Liquid cooling
Can liquid cooling help data centres avoid water wars?
Thanks to the AI boom, we need to build new data centres. But we also need better ways to cool them, says Nicole Kobie
@PCPRO
FACEBOOK.COM/PCPRO
The trends and technologies that are the shape of things to come
As the hillsides surrounding LA burned, reports emerged of firefighters running out of water to douse the flames. In response, California lawmakers introduced bills designed to force data centres to publish how much water they use and work towards using less – but two weeks later, President Trump was promising to go big on data centres to power the AI boom, with $500 billion pledged to make it happen. The Labour government has promised something similar here in the UK, in the hope AI will boost the economy.
Can we build the next generation of data centres without dooming ourselves to water shortages, or even water wars? It’s possible – and may even be technically necessary.
Data centres already use as much as 2% of global electricity, on a par with aviation, but it’s predicted to climb to 4% within five years. About 40% of that power is used for cooling. And, as AI sparks demand for bigger, more complex data centres with cutting-edge chips, more energy and therefore more cooling is required.
Already, a small one-megawatt data centre using traditional air cooling slurps through as much as 26 million litres of water annually. Such big numbers are hard to comprehend, but in the city of The Dalles, Oregon, a legal fight between locals and Google revealed the latter’s data centre used a quarter of the region’s public water. And that’s a problem when a fifth of data centres are located in water-stressed areas, according to research from the University of Tulsa.
The rise of AI is exacerbating an already growing problem. Research from the University of California Riverside and the University of Texas Arlington revealed that training OpenAI’s GPT-3consumed 185,000 gallons of water, while answering 20-50 questions requires about half a litre of water. Now imagine every search query running via AI.
“Big data centres... are hungry and thirsty,” said Colm Shorten, senior director for data centres at JLL, a global consultancy and real estate firm.