TOKYO: As the use of artificial intelligence (AI) spreads and the technology grows in scale, it is also soaking up an increasing share of essential resources needed to power and maintain datacentres.
The International Energy Agency estimates that running a question by a chatbot uses 10 times as much energy as a Google search. But the growing number of datacentres that power AI not only burn through vast amounts of electricity, but also water.
The water needs to keep giant servers cool are so immense that by 2030 AI could use as much water as all of Sub-Saharan Africa, rivalling the needs of 1.3 billion people, according to a June estimate from the United Nations University (UNU) in Tokyo.
Unless there are major efficiency gains, datacentres are also set to triple their cooling water needs by the same year, according to Rystad Energy, an independent research and energy intelligence company.
Much of the water used in datacentres evaporates and can't be recycled. The researchers attribute the thirst to the "massive" scaling up in datacentre capacity currently under way and "the growing adoption of artificial intelligence in everyday life."
AI datacentres are meanwhile on track to need nearly three times as much electricity as Pakistan, Bangladesh and Nigeria combined, the UNU researchers say.
To help limit AI's strain on water and energy supplies, environmental experts at UNU recommend that consumers practise "fit-for-purpose" use of chatbots.
This means opting for simple text queries instead of generating unnecessary AI images or videos, using lighter and less resource-intensive models and turning to traditional search engines for basic questions.
How much water a datacentre needs for cooling depends on the technology being used as well as factors such as climate and location, Rystad said, with datacentres in cooler regions needing less water.
In turn, the effect on water supplies also depends on location and the nature of the local watershed and nearby agricultural requirements.
"Major datacentre operators, including hyperscalers such as AWS, Google, Microsoft and Meta, have committed to using water more sustainably," Rystad said. However, the impacts of pledges will vary depending on location and the level of water stress in a region, the researchers said. – dpa
