Water for People, Progress, and AI

Sabine Blumstein

Principal Consultant

Close

Sabine Blumstein

Contact Sabine Blumstein

Jason Clarke

Director

Close

Jason Clarke

Contact Jason Clarke

Samantha Almonacid

Consultant II

Close

Samantha Almonacid

Contact Samantha Almonacid

Table of Contents

• Data centres caught between people and progress
• What we know, and what we don't
• How much water do data centres need?
• Water for people and progress
• How Anthesis can help

Share this article

• Share via Linkedin
• Share via Facebook
• Share via X
• Share via Email

This month, the global water community converges on Stockholm for World Water Week 2026 under the theme “Water for People and Progress.” The topic captures a tension frequently at the centre of water-based decisions: growth depends on water, but so do households, farms, communities, ecosystems, and the local systems already under pressure to serve them.

While the underlying dynamic of growth-driven site selection colliding with basin-level water limits is not unique to a single industry, the race to scale artificial intelligence (AI) does make tech one of the most visible sectors testing that balance right now. The AI data centre boom promises economic development, digital capacity, and a stake in the future of AI, but it also demands power, land, cooling, and water – often in communities that are already asking how much capacity is left.

In May 2026, for example, commissioners in Box Elder County, Utah, voted unanimously to approve the Stratos Project, a proposed 40,000-acre, 9-gigawatt AI data centre in one of the driest corners of the state.  Within days, residents petitioned for an independent review and a local group filed to force a public vote. Organisers are now fighting in court and the developer has paused the project.

This case illustrates a pattern emerging across the United States and beyond: decisions about water and land-use for AI data centres are often made by a small group of officials while the consequences can spread across entire basins, affecting ranchers, farmers, and towns that never asked for the project in the first place.

While pushback has been loudest in the United States, the same dynamic is emerging wherever new data centres are landing on already-stressed water systems with aging and underfunded water infrastructure. Chile offers a close parallel: a Chilean environmental court halted Google’s Cerrillos data centre outside Santiago in 2024 over its draw on the region’s drought-strained aquifer, and Google is only now reviving a redesigned, air-cooled version of the project.

Data centres caught between people and progress

AI data centre site selection is rarely driven by water availability. Instead, it often prioritises energy cost, land price, tax incentives, and speed of approval – characteristics that often overlap with dry, sparsely populated land. That is why projects can end up in places where water systems are already under pressure, even when water is central to the project’s long-term risk.

The promise to host communities can also be significant. Across the country and around the world, similar projects are often pitched as sources of tax revenue, construction work, permanent jobs, and a foothold in the AI economy. In Utah, for example, Stratos secured approval in exchange for an estimated $30 million in initial county revenue and roughly 2,000 jobs.

Those local benefits sit alongside impacts that are much harder to localise neatly: added demand for power, land, and water; pressure on utilities built for a different scale of use; and environmental effects that extend beyond the community where the project is approved.

That wider footprint of impact helps explain why public reaction is often skeptical, even when the sector’s direct water footprint looks modest compared with agriculture or manufacturing. Residents are not reacting to global averages; they are asking whether one new facility will compete with homes, farms, and ecosystems for the same limited supply, particularly during droughts or peak summer demand.

That concern shows up clearly in public opinion. A Gallup survey found that 70% of Americans don’t want to live near a data centre, with half citing water and resource use as a primary reason. For developers, that opposition is not only reputational: petitions and ballot measures can stall permitting timelines and strand capital already committed to a site, as the Stratos case above shows.

Unfortunately, the water question is difficult to answer with a single number. On site, a specific data centre’s direct water footprint is relatively small next to other water-intensive industries, but the upstream footprint – the water embedded in the electricity that runs data centres – is larger and gets less scrutiny. With the AI-driven growth arriving faster than most utilities plan for, the strain shows up less in national totals and more in local systems sized for a different era of demand. That mismatch is what’s driving much of the scrutiny data centres now face.

The debate, however, suffers an evidence problem: we know enough to see why communities are worried, but don’t have the data needed for precise, localised answers.

What we know, and what we don’t

The conversation around AI’s water footprint runs on a handful of big, contested numbers. Some are useful for understanding the scale of demand. Others can obscure the local timing of water use, infrastructure capacity, and basin-level conditions that determine whether a specific project becomes a water problem.

Data centres are overlapping with water stressed areas

Water stress measures how much of a region’s renewable water supply is already being withdrawn each year. “High stress” indicates that demand is closing in on, or exceeding, what’s naturally available. By that measure, 40% of the world’s data centres already sit in areas of high or extremely high water stress.

To understand what this means in practice, we need more granular, basin-level evidence from the communities and ecosystems around specific sites: who else draws on the same supply, how much stress already exists, and whether the system can absorb new demand without shifting risk onto others.

Timing can be particularly misleading

Water withdrawal at data centres often peak in summer, the same time as farms, utilities, and households. The driver is temperature: air-cooled and hybrid systems only switch over to water-based cooling once outside air gets too hot to reject heat efficiently. That means the hottest weeks of the year are exactly when data centres pull the most water from the same systems everyone else is also leaning on.

What we’re still missing is more granular mapping of specific water systems: when demand peaks, how much spare capacity exists, and whether infrastructure built for one scale of use can absorb a new industrial load without affecting existing users.

Scale, at least in the United States, is not in dispute

About two-thirds of the 809 data centres planned in the United States are set to be built in counties that have been in drought over the past year. Total United States data centre water demand is projected to reach as much as 73 billion gallons a year by 2028, up from about 17 billion in 2023. The largest single facilities can draw up to 5 million gallons a day, roughly the water use of 50,000 people.

Existing infrastructure is not prepared

Infrastructure in many of the regions seeing the highest demand was not built for industrial users of AI scale. Texas is a well-documented case: state law requires utilities to serve any customer who requests it, but there’s no mechanism for a utility to deny or limit a large new water user, or to require advance notice of how much water it will need. For developers, that gap cuts both ways: a project can clear local permitting and still face water supply risk the approval process never tested.

How much water does a data centre need, and how can we minimise impact?

Water demand can vary widely from one data centre to another, and even within the same data centre over time. Key variables include: server load and utilisation, computing density, cooling technology and efficiency, operational efficiency, local climate, facility size and capacity, energy source and grid mix, water source and quality requirements, and local infrastructure constraints. These factors make determining a precise quantity of water estimate difficult without site-level data.

Still, it is imperative that we approach AI growth and associated water use with both ecosystem boundaries and human rights in mind.

Some large data centre operators are already stepping in voluntarily. For instance, Microsoft partnered with the city of Quincy, Washington, to build a dedicated Water Reuse Utility that treats cooling wastewater for reuse. Meanwhile, Amazon has funded leak-detection and network upgrades in Mexico City and Monterrey which is projected to save more than 1.3 million cubic meters of water annually.

Regulation is also starting to catch up, primarily by mandating disclosure.The EU’s Energy Efficiency Directive already requires data centres above 500 kW to report annual water use, and California, Michigan, and Iowa are advancing similar rules. A smaller number of jurisdictions are going further: where community pushback has been loudest, moratoriums start to become politically viable. Texas paused new data centre approvals in August 2026 pending a state audit, following months of public backlash over water and energy use, and New York has floated a similar freeze.

The split between these two tracks, disclosure versus restriction, represents the split in the debate itself: most regulators are trying to understand the scale of the problem before deciding how to act on it.

Water for people and progress

Underneath the headlines, the threat of social licence to operate has two roots. The first is physical competition: AI’s growth is accelerating and its demand is concentrated in the same peak weeks when surrounding communities need the same water. The second is transparency: most operators still do not disclose what they use on site, what is embedded in their upstream supply chain, or where they are investing to offset impact.

Closing both gaps takes assessment at more than one scale: catchment-level data to understand exposure to drought and water stress, and site- or network-level data to understand whether the specific utility or supply system a facility draws on can absorb that demand without affecting other users.

Right now, most of what gets measured and disclosed sits at the aggregated company or industry level. That scale can’t answer the questions that determine local impact: how much competition already exists for water in a given catchment, how much spare capacity a specific network has at peak demand, and who else depends on the same supply.

“Water for People and Progress” is exactly the balance regulators, communities, and the companies building this infrastructure now have to strike. AI is neither inherently good nor bad for water security. Its outcomes depend on governance, infrastructure investment, transparency, and who bears the cost of the tradeoffs.

How Anthesis can help

As water demand accelerates across every capital-intensive sector, most companies do not yet know where their own exposure sits. Anthesis helps organisations build that picture: quantifying direct operational water use alongside embedded water in energy, supply chains, and surrounding infrastructure, then mapping that demand against basin-level stress, climate exposure, ecosystem needs, and community dependence on the same water systems. That mapping is what separates a relatively small enterprise-wide footprint from a site-specific impact that may be material in a stressed catchment.

The same exposure runs well beyond data centres: semiconductor fabrication, beverage and food manufacturing, mining and battery materials, and textile finishing all share the pattern of site selection driven by energy cost and land price landing operations in already-stressed basins.

We also support clients in identifying opportunities to reduce water-related risk at both the site and value-chain level. Our water stewardship approach brings together operational performance improvement, risk and opportunity assessment, benchmarking, target setting, stakeholder engagement, governance, reporting and disclosure, and value chain performance improvement.

Additionally, we walk clients through credible third-party benchmarking processes, including the Alliance for Water Stewardship certification process.

Data centres are the most visible face of a pattern that extends well beyond the sector. Water and the natural systems it depends on are moving from a reporting obligation to a material financial risk for any company with capital-intensive operations or long supply chains, whether that exposure sits in a factory, a farm, a mine, or a server hall. Anthesis helps clients get ahead of that shift: identifying where nature-related risk actually sits, quantifying its financial materiality, and building water and nature stewardship into resilient enterprises and value chains, turning disclosure from a compliance cost into a source of competitive advantage.