As large language models grow bigger and chips run faster, a seemingly simple problem has become increasingly urgent: servers are getting too hot.
For years, data centers relied primarily on fans and air conditioning to keep cool. But with the rapid deployment of high-power computing clusters such as exascale artificial intelligence data centers and supernodes, chip power consumption continues to climb, pushing traditional air cooling to its limits. Liquid cooling has quickly emerged as the solution.
The computing industry is heating up — literally. Server heat output is rapidly approaching the capability limits of conventional air cooling. A cold plate no bigger than a palm now must dissipate 2,000 to 3,000 watts of heat, drawing warmth away from high-power chips through circulating liquid, experts said.
The market opportunity extends well beyond new data center construction. According to the market consulting company CCID Consulting, more than 80 percent of existing air-cooled data centers in China have the potential for liquid cooling retrofits. With new AI data centers multiplying and old facilities undergoing energy-efficiency upgrades, the liquid cooling industry is capturing growth from both "new builds" and "retrofits".
CCID Consulting data show that China's liquid cooling data center market reached 15.98 billion yuan ($2.38 billion) in 2025, up 45.2 percent year-on-year. Industry projections indicate the global AI data center liquid cooling market could surpass 100 billion yuan in 2026. As long as AI server power consumption continues rising, rack density keeps increasing and AI data centers keep expanding, the liquid cooling market will continue to grow.
At Range Intelligent Computing Technology Group Co, company President Li Li said the company's independently developed next-generation AI data center, or AIDC, has achieved systematic breakthroughs in multiple dimensions, including ultra-large single-building construction, convenient networking, ultra-high power, converged architecture and intelligent operations.
Meanwhile, leveraging its self-developed next-generation cold-plate liquid cooling technology, combined with high-efficiency energy-saving equipment and AI algorithms that dynamically optimize cooling and power supply chains, the company's AI data centers are steadily advancing energy conservation and carbon reduction. The technology can reduce Power Usage Effectiveness, or PUE, to as low as 1.15.
"Future AI data centers should be located close to densely populated areas. Langfang sits at the geometric center of the Beijing-Tianjin-Hebei region, within easy reach of Beijing, Tianjin and Xiong'an New Area, with access to the country's densest population resources," Li said.
Headquartered in Langfang, Hebei province, the company has mapped nine AIDC intelligent computing industrial parks globally. In Langfang alone, it has already put into operation eight internet data center buildings and four AIDC buildings, making its computing infrastructure scale one of the largest in China.
The first phase of the company's Artificial Intelligence Application Center has a total investment of 21.8 billion yuan. Its first B1 intelligent data center building is capable of supporting the training and inference of trillion-parameter-scale large language models.
China Unicom Digital Technology Co is also doubling down on AI computing infrastructure. Wen Yuan, director of the AIDC department at the company's cloud computing division, said the company is making tiered national deployments in line with the "east data, west computing" national strategy.
"Langfang is a key node in our layout," Wen said. "We will continue to increase our intelligent computing power construction, create stronger agglomeration effects, and optimize cost structures. We firmly support national policies and will fully leverage domestically produced computing power to meet diverse business needs. At the same time, we are further developing Unicom Cloud services to deliver better computing products to our customers."
masi@chinadaily.com.cn