The Right Cooling for the Right Site: Rethinking Data Centre Water Use

AI is changing the data-centre sector at extraordinary speed. Facilities are becoming larger, rack densities are increasing and the power required to support new generations of compute is rising rapidly.

Yet as attention focuses on securing grid connections, land, and the infrastructure required to deliver this growth, another constraint is moving steadily up the agenda: water.

Water is not simply an operational input to be addressed once a site has been chosen. Decisions around water availability, cooling, treatment, discharge and resilience can influence whether a site is appropriate in the first place, how efficiently it can operate, and what infrastructure will be required to support it.

This was the central theme of the final session of The Foresight Water Webinar Series, supported by Evides Industriewater and Arup. Bringing together Catherine Buckley of Arup, Vincent Toussaint of Evides Industriewater, and Michael Pollock and Oliver Thomas of Integrated Environmental Solutions (IES), the webinar examined how the industry can make better decisions around water, cooling, energy and site selection.

What emerged was a clear conclusion: the objective should not simply be to build data centres that use no water. It should be to identify the right combination of water, cooling, energy and infrastructure for the right site. 

Water From Day One

Power remains one of the defining constraints on data-centre development. As Oliver Thomas explained during the discussion, developers understandably begin by asking whether a site can access the hundreds of megawatts increasingly required by hyperscale facilities.

“Power typically dominates,” he observed.

But a power-ready site is not automatically a water-ready site. Catherine Buckley argued that water needs to enter the development process at the very beginning. Once a site has been selected, developers may already have lost options around alternative water sources, reuse, storage, treatment infrastructure and co-location with other industries.

This becomes particularly important as hyperscale developments move towards larger and sometimes more rural sites. In these locations, connecting to water and wastewater infrastructure can become a major infrastructure project in its own right, bringing additional consenting, construction, environmental and land requirements.

The issue is therefore not simply whether sufficient water can ultimately be delivered. It is whether water has been considered early enough to influence the fundamental development strategy.

Vincent Toussaint reinforced this from the water-treatment perspective. Before determining the appropriate solution, developers need to understand what sources are locally available, how much water will be required at different points in the year, the quality required by the cooling system, what storage may be needed and what options exist for discharge or reuse.

The earlier those questions are asked, the wider the range of solutions available.

There Is No Single Best Cooling Strategy

A second theme emerged repeatedly throughout the discussion: the data centre with the lowest water consumption is not automatically the most sustainable data centre.

Cooling decisions sit at the centre of a complex relationship between water consumption, energy demand, carbon, resilience, climate, cost and available space.

A fully dry cooling system may dramatically reduce operational water requirements, for example, but can increase electrical demand. Conversely, evaporative cooling can offer substantial energy-efficiency benefits while increasing water consumption. The balance can change completely depending on location.

Buckley described a hyperscale project where substantial initial water requirements presented a significant challenge. Once the availability of green electricity was confirmed, the project was able to pivot towards a fully dry cooling solution.

The significance of the example was not that dry cooling is inherently preferable. It was that the optimum solution emerged from the specific energy, water and site conditions of that project.

“There is no single best cooling strategy,” she told the session.

Applying the same cooling template or sustainability target across Northern Europe, Southern Europe, Asia and other markets therefore risks missing the wider resource implications of the decision.

Looking Beyond WUE

This raises an important question around how data-centre sustainability is measured.

Water Usage Effectiveness, or WUE, is becoming an increasingly important metric alongside Power Usage Effectiveness (PUE) and Carbon Usage Effectiveness (CUE).

However, the panel cautioned against considering any of these measures in isolation.

Michael Pollock highlighted an important limitation of annual WUE figures: they can tell developers how much water a facility consumes relative to its IT load, but they do not necessarily reveal when that water is required.

A facility using evaporative cooling might consume comparatively modest quantities across the year as a whole, while concentrating much of that demand into the hottest periods. Those periods can coincide precisely with drought conditions and the greatest pressure on local water supplies.

Vincent Toussaint added that alternative sources have their own availability profiles. Treated wastewater, for example, may vary seasonally too. Sustainable design therefore requires matching not just total annual demand with total annual supply, but understanding when each is available.

This moves the industry towards a more sophisticated question. Rather than simply asking what is our WUE?, developers increasingly need to ask: what is our water demand, when does it occur, where does that water come from, and what happens elsewhere in the system when we reduce it? 

Modelling Before the Options Disappear

Dynamic simulation can play an important role in answering those questions.

Pollock explained that detailed modelling does not need to wait until a project has reached detailed design. Some of its greatest value can come much earlier, while developers still have flexibility over the site, cooling technology and system architecture.

At feasibility stage, the objective is not necessarily to predict the finished facility's performance with absolute precision. Instead, modelling can help determine which options appear viable and which do not, comparing factors including IT loads, cooling technologies, climate, energy use, water demand and resilience before major capital decisions have been locked in.

As the design develops, those models can become progressively more detailed.

This approach is increasingly important as AI accelerates the adoption of higher-density systems and technologies such as direct-to-chip liquid cooling.

Liquid cooling itself does not necessarily mean significant operational water consumption. Closed-loop systems can circulate cooling fluid with relatively limited losses. But, as Toussaint stressed during the audience discussion, the heat still has to go somewhere.

It must ultimately be rejected through air cooling, evaporative cooling, chillers or another form of heat-rejection infrastructure.

The question therefore cannot stop at the server rack. The performance of the complete system needs to be considered.

From the Data Centre to the Catchment

Perhaps the most significant shift discussed during the webinar was the need to move beyond the physical boundary of the data centre entirely. A facility can be extremely efficient when measured internally while still creating significant pressure on the system around it.

A genuine catchment-led approach means understanding where water originates, how reliable that source will be throughout the year, who else depends upon it, what environmental requirements need to be maintained and what happens to water after it leaves the site. It also means planning for future demand rather than assessing the development against today's conditions alone.

Buckley highlighted the importance of considering both upstream and downstream impacts. Existing communities, ecosystems, agriculture and industry may already depend upon the same resources. Meanwhile, wastewater capacity and discharge requirements can prove every bit as important as securing the original supply.

This creates opportunities as well as constraints.

A data centre located alongside existing industry or wastewater infrastructure may be able to use treated effluent, share infrastructure or potentially provide water streams that can be reused by another nearby user.

But achieving these outcomes requires coordination that extends well beyond the developer itself.

Buckley noted that an individual project may recognise an opportunity for another future industry to reuse its cooling water, for example, while lacking the governance or institutional mechanisms needed to make that arrangement happen.

The challenge therefore becomes one of planning and coordination as much as engineering.

Reuse Is an Opportunity, Not a Universal Answer

Reducing dependence on potable water is an increasingly important objective for the sector. The panel discussed a wide range of possible sources, including treated wastewater, freshwater, brackish water and seawater.

However, switching source does not remove the engineering challenge. It changes it.

Different sources can require different levels of treatment and introduce new considerations around water chemistry, filtration, discharge and permitting.

Toussaint highlighted reverse osmosis as one example: treating a source in this way also produces a concentrated waste stream that subsequently needs to be managed and discharged.

Water quality regulation therefore affects both sides of the system: developers need to consider not simply what can be abstracted, but what can ultimately be returned to the environment and under what conditions.

The implication is important: Reuse should be evaluated as part of the whole system rather than pursued as an isolated sustainability target.

Where reclaimed or alternative water can reduce pressure on potable resources while maintaining efficient and resilient operation, it can offer substantial benefits. But treatment energy, infrastructure, environmental impacts and local conditions all need to form part of that calculation.

Water as Strategic Infrastructure

The audience discussion ultimately returned to a deceptively simple question: if water is scarce, why not eliminate it from cooling altogether?

The panel's response captured the wider conclusion of the webinar.

Water is an exceptionally effective resource for transferring and rejecting heat. Removing it completely can move the environmental burden elsewhere through additional power demand, greater equipment footprints or other infrastructure requirements.

As Oliver Thomas argued, the direction of travel should not automatically be “no water”, but better and more environmentally conscious water use.

Buckley reached a similar conclusion: what matters is finding the solution that is appropriate for the location.

This is why the relationship between water and data-centre development needs to be reframed.

Water should neither be treated simply as an unlimited utility nor solely as a constraint to be minimised. It is critical enabling infrastructure, interacting directly with energy, climate, technology, planning and long-term resilience.

Conclusion: Water Should Not Be a Downstream Constraint

Across the Foresight Water Webinar Series - spanning industrial development, advanced nuclear, sustainable aviation fuel and data centres - one message has consistently returned: Infrastructure projects work better when water is considered early.

For data centres, that principle is becoming particularly urgent. AI is accelerating the scale and density of development at the same time that water resources, energy systems and planning infrastructure face growing pressure.

The answer will not be a universal cooling technology, a single efficiency metric or the elimination of water from every facility. It will require developers to understand local conditions, model the interaction between water and energy, consider the wider catchment and preserve as many options as possible before committing to the site.

As Michael Pollock summarised in his closing remarks:

“Water shouldn’t be a downstream constraint.”

It should form part of the planning process from the beginning.

For a sector increasingly accustomed to treating access to power as a strategic determinant of where development can occur, the next step is clear: water deserves the same attention.