Water consumption in AI data centres is frequently misunderstood, say Nick Remington of PolarDC and Simon Wyatt of Cundall. With the right cooling strategies, site selection and energy sourcing, the UK can expand its digital infrastructure without a corresponding rise in water use

Simon Wyatt Nick Remington

Simon Wyatt is sustainability partner at Cundall (left), Nick Remington is technical director at PolarDC and

Data centre demand is at an all-time high, driven by artificial intelligence. This growth has also increased scrutiny of water consumption, with concerns that AI could further strain water resources. In the UK, however, the relationship between AI and water use is often misunderstood.

AI itself does not drive water consumption, but decisions about cooling systems, site location and energy supply do. Data centres can support significantly greater computing capacity while consuming little, and in some cases virtually no, operational water.

Water use is a design choice, not a function of AI

Historically, many facilities relied on evaporative or adiabatic cooling systems. These are highly energy efficient; but consume water and, as such, have attracted attention in regions with limited water.

AI has accelerated the shift to direct-to-chip liquid cooling and other systems that can manage higher rack densities. Liquid cooling doesn’t necessarily consume water, giving it an advantage over other systems. Instead, they operate in closed loops, transferring heat to external infrastructure like dry coolers without evaporating water.

When combined with closed-loop dry cooling solutions, operational water demand can be reduced dramatically and, in some cases, virtually eliminated

When combined with closed-loop dry cooling solutions, operational water demand can be reduced dramatically and, in some cases, virtually eliminated. PolarDC’s AI-ready facilities, for instance, have been designed around this principle, supporting high-performance computing infrastructure whilst minimising operational water consumption.

Looking beyond the data centre fence

Assessing water consumption solely at facility level provides an incomplete picture of environmental performance. Electricity generation itself can be a significant consumer of water, particularly where fossil-fuelled and nuclear power stations rely on cooling processes. By contrast, renewable technologies such as wind and solar have very low operational water requirements.

As the UK electricity system continues to decarbonise, the indirect water footprint associated with data centre operations is also falling. Consequently, two facilities with identical IT loads can have materially different overall water impacts depending on both their cooling strategy and the source of electricity supplying them.

Why public perception and reality often differ

Much of the concern surrounding data centre water consumption is driven by examples from hot, water-stressed regions, particularly parts of the United States. In these locations, operators have historically favoured evaporative or adiabatic cooling systems to minimise energy consumption. At the same time, around 76% of electricity generation is supplied by fossil fuel and nuclear power stations, both of which can have significant cooling water requirements.

As a result, some large hyperscale campuses consume substantial volumes of water. Global data centre water consumption is estimated at approximately 500 to 700 million gallons per day, with evaporative cooling systems typically using between 2 and 4 litres of water per kWh of IT load. At peak demand, individual hyperscale campuses can consume several million gallons of water per day.

The UK presents a very different picture. Facilities operating in cooler climates, particularly in Scotland, can make extensive use of dry cooling, free cooling and heat recovery. These approaches can reduce water consumption dramatically, achieving Water Usage Effectiveness (WUE) values as low as 0.1 litres per kWh and, in some cases, near-zero operational water use.

How AI is changing the cooling equation

For many years, the industry focused heavily on reducing power usage effectiveness (PUE), often creating a trade-off between energy efficiency and water efficiency.

Direct-to-chip liquid cooling removes heat far more effectively than traditional air-based systems, allowing facilities to operate at higher temperatures and maximise the use of dry cooling and free-cooling technologies. As a result, operators can increasingly achieve both low energy consumption and low water consumption simultaneously.

A uniquely British opportunity

While parts of South-east England are experiencing increasing pressure on water resources, the UK’s temperate climate provides significant opportunities to utilise dry cooling and free-cooling technologies for much of the year. This reduces reliance on water-intensive cooling systems and supports lower operational water consumption.

Scotland is well-positioned. Its cooler climate and excess of renewable energy have created favourable conditions comparable to the Nordic markets. The main constraint is the proximity to demand and the latency requirements of certain applications.

What should we be measuring?

As AI continues to drive investment in UK digital infrastructure, both energy and water performance must become central measures of success.

Whilst PUE remains an important indicator, Water Usage Effectiveness (WUE) is increasingly critical. A facility achieving a low PUE through significant water consumption may not represent the most sustainable outcome, particularly in regions facing water stress. Equally, a facility designed around closed-loop cooling systems can achieve exceptionally low water consumption whilst maintaining excellent operational efficiency.

Local authorities, planners and policymakers should therefore place greater emphasis on operational results than on the type of technology used. Setting minimum requirements for PUE and WUE performance, together with continuous operational reporting and verification, would provide a much stronger foundation for evaluating environmental performance.

The future of AI data centres in the UK

AI growth does not have to increase water consumption. The UK’s climate, evolving cooling technologies and decarbonising grid create an opportunity to increase computing capacity without a corresponding rise in water use.

The key question is no longer how much compute a data centre contains. The more important question is how effectively it is cooled, powered and operated throughout its life.

Nick Remington is technical director at PolarDC and Simon Wyatt is sustainability partner at Cundall