption tied to electricity generation: ~4.5 L/kWh in 2023, often exceeding direct site use depending on grid mix

In short, cooling water is only part of the story. Power generation water often dwarfs facility-level usage.

Note, this topic resonates locally in Utah. In fact, Salt Lake City and the broader Utah grid are facing mounting energy stress as explosive growth in data centers, tech campuses, and electrification (all needing rare earth elements, of course) collide with the steady retirement of coal-fired baseload plants that once provided round-the-clock reliability.

Rocky Mountain Power is struggling to keep pace, leading to interconnection delays, reliability concerns, and even high-profile project slowdowns such as Google’s postponed Eagle Mountain campus. Coal’s share of Utah’s electricity mix has fallen sharply—from about 75% in 2015 to 45% in 2024—and roughly two-thirds of legacy baseload capacity could be offline within two decades. At the same time, transmission bottlenecks prevent lower-cost power from reaching the Salt Lake region, while prolonged megadrought conditions undermine hydro reliability and increase water-related energy demands.

In response, state leaders have launched “Operation Gigawatt,” (opens in a new tab) an initiative to double generation capacity, expand transmission infrastructure (including projects like Gateway South), and diversify supply with natural gas, solar, geothermal, and potentially small modular nuclear reactors. The strategy aims to restore reliability and fuel economic growth, but it underscores a broader tension between rapid electrification, environmental constraints, and the pace of infrastructure development.

Structural Water Stress

Arizona’s Department of Water Resources projects 4.86 million acre-feet of long-term unmet groundwater demand in the Phoenix Active Management Area over 100 years under current conditions. “Assured supply” debates that once focused on housing now increasingly spill into large industrial siting.

Across the West, at least eight states introduced legislation last year seeking mandatory data-center water reporting. NDAs and proprietary claims are colliding with municipal planning needs.

Transparency is becoming law.

From Infrastructure Boom to Bubble Risk?

Capital expenditures are exploding. Alphabet reported $52.5 billion in capex in 2024. Amazon disclosed $77.7 billion in 2024 cash capex, much tied to technical infrastructure. Microsoft has indicated that cloud and AI-related investments now dominate capital spending.

Waterscarcity converts that capex surge into the potential for stranded assets.

Permitting delays, mandated redesigns, wastewater constraints, and cooling-system shifts toward higher-power dry configurations all erode margin assumptions. In water-stressed basins, water-use models can break down.

When capex outruns physical resource constraints, bubbles form around optimism — not throughput.

The China Contrast

China’s “East Data, West Computing” initiative uses centralized coordination to steer data-center geography based on climate and energy profiles — though some western provinces face water stress of their own.

At the same time, China’s digital economy has emphasized app-mediated profit systems. OECD estimates suggest Ant Group and Tencent dominate China’s mobile payments ecosystem — a reminder that data monetization can scale without proportional infrastructure duplication.

Meanwhile, upstream leverage persists. As the Rare Earth Exchanges community knows well, the U.S. Geological Survey reports roughly 72% of U.S. rare-earth compound and metal imports (2019–2022) originated from China.  And of course, about 80+ of the rare earth refined output comes from China. Export controls on medium and heavy rare earths add additional pressure to electrification and grid buildouts required by AI expansion.

The U.S. is building hardware scale. China is optimizing systems scale.

Risk Outlook and Mitigation

Risk Assessment:

• High local operational risk in water-stressed basins
• Moderate-to-high systemic risk if infrastructure capex continues outpacing permitting, water, and grid constraints

Mitigation Imperatives:

• Mandate auditable reporting of both direct (site) and indirect (source) water footprints
• Condition approvals on reclaimed-water use or closed-loop/dry cooling in high-stress basins
• Add “water + grid” diligence screens to AI infrastructure investment models
• Treat water as a first-order engineering constraint — not a public-relations afterthought

The AI boom is real. But so are aquifer limits.

In the West, silicon may be abundant. Water is not.