A ChatGPT-led boom in artificial intelligence (AI) is expected to fuel an exponential growth in demand for liquid cooling technology from cooling module suppliers and data centre operators, given the need to keep temperatures and carbon emissions from servers down.
The global market for overall cooling systems could exceed US$9 billion by 2027 from about US$2 billion this year, according to Macquarie Capital. The adoption of liquid cooling systems in total data-centre units is projected to reach 22 per cent from 1 per cent over the same period, it said.
Liquid cooling solutions, which use liquids such as water or a refrigerant instead of air to cool data centres, are expected to fuel that growth. Its market value is forecast to grow almost 25 times to US$7.8 billion over the next three years, from US$317 million currently, according to Macquarie.
“How to get the heat out and cool down the system has become a major challenge for the [data centre] industry,” Arthur Lai, head of technology research of Asia at the Sydney-based investment bank, said in Hong Kong on Tuesday. “This would also be a big opportunity for investors.”
18:59
Why the EU, US are concerned about China’s overcapacity
Why the EU, US are concerned about China’s overcapacity
Cooling, which reduces the temperature generated by server processors, consumes around 40 per cent of total power demand in data centres. It is the second-largest source of electricity consumption in such facilities, after the computer servers’ own electricity demand.
Because of the growing importance of cutting-edge AI, data centres are expected to account for 3 per cent of global electricity consumption by 2028 and 7 per cent by 2035, according to Macquarie. That means hitting about 3,100 terawatt-hours in a decade, the investment bank added.
Sam Altman, the CEO of ChatGPT creator OpenAI, said the future of AI depends on energy breakthroughs. Tesla and xAI founder Elon Musk said “the next shortage will be electricity.” Next year, Musk added, “you’ll see they just cannot find enough electricity to run all the chips [for AI data centres]”.
Currently, most data-centre servers are still air-cooled by fans. However, this method is no longer enough to overcome the heat generated by AI chips and servers, nor is it energy-efficient. Shifting to liquid cooling systems can cut data centre power consumption by about 10 per cent, Macquarie said.
Although current liquid cooling systems are normally 11 times more expensive than air cooling ones, they can generate cost savings in the long run through lower energy bills, according to Lai. In Asia, cooling module makers Asia Vital Components and Auras of Taiwan and Shenzhen-based Envicool, are among the market leaders.
“This is not an option but a must for the [data centre] industry to go for liquid cooling,” he added. “There’s already a big industry [in liquid cooling solutions] at this moment.”
Singapore, Germany, and Japan have also set requirements for data centre energy efficiency.
Standardised environmental metrics are needed to measure the energy efficiency of data centres, said Pankaj Sharma, senior vice-president of secure power division at Schneider Electric. Data centre operators should look into energy, greenhouse gas emissions, water, waste, and impacts on the local ecosystem, when delivering their sustainability goals, he said.