Rethinking water in the AI economy: A path to resilient data centers

Much of the conversation around AI infrastructure has focused on how quickly new capacity can be built. While water is considered in data center planning today, less attention has been paid to the readiness and resilience of the systems that need to support this growth over time.

Data centers are expanding rapidly, often in regions already under strain. Water demand is rising alongside that growth, but the real challenge is not just volume. It is timing, location, and whether infrastructure can sustain that demand in practice. That gap is starting to shape what can be built, and where.

Ann Perreault, Vice President, Vertical Strategy, Growth and Innovation at Xylem, sees this playing out in real time, and how water is moving from a secondary consideration to a defining factor in how data center projects are planned and delivered.

How should water shape data center planning in high-growth regions?

Water has traditionally been treated as an operational input in data center design, something to be managed later rather than a constraint that shapes decisions upfront. That approach worked when workloads were more predictable and thermal demands were easier to manage. But the new generation of AI infrastructure works very differently.

Training and operating advanced AI models rely on dense clusters of high-performance chips that generate far more and less predictable heat than traditional workloads. As power densities increase, cooling needs change, along with the increased need for reliable water supply. This is because water remains the most efficient heat transfer medium. As a result, water availability, quality, reuse potential, and regulatory exposure are now directly tied to uptime, scalability, and cost — not just sustainability optics.

This shift is already increasing pressure in key regions. Close to 40% of the world’s data centers are located in areas experiencing high or rising water stress, including the American Southwest, Taiwan, parts of Asia, and other fast-growing data center hubs. As AI workloads scale, demand is rising quickly, especially during peak periods when communities rely on the same systems. These pressures are becoming more visible. In many regions, demand is rising at the same time communities need water most.

Policy and regulation are still catching up. Much of this growth is happening without clear guidance on how water should be sourced and managed, or how the underlying infrastructure should be funded or expanded. This is creating a gap between the pace of development and the readiness of local water infrastructure.

Addressing this means looking beyond the facility itself. Some of the challenge sits inside, but much of it sits in the systems that supply it.

How can data center operators improve water efficiency inside the facility?

Operators have the most immediate control over what happens inside the fence. That is where many of the first changes are happening. Cooling efficiency is improving, and that progress will continue. New technologies are helping reduce water use per unit of compute, and operators are getting better at managing thermal loads. But efficiency alone isn’t enough to keep pace with rising demand.

Many large data centers still rely on evaporative cooling or hybrid cooling systems. These systems work well, but they consume most of the water they withdraw through evaporation, returning only a small portion to the basin or for reuse. As AI workloads scale and operate at much higher heat loads, total water demand continues to rise — even as per-unit cooling efficiency improves. This turns data center growth from an abstract sustainability concern into a direct dependency on local water availability, resilience, and long-term stewardship.

Newer cooling approaches are beginning to change that equation. Hybrid and liquid cooling architectures significantly reduce reliance on evaporation, with many designs operating in closed or semi-closed loops that keep water in circulation rather than consuming it.

Advanced treatment processes — including reverse osmosis, nanofiltration, and electrodeionization — are expanding reuse options while maintaining the water quality required for high-performance cooling systems. At the same time, digital monitoring and controls enable real-time optimization, helping operators balance performance, efficiency, and water use as workloads fluctuate.

In some cases, these approaches can reduce water use at the facility level by as much as 80 to 90%. These are important steps forward, but they don’t address the limits of the systems they depend on. In many regions, water infrastructure is already operating near its limits. Closed-loop systems can reduce on-site water use, but often rely on air cooling, increasing power use and shifting water demand to the power source.

Finally, the aging distribution systems lose significant volumes of treated water each year — often exceeding the incremental demand anticipated from new AI‑driven data centers. This highlights a broader issue: the challenge is not only how facilities are cooled, but whether the surrounding water systems are resilient, efficient, and modernized enough to support what is being built.

How can operators strengthen the water systems they depend on?

Even with improvements inside the facility, the bigger constraint often sits outside it. Long-term performance depends not only on how water is used on-site, but on the resilience of the surrounding infrastructure.

Water networks in many regions already experience significant losses. Addressing this gap is critical to ensuring reliable supply. Operators are increasingly working with utilities and communities to strengthen these systems. Data center developers and hyperscalers that engage local utilities early in the process can co-plan for needs and investments that accrue value to all players in the watershed.

In Mexico, for example, Amazon has partnered with utilities to use digital monitoring, pressure management, and leak detection to reduce water loss across distribution networks. These efforts are expected to save more than one million cubic meters of water each year while improving supply reliability for surrounding communities.

This reflects a broader shift. Reuse, alternative water sources, and closer coordination with utilities are increasingly being built into projects from the start. Early planning helps operators understand local conditions, align with community needs, and design systems that can perform reliably over time.

In some cases, this also means designing facilities to operate at higher temperatures, reducing reliance on water-intensive cooling during peak periods.

As scrutiny increases, operators are expected to demonstrate that their investments support — not strain — local water systems. Where this is not addressed early, projects have faced delays or opposition, making water a critical factor in site selection and long-term viability.

What will define successful data center development moving forward?

The tools to address these challenges already exist. Digital monitoring, advanced treatment, reuse systems, and more efficient cooling approaches are proven and increasingly being adopted. Many hyperscalers have also committed to water-positive or replenishment goals, accelerating this direction.

What’s needed now is earlier action. The companies that succeed will not be those that focus on power alone, but those that consider water from the start and plan for it alongside everything else.

Done well, this approach creates real opportunity. It supports uptime, reduces permitting risk, and builds alignment with communities. It turns water from a hidden constraint into part of how these systems perform over time. 

For all the attention on AI’s hunger for energy, its thirst for water is only just coming into focus, and it will shape where and how far this next phase of growth can go.