As the UK seeks to compete on the global AI stage, the commensurate demand for infrastructure has brought concerns over data centre sustainability and water usage to the fore.
Scotland is famous for its abundance of water, with its iconic wet weather all year round, boggy grasslands, and plethora of lochs. But even we’re feeling the strain, with increasingly less rainfall equating to less available freshwater due to climate change.
Seventeen catchment areas in Scotland faced significant water scarcity in September this year, particularly on the nation’s east coast, as rivers are running lower each year. Further, five regions in England are in drought as of October – water scarcity is quickly becoming a UK-wide problem.
This comes at a time when data centre infrastructure investment is also on the rise, with the UK Government striking numerous deals with major tech giants through the UK/US Tech Prosperity Deal, bringing £31bn in AI investment. But all this comes with a potentially grave environmental cost.
The stats of how much water and energy AI uses are sobering – even small figures quickly jump to terrifying when scaled with the amount of users asking AI chatbots to draft emails or answer queries.
Looking at global AI water consumption, by 2027 AI could guzzle up about 6.6 billion cubic metres of water, which is equivalent to almost two thirds of the annual consumption of England.
But just how much water are data centres really using? Are they using the same supply that could go to safe drinking water, to water crops and livestock, or support a healthy environment?
In the UK, the answer is not a simple one.
To try and understand the connection between the UK’s water infrastructure and data centres, DIGIT spoke with Stephen Slessor, CEO of water technology company RSE and chair of British Water, a membership body for water businesses in the UK.
Slessor explains some of the issues of the UK’s water infrastructure system, and how these gaps in regulation and understanding are being exacerbated by the gung-ho race for computational power and sovereign data centres to fuel AI ambitions.
Some Of the Basics
How is water even connected to data centres and AI?
A bit of the basics: the internet and the data underpinning it are physical – or, at least, their ability to function is predicated upon physical infrastructure. Data centres are essentially where the magic happens – servers store the data that underpins the internet, AI, and any of your files stored in the ‘cloud’. In order to do this, data centres require a lot of energy, which generates a lot of heat – in order to stop servers from overheating, these data centers need to be cooled.
Water is one of the most efficient ways of cooling data centres en-masse, via cooling towers or air cooling systems. The more computational power – the more data being run on a server – the more energy required and therefore, the more water to dissipate heat.
Data centres use various methods to cool using water, with traditional data centers often using potable – or drinkable – water to avoid any sediment build up in the cooling system.
Some cooling systems are ‘open-loop’ systems, meaning that they are directly connected to the same mains pipes that go into your kitchen sink, and are continually replenished as water evaporates in the cooling process. Though potable water enters these systems, it does not exit as drinkable water due to the use of chemicals to stave off bacteria, and would need to be retreated to reenter mains pipes.
Other ‘closed-loop’ systems are more water efficient, using a set amount of water that is reused in the same system. Water is typically ‘caught’ following its evaporation and recondensed to be fed back into the same system. Still, these systems often rely on potable drinking water, and depending on the system, water could be lost and would need to be replenished.
Water is also required for the power plants supplying energy to the data centres, which will undergo more strain and require more water. Water will also be used in any wastewater treatment facility that is used in the data centre water supply chain.
While these data centres do not always use potable drinking water, they do require water that has gone under some form of treatment and would not be able to use salt water so pipes do not corrode and sediments do not build up in these intricate cooling systems.
All of this adds up. In the US, it constitutes potentially billions of gallons of freshwater annually. While the Earth is composed of 70% of water, only about 2.5% of this water is fresh, 1.2% of that freshwater is on the surface of the earth, and 0.3% is available in lakes and rivers.
What systems most UK data centres use, how much water they use, and if they really need to be using potable water, are all valid questions – but the answers depend on who you ask.
This leads to one of the major problems of understanding the impact these data centres will have on the UK’s water supply – which, again, is not nearly as plentiful as the grey skies and dreich days would have you believe.
Murky Waters
Part of the issue lies in reporting, both within and outside of data centres.
While there is empirical evidence on how much water ChatGPT uses per prompt (just under a half litre bottle of water per 100 word email) or how much water some AI training uses, the day to day running of a data centre and its associated water usage ranged depending on the system employed.
“There’s no data anywhere to give an accurate usage forecast for water requirements,” Slessor said.
Councils do not require data centre developers to include water usage on their delivery plans in the UK, and there is a lack of proper regulation in industrial water consumption reporting. Industry often self-report using their own metrics, creating an opaque picture for regulators and the wider water industry to sift through.
This leaves the water industry, environmental regulators, and the government largely in the dark, with potentially devastating consequences.
Further, the water usage within data centres is still largely unregulated and unknown – a myriad of systems exist using water to cool large data centres.
“There’s very little empirical evidence because these are all privately closed systems that nobody sees,” Slessor said.
While AI companies have issued their own estimations and studies on how much water it takes to train their AI models, and independent reviews have shown how much water goes into an AI search query or prompt response, how this actually works largely depends on a number of factors.
The location of a data centre, the computational requirements of the system it is cooling, as well as the water cooling system itself, will all affect the amount of actual water used to cool a data centre.
While data centres currently exist in the UK, the country has yet to host any hyperscaler data centres that can run AI.
“AI takes it to the next level purely because it’s a continual process of large language models learning constantly all the time – the computational power required to do that, is about five times that you would see in things like the cloud,” Slessor explains.
AI data centres will put even more strain on energy and water resources, with environmental advocates and politicians calling for more transparency surrounding the environmental impact of data centres.
The Labour government’s first AI growth zone was designated to Oxfordshire, a region that’s often noted as being water-stressed, located next to a new reservoir, sparking fears about potential water usage.
While the UK government is calling on a reduction of non-household water usage by about 9% by 2037-38, new data centre water usage could mean that any gains in water sustainability by businesses simply do not measure up to increasing demand.
Without proper reporting of data centre usage in planning reports or environmental assessments, the UK will be largely left in the dark when it comes to tracking water consumption as water-stress expands in areas across the country.
We Have the Technology
According to Slessor, we have the technology to mitigate a large part of the problematic water usage that comes with data centre hyperscalers.
While most UK data centres use potable water “there is no technical reason that we have to use potable water in data centres,” Slessor said.
However, this idea seems to be dominating current data centre discourse, Slessor said.
Treated waste water would be usable in data centres to cool them, according to Slessor. While potable water is often used to avoid sediment build up in the expensive equipment used to cool data centres, the technology to treat and reuse waste water for data centres exists and is being used in different hyperscaler AI data centres.
Slessor provided examples of an Amazon Web Services data centre in Arizona – a US state known for its dry heat – using flat sheet ceramic filtration. AWS is using reclaimed, treated water at around 20 data centres in some of the driest regions in the US. Another data centre in the US run by xAI uses a ceramic membrane bioreactor to filter municipal wastewater for reuse in a data centre, allowing little water to be wasted in the cooling process.
“The technology exists to transform water from a sewage treatment centre into water suitable for a data centre,” Slessor assured. “None of this is new technology.”
Location, Location
This brings us to the location of data centres. While Scotland may be touted as an ideal place to house heat-creating data centres, with its cooler climate and more readily available water sources, not all of Scotland is as suitable as popular opinion would have you believe.
Water scarcity is increasing in the east of Scotland, strained by a lack of rainfall and a higher population. Despite this, data centres continue to pop up in water-strained areas, such as the data centre in Cockenzie just outside of Edinburgh.
“I understand why they would do this – it was an old power plant, it works for jobs in the area, and we all want data centres, we need data centres,” Slessor said. “But when you look at water consumption, it just doesn’t make sense.”
Locating data centres on the wetter West coast which has more historical pipes, or in the colder, wetter highlands, would make more logical sense.
Misplacing these data centres has a huge infrastructure cost – “Scottish Water is having to build big transfer mains and new reservoirs to move water around the country to make it easy to get water for these areas.”
If potable water is continually used in areas facing increasing water scarcity, then the results could be dire for the public as well.
“People will turn their taps on and nothing will come out,” Slessor warned.
Recommended reading
- How Much Water Are UK Data Centres Using?
- AI Model Training Guzzles Water and Spits Out Carbon, Mistral AI Finds
- Report: UK Public Unaware of GenAI’s Environmental Impact
A Systems, Common Sense Approach
“There’s a lack of forethought going into this,” Slessor lamented.
From a lack of robust reporting requirements and data, to a piecemeal approach to planning, the data centre issue really exposes a long-time issue in the UK’s approach to managing one of life’s most precious resources, water.
“There’s no sort of single plan as to what data centre planning will look like, there’s no one single approval authority that looks at this from a system thinking approach,” Slessor explained.
Without proper reporting from data centres on their current water usage, minimal understanding of the technical opportunities and limitations of water cooling, and little forethought in the infrastructure demands of water, the UK’s entire water policy seems to be based on a fundamental falsehood that water is plentiful.
“There’s zero planning for any of the planning authorities to consider what the water resource needs are,” Slessor said, discussing data centres planned for the water-scarce area between London and Swindon.
At the moment, planning authorities are facing difficulties building houses in the area due to a lack of potable water resources, but data centres do not face the same environmental hurdles.
“The reality is that if data centres take commercial water out of these systems, then it’s going to increase the cost of water for everybody,” Slessor said.
Right now, hyperscaler data centres are largely concentrated in the US, which has its own faults when it comes to data centre environmental regulation. Households in areas adjacent to data centres have seen their water compromised, either through prices or through pollutants.
When asked if the UK government has enough regulation or infrastructure to prevent this from being a reality in the UK, Slessor said: “There’s nothing. Nobody’s coming to say, ‘hang on a minute, what about water?’”
When it comes to what is needed, Slessor wants data to start at the very basics, because that’s where the UK water regulation currently is.
“Just basic usage figures of data centres, water in versus water out, would be huge to generate a better understanding for future planning,” Slessor said.
“The infrastructure is there,” Slessor said. Sewage and water treatment facilities exist, and placing data centres near these instead of near urban centres would ease water strain in water scarce areas, which are only increasing in size.
Using non-potable water, and “not making 21st century decisions based on 20th century assumptions,” would also be a common sense regulatory way to improve the water efficiency of data centres.
“This would actually be an opportunity for water companies to reduce bills because instead of paying for treated water to go back into the environment, water companies could sell the treated water to data centres,” Slessor said.
This would increase water revenue for water companies, and the UK, with the potential to even lower prices for household water usage.
Further, legislation about what happens inside of a data centre – the cooling systems employed, their daily water usage and water reuse schemes – would also ensure that these data centres mitigate their environmental and societal harms.
But data centre water discourse is exposing a much deeper issue in the UK, from the government to the public. Water is not viewed as a valuable resource in the UK, not when you look directly at how it is regulated.
“The UK water infrastructure is years behind,” Slessor says, considering how water recycling is “endemic” in much of the world and is only beginning to take hold across the UK.
While the UK government has set out some guidelines and targets when it comes to water consumption, Slessor says that future planning when it comes to population growth and climate change is not taking industry into account.
“We’re only scratching the surface with AI,” he said. This is true in the technology in itself, as well as its environmental effect and strain. Current planning “will be out of date in the next five or ten years,” as water strain and technological innovation accelerate.
This all does play into a fundamental issue in Scotland. Scots on average use about 40% more water than the rest of the UK, largely under the assumption that Scotland is lush with water. But after a record-breaking dry summer and a winter that is unlikely to fully replenish our water supplies, we may need to change our tune.
Data centres and their water strain can present an opportunity for Scotland and the rest of the UK to reassess their relationship with water as a precious important resource integral to life, but also used as a utility in industry and technological advancements. Data centres are large and physical – the cloud is a real place – it’s hot metal that needs to be cooled down.
“We need data centres,” Slessor said. “But in the long term, how do we do that? We make them environmentally sustainable, and we make sure we are using water in a way that is suitable and values it as a resource rather than just something that comes out of a tap.”
Part of this is seeing water as a resource, but also understanding how it is used in myriad areas of our everyday lives – to produce food, materials, and sustain technology.
“It’s a wider societal change around valuing our environment and people don’t often see the connection between the environment, energy and water in their everyday lives,” Slessor said.





