Endless Thirst: Addressing Data Centers' Soaring Water Demands at the Local Level
AI is thirsty – what should local governments know about the water demands of data centers, and how might public agencies promote more sustainable practices?
By Lindsey Washburn, Leila Doty, & Eleonor Bounds
Last month in our blog series on “AI & the Environment,” we looked at AI’s intense energy demands. This month we focus on a lesser discussed, but equally critical aspect of AI’s impact on the environment: water.
As always, our aim is to spark conversation and action from those in local government so that together, we can work towards shaping a more sustainable future for AI where the benefits outweigh the costs.
Water: Essential to Life (& Data Centers)
What does water have to do with AI anyway? Well, a lot actually. The chips used to power AI, generative AI in particular, get extremely hot. So hot that they can start melting if not properly cooled. These chips sit on server racks and are stored in data centers. Massive water-based cooling systems are used to cool the data centers and operate 24/7. These systems require up to millions of gallons of water at any given time, both directly for cooling and indirectly depending on how the data center is powered. The water used on site to cool both the servers and the data center itself are considered direct (or scope-1) usage.
As we learned last month, AI is an energy hog, largely due to the energy demand of data centers. While some data centers utilize clean energy through renewable sources like solar and wind power, the rest run primarily on electricity generated by fossil fuels. Large amounts of water are also needed to generate electricity with fossil fuels (and nuclear, by the way), further adding to the water footprint of a data center through indirect (or scope-2) usage. To get a complete picture of AI’s impact on water, we must consider water usage in the manufacturing process for AI chips and servers (or scope-3), as this is also significant.
We highly recommend checking out the research paper, “Making AI Less “Thirsty”: Uncovering and Addressing the Secret Water Footprint of AI Models,” for a deeper dive into how AI uses water. For the purposes of this piece, we’ll focus on the direct water usage of data centers through the cooling system. Below is a simplified illustration of how this works from the U.S. Department of Energy:
Figure 1: U.S. Department of Energy Schematic of Typical Data Center Evaporative Cooling System
To put this in perspective for the individual user, researchers found that a 100-word email generated by GPT-4 was equivalent to a 500ml bottle of water.
Growing Demand
While data centers have been around for several decades, the rate at which companies are planning and building new ones is unprecedented. The skyrocketing demand for more data centers is largely due to generative AI as it requires a massive amount of computing power (compute) achieved through the chips called graphics processing units (GPUs). By 2027, global AI demand is expected to account for 1.1 to 1.7 trillion gallons of water withdrawal, about half the total annual water withdrawal of the UK.
With the U.S. home to nearly 40% of the world’s data centers, counties, municipalities, and special districts are going to experience significant challenges balancing AI’s water demand with their constituents’ needs. Let’s not forget that we also need water to grow our food, fight fires, and produce almost everything else we use in daily life.
During Climate Week in Washington D.C. this past April 2025, policy experts discussed the sustainability of AI. “We’re now at a stage where AI and data centers that power it are competing directly with humans for land and water and energy,” said Jonathan Gilmour, a data scientist for Harvard’s school of public health. AI’s environmental impact remains an ongoing challenge, despite the increase in use cases leveraging AI for climate challenges.
Location, Location, Location
Dry and sunny land is great if you’re maximizing for renewable energy like solar - plus it’s cheap! Unfortunately, dry and sunny land is not ideal when you need millions of gallons of water to keep your data center cool. Yet, this is the situation data center operators often find themselves in because water (yes, essential-to-life water) has been a systematic afterthought.
Data centers are disproportionately built in already water-stressed areas. Two thirds of data centers that have been built or planned since 2022 are in places with scarce water resources (think Arizona, California, and Texas). Some communities are starting to push back via protest and legal action as we’ve seen in Arizona, Oregon, South Carolina, and globally, including Uruguay. We’ve also seen a concerning trend in data centers operating in unincorporated regions where local advocacy tends to be weaker.
To learn more, we interviewed Masheika Allgood, a lawyer-turned-technologist who founded AllAI Consulting to address the AI knowledge gap in non-techie industries and professions. She’s an AI ethicist who specializes in generative AI’s environmental impact. When talking to Allgood about her work on data centers, she said, “I have yet to see a water district partner with a data center during the development process to ensure sustainability. I think that’s where the gap is, water is an afterthought, but water is just as important to a data center as electricity.”
Places with lower ambient temperatures reduce cooling requirements and lower cooling requirements reduce both direct and indirect water consumption. Achieving sustainable AI requires operating data centers in locations that can support renewable energy and meet the water demand.
Lack of Transparency
Another concern with the current state of data centers is the lack of transparency around water usage. It’s challenging to get an accurate picture of a data center’s water demand because typically very little data is publicly available for research and analysis. According to one study, less than a third of data center operators even measure their water consumption. And why should they when there is little to no incentive to use water efficiently? Data centers pay the same rate for water regardless of how much they consume and without penalty for overconsumption.
Concerningly, some jurisdictions are signing Non-Disclosure Agreements (NDAs) with data center operators, further obscuring the real water demands and impact to the community. “It’s hard to have a meaningful conversation without the numbers,” said Allgood. To address this gap, Allgood created the Data Center Water Consumption Calculator.
Figure 2: Data Center Water Consumption Calculator
To use the calculator, you’ll need to know the megawatt-hours (MWh) used by the data center, which is the amount of electricity either used or estimated on an annual basis, and then enter that number and select your unit of measurement. The calculator was designed to empower municipal leaders in the contract negotiation process with data centers. “Our municipal leaders need to have some understanding of how much water the data center will require before approving the build - during the permitting and zoning process,” Allgood states. For the math nerds out there, check out Allgood’s blog post about how she came up with the calculations used.
What Local Governments Can Do
In many ways, water is a local issue. This makes municipal governments uniquely positioned to negotiate with data center operators in a way that serves the public interest.
Utilize the Data Center Water Consumption Calculator: The estimates from the calculator help validate the data center operator’s claims about water usage. A few questions Masheika suggests to ask data center operators during negotiations:
How do the assumptions in your estimate differ from those in the tool?
What is your average facility-level water consumption percentage?
On average, how many hours in a year do your facilities run at peak usage?
What is your local water supply chain and what percentage of the estimated usage can it meet today?
How much of the estimated consumption do your replenishment efforts account for, and in what timeframe?
Check out the GovAI Coalition’s AI FactSheet: Specifically, check out the section for vendors to disclose the environmental impacts of their AI product. Agencies should implement the AI FactSheet and require prospective vendors to share “energy and water consumption for training and/or per use of the model, sources of energy and water for [their] solution, and [their] organization’s approach to reducing environmental impact” during the procurement process. If the vendor shares unsatisfactory numbers or even outright refuses to disclose this information, agencies should refrain from contracting with them, to the extent possible.
Increase transparency: Require greater visibility into energy and water agreements with data center operators. Public engagement around the impact of data centers on communities has been minimal at best. Involving local communities early on can ensure that their concerns are addressed and that the benefits of data center operations are shared equitably. Requiring regular reporting on water usage will also increase transparency through the operation lifecycle of the data center.
Set standards for water usage: Setting limits or efficiency standards for water use in data centers can drive innovation and encourage the adoption of less water-intensive cooling technologies.
Create incentives for sustainable practices: Leverage incentives like tax breaks or expedited permitting for data centers that implement water recycling, rainwater harvesting, or closed-loop cooling systems that significantly reduce freshwater consumption.
As Masheika Allgood put simply, “Water cannot continue to be an afterthought.”
Thanks for reading. See you next month for our discussion on the public health impacts of data centers.
About the authors
Lindsey Washburn is the Program Manager for Strategic Initiatives at the City of Tigard, Oregon.
Leila Doty is a Privacy and AI Analyst at the City of San José.
Eleonor Bounds is a Senior Analyst in Data Privacy and Responsible AI at the City of Seattle.
Note: The opinions expressed in these articles are solely those of the author(s) and do not necessarily reflect the views, positions, or policies of the GovAI Coalition or the authors’ affiliated professional organization(s).
—
This is the blog for the GovAI Coalition, a coalition of local, state, and federal agencies united in their mission to promote responsible and purposeful AI in the public sector.
