Scottish data centres powering AI already using enough water to fill 27 million bottles a year
3 hours ago Share Save Georgina Hayes BBC Scotland Share Save
Getty Images The US is the location of most new data centres but the UK is expanding quickly
Data centres powering artificial intelligence (AI) in Scotland are using enough tap water to fill 27 million half-litre bottles a year, according to data obtained by BBC News. AI systems such as the large language models (LLMs) that power OpenAI's ChatGPT and Google's Gemini require warehouses full of specialist computers. The equipment is power-hungry, consuming large amounts of energy, but they also use tonnes of water in their cooling systems to stop the servers overheating. Freedom of Information data shows the volume of tap water used by Scotland's data centres has quadrupled since 2021.
There are currently 16 data centres in Scotland and this number is set to increase in the years to come. Such centres have been powering the digital world for years - running everything from movie streaming to online banking - but the boom in generative AI tools has rapidly increased the amount of energy and water they use. In an interview with BBC Scotland News, Scottish Water described the increase in tap water used by data centres as "significant" - although it pointed out that it still only amounts to about 0.005% of the water supply. As AI booms - with 60% of the UK population already using it - Scottish Water wants the sector to look at sustainable alternatives such as wastewater systems. "We would like to try to look for other alternative solutions rather than using precious tap water", operations manager Colin Lindsay said.
Scottish Water's Colin Lindsay is concerned about the growing use of "precious" water supplies for data centres
Data centres are being built all over the world, especially in the US, and people in Scotland using AI tools will not necessarily being using servers in Scotland. However, many more data centres are expected to come to Scotland in the near-future, including a massive AI industrial park near Irvine in Ayrshire which was confirmed last month. "If we had to supply all that with tap water, then that would be a real concern," Mr Lindsay said.
Scottish data centres powering AI already using enough water to fill 27 million bottles a year
The BBC understands that the majority of data centres in Scotland currently use "open loop" systems, which need a constant supply of mains water. However, the industry is moving to towards more efficient methods such as "closed loop", meaning they would recirculate a fixed amount of water. Mr Lindsay said: "Open loop systems use enormous amounts of water. "We're working with developers on a case-by-case basis to explore sustainable water sources to reduce demand on public drinking water." He said closed-loop cooling systems could increase energy use so Scottish Water were encouraging open-loop systems near wastewater treatment works. These would use treated effluent to supply the volumes of water needed and minimise energy use.
In the UK alone, it's estimated that another 100 data centres will be built over the next few years to meet the demand for AI processing. The tech industry does not release figures on water consumption - and all Scottish data centres contacted for this article did not respond to our inquiries. It is estimated that 10-50 responses using AI model GPT-3 could consume 500ml of water. Experts at the University of Glasgow said the figures, revealed by BBC News, suggested that the water consumed by data centres in Scotland was equivalent to every person in the country drinking an extra 2.48 litres a year. By another measure, it amounts to more than 27 million 500ml water bottles. The university modelling also found the carbon footprint of these data centres could be the equivalent of every person in the country driving up to an extra 90 miles, or 145 kilometres, every single year. This is before any expansion in data centres in Scotland. And it does not account for the environment impact in the rest of the world of Scottish AI users.
... continue reading