How AI is changing water use in data centres?

Do AI data centres consume large amounts of water? Increasingly, the answer is no. Water consumption is driven less by computing demand itself and more through factors such as cooling strategy, local climate, overall facility design, and the carbon intensity and generation mix of the electricity supplying the data centre. Closed-loop cooling is the preferred approach for AI infrastructure, enabling operators to significantly reduce operational water consumption while creating additional value through waste-heat recovery.  

Water use has become one of the most debated aspects of global data centre growth. Headlines are regularly suggesting that hyperscale facilities consume as much water as entire cities. Meanwhile, local communities are growing more aware of AI and data centres in general and are increasingly questioning whether digital infrastructure can exist alongside long-term water security goals.  

In reality, the issue is much more subtle.  

Water consumption is not an inherent characteristic of a data centre. It depends mainly on the chosen cooling strategy, local climate, and how the facility interacts with wider energy systems.

In the past, a great many data centres have used evaporative cooling systems, which involve the use of water in order to improve cooling efficiency. Nevertheless, the fast growth of AI is causing more and more people to adopt advanced liquid cooling solutions. Such systems are capable of handling much higher heat loads and do so by using closed-loop arrangements that greatly reduce, and in some instances completely eliminate, the amount of water needed for operation.

Understanding a data centre's true water footprint also requires looking beyond the facility itself. Electricity generation also carries its own water demand, which varies considerably by energy source. Conventional thermal power stations, including coal, gas and nuclear, can consume substantial quantities of water through their cooling operations. In contrast, wind and solar generation have negligible operational water requirements.

As a result, a data centre's overall water impact relies not only on how it is cooled, but also on how its electricity is generated. Norway provides a convincing example. With around 89% of its electricity supplied by hydropower, data centre developer PolarDC's facilities benefit from both a low-carbon grid and a comparatively low water footprint. Assessing water consumption solely at the facility level can therefore misrepresent overall impact.

AI is also driving a fundamental shift in where and how data centres are deployed. Regions such as Northern Europe are particularly well positioned to host the next generation of AI infrastructure, benefiting from cooler climates, abundant renewable energy resources and highly decarbonised electricity grids. Many locations also have established district heating networks, creating opportunities to recover and reuse waste heat from data centres to support local communities and businesses.

As operators adopt water-efficient cooling technologies and increasingly deploy infrastructure within regions having abundant low-carbon electricity, the historic link between data centre growth and water consumption is weakening. The key question is no longer how much compute a data centre contains, but how effectively it is cooled, powered and integrated into the wider energy system.

Why the perceptions around water use can be misleading

A relatively small number of developments often drives public concern about data centre water consumption. These are typically situated in hot, water-short regions, such as parts of the United States (US). The grid generation mix is approximately 58% fossil fuels, 18% nuclear, and 24% renewables.

In these locations, operators typically prioritise reduced power consumption and energy efficiency, so they choose evaporative and adiabatic cooling systems. While highly effective from an energy standpoint, these systems consume water through evaporation and remain the largest source of operational water demand in many facilities.  

Global direct water usage by data centres is generally stated to be about 500 to 700 million gallons each day since evaporative cooling systems usually function at a rate of about two to four litres per kWh, this varying according to the local climate and the general design of the facility. However, when operating at full capacity, individual hyperscale campuses alone can use millions of gallons of water each day.

But by comparison, a facility in Norway using free cooling, waste heat recovery, and non-evaporative cooling might consume just 0.05 million gallons per day. Their water usage effectiveness (WUE) values could approach only 0.1 litres per kWh. This is effectively near-zero water use.  

Water consumption is therefore less a function of compute demand and more a function of how heat is rejected from the data centre and location. Evaporative cooling essentially trades water for lower energy consumption, while dry cooling and air-based systems reduce water demand but rely more on electricity.  

How AI is changing the relationship between cooling and water  

AI workloads are accelerating a shift that was already underway in the industry. Training and inference clusters now exceed +200kW per rack. This is pushing conventional air cooling far beyond typical operating envelopes and requiring the adoption of liquid cooling technologies.  

Importantly, despite its name, liquid cooling does not intrinsically consume water.  

Technologies such as direct-to-chip and immersion cooling operate using closed-loop systems that circulate coolant to remove heat directly at the source. Water consumption occurs at the heat-rejection stage, whether using evaporative, hybrid, or dry cooling solutions.  

Liquid cooling allows heat removal more efficiently and at higher temperatures, enabling greater use of dry coolers, free cooling, and fully closed-loop heat rejection systems. In many use cases, these can drastically reduce or nearly eliminate operational water demand.  

The industry is already moving in this direction. Both hyperscale operators like Microsoft and developers like PolarDC have introduced data centre designs that eliminate evaporative cooling water through a closed-loop, chip-level cooling system. This avoids more than 125 million litres of water consumption per facility each year.  

Looking beyond power usage effectiveness  

For many years, power usage effectiveness (PUE) has been the primary metric for data centre efficiency. To improve PUE, facilities needed to reduce cooling energy consumption, even if this increased water use elsewhere. As a result, water usage effectiveness (WUE) has become more widely used to make the trade-off more visible.  

Evaporative cooling systems can improve energy performance, but they require significant water use. Dry cooling technologies can minimise water use, though they typically come with an energy penalty. However, the rise in AI infrastructure is changing that relationship.  

Direct-to-chip liquid cooling removes heat much more efficiently than traditional air cooling approaches. This means the coolant can operate at higher temperatures, reducing the heat released into the data hall. Overall, the thermal efficiency improves, providing more opportunities for free cooling and dry heat rejection than evaporative systems can.  

Why waste heat may be the bigger opportunity  

The largest opportunity may sit beyond cooling altogether.  

AI infrastructure generates huge quantities of relatively waste heat. In isolation, this heat is a by-product. However, when connected to district heating infrastructure, it becomes a valuable energy resource.  

Across Denmark, Sweden, Finland, Germany, the Netherlands, and increasingly the UK, district heating networks are already designed to absorb low-carbon heat from multiple sources.  

In these environments, data centres can contribute to local energy systems rather than remain isolated energy consumers. They also benefit from lower water consumption. Combined with lower ambient temperatures and greater availability of free cooling, this creates a fundamentally different operating profile than facilities in hotter, water-constrained regions.  

This, combined with chiller ambient temperatures and greater availability of free cooling, creates a significantly different operating profile from that of the hotter, water-constrained regions in the US.  

Why Northern Europe is well positioned for AI infrastructure  

Northern Europe combines several advantages for next gen AI infrastructure:  

• Cool, stable climates that maximise free-cooling opportunities.  
• Lower levels of water stress than many competing markets.  
• High levels of renewable energy generation reduce indirect water consumption associated with power production.  
• Established district heating networks capable of utilising waste heat at scale.  
• Regulatory systems that increasingly support low-water, high-efficiency design approaches.  

Together, these conditions generate the potential for AI-scale facilities to operate with near-zero water consumption while delivering wider energy system benefits. 

To learn more about designing low-water, high-performance AI infrastructure, speak to a member of our data centres or sustainability team.