By now, most people understand AI requires enormous computing power and energy. U.S. data centers consumed about 183 TWh of electricity in 2024. That’s more than 4% of total U.S. electricity use in 2024, and demand is projected to more than double by 2030.
Although the conversation around AI’s water usage footprint has started, it’s focused on the wrong part of the problem.
Water is essential for cooling data centers and keeping servers operational. Water moves through pipes next to the equipment to absorb that heat, then carries it away to cooling systems so the servers don’t overheat and shut down. Water is often the most efficient way to cool infrastructure. Some facilities consume millions of gallons daily, and AI expansion is accelerating this demand dramatically.
But that is not the only issue. Electricity generation and chip manufacturing also use an extensive amount of water. Yet, few are focused on that issue.
Even as major chipmakers announce initiatives to reduce data center water consumption, research shows that water used outside the data center in electricity generation and chip manufacturing can double or triple a facility’s total water footprint. That means the solutions getting the most attention may only address one-quarter to one-third of AI’s actual water impact.
Most leaders aren’t planning for the rest of the supply chain at a moment when the world is already experiencing a broader water crisis and water systems are under growing strain.
The reason the water footprint is so much larger than most leaders realize comes down to what happens outside the facility walls. Cooling towers and on-site water systems get the most attention, but the water embedded in electricity generation and semiconductor manufacturing tells a different story.
Thermoelectric power plants, which generate much of the electricity powering data centers, withdraw and consume significant volumes of water in the cooling process. Chip fabrication is similarly intensive: producing a single advanced processor can require thousands of gallons of ultrapure water. When those upstream demands are factored in, research shows the total water footprint of an AI facility can be two to three times higher than what’s visible on-site. A data center that looks water-efficient by conventional metrics may still be driving substantial water consumption somewhere else in the supply chain.
The scale of what’s coming is already measurable. Researchers estimate U.S. data centers could require an additional 697–1,451 million gallons of water capacity per day by 2030 if current water intensity trends continue. This is almost equal to the amount of water New York City uses in an average day.
This points to a larger debate over whether communities are carrying the environmental and resource costs of AI growth while companies capture most of the benefits. AI’s hidden infrastructure challenge is not energy alone. Water is already an equally important constraint on growth.
What makes this challenge unique is that most consumers, and even many business leaders, never see this side of the AI boom. AI feels digital and invisible, but behind every prompt is a physical network of buildings, energy systems, and water infrastructure. Until recently, these tradeoffs were largely discussed inside infrastructure, utility, and industrial circles. That is beginning to change as AI scales and communities start asking what resources support that growth.
This becomes a business issue because water availability may increasingly shape where data centers can be built. Drought-prone regions are already beginning to confront those tradeoffs. As negotiations continue over declining Colorado River supplies and states across the Southwest prepare for a future with less reliable water access, communities are becoming more sensitive to large industrial users entering the market. In Utah, proposed data center development has already sparked public pushback tied to concerns around utility strain, water use, aquifer pressure, and broader quality-of-life impacts. The result is that AI’s next major infrastructure constraint may not simply be electricity generation; it may be access to sustainable, locally supported water infrastructure.
Most U.S. water infrastructure was not designed for hyperscale data centers, AI, and today’s industrial demand. Many systems are decades old and cannot handle the collusion of population growth, AI growth, and climate pressure. Utilities and municipalities are being forced into reactive instead of proactive planning, and infrastructure conversations are lagging behind technological acceleration.
At the same time, cities and states across the Southwest continue searching for ways to stretch existing supplies as pressure grows on the Colorado River system, which is a reminder that securing more water during periods of scarcity is becoming increasingly difficult. Rather than assuming additional supply will always be available, I believe data centers will increasingly need to think about water the way they already think about energy: as a strategic resource to manage, reuse, and optimize.
That shift may accelerate approaches like closed-loop systems, where water is continuously reused within operations, and decentralized treatment models that reduce reliance on centralized municipal infrastructure. Operators that begin planning for long-term water resilience now may gain advantages in permitting, cost predictability, and scalability later. To me, that’s the more important signal: communities and operators that act proactively instead of waiting for shortages may ultimately be better positioned for long-term growth.
Water strategy is becoming infrastructure strategy.
AI will absolutely continue to transform industries and economies at higher levels than we’ve already seen. But physical infrastructure still matters, and water is becoming a hidden but foundational issue beneath the AI boom. The smartest companies won’t wait for water constraints to become a crisis before adapting. The companies that lead the next era of AI may not simply be the ones with the most computing power, but the ones that understand the infrastructure realities supporting it.
Kevin Gast* is co-founder, chairman, and CEO of VVater.*