Water pressure on AI computing power centers rises, Shanxi, Inner Mongolia, and Henan explore computing-water coordinated water-saving solutions

Related topics top start Related topics top end Futures promotion begin Futures promotion end Miaopai begin Miaopai end Video player start Video player end Market chart begin Market chart end

This article is reproduced from [CCTV Finance];

Currently, demand for hardware such as computing power in the artificial intelligence industry is growing exponentially, and domestic computing power centers are also accelerating their layout. Large model training and inference will cause computing power centers to continuously generate heat, which needs to be dissipated in a timely manner through cooling systems. Behind this lies not only electricity consumption but also water resource consumption. How to achieve computing-water coordination and improve water resource utilization efficiency has become a real challenge in computing infrastructure construction.

Under different water resource conditions, what new models have computing power centers explored in terms of water conservation, energy efficiency, and regional development? Let's take a look at the reporter's research in Shanxi, Inner Mongolia, and Henan.article_adlist[

]article_adlist

Guangling, Shanxi: First water storage project for a computing power center to be completed by the end of the year, water resource utilization yields multiple benefits

In Guangling County, Datong, Shanxi, Shanxi's first water storage project serving a computing power center is underway. The project's water source is surface water, with a storage capacity of 4 million cubic meters, which can meet the annual water demand of a nearby 1GW computing power center. This water storage project draws water from the Mentouyu Reservoir in Lingqiu County, Shanxi, and builds a new water transmission trunk line with a total length of about 60 kilometers, crossing Lingqiu and Guangling counties; through surface water diversion and cascade utilization, it can reduce the use of local groundwater resources.article_adlist[

]article_adlist

Ju Jing, founder and CEO of Qineng Technology, introduced that this 4 million cubic meters of water can approximately meet the usage capacity of the entire data center together with agricultural water supply for nearly two to three months; during the wet season, the water storage project can also effectively regulate the capacity of upstream reservoirs.

Near the water storage project, cooling equipment for the second phase of the computing power center is being installed. The project's main water use period is concentrated in the summer wet season, using spray facilities on the outdoor units of the cooling equipment to carry away heat and achieve cooling. Xu Diqian, operations and maintenance manager of Qineng Technology's Guangling Park, explained that water consumption mainly comes from the wet film humidification process: water is sprayed onto the wet film, and evaporation lowers the air temperature, similar in principle to an air conditioner outdoor unit; in the winter dry season, when the local environment is cold and dry, water demand basically drops to a very low level.

The reporter learned that the water supply project of the Guangling computing power center is expected to be completed by the end of this year. The value generated by the project includes operating income from residential water supply, agricultural water supply, and computing park water supply, achieving synergy between public infrastructure investment and regional industrial development. After the project is completed, it will also become an important supplement to local residential and agricultural water use.

Ulanqab: Water-scarce areas explore a new liquid cooling model, using more electricity and less water

At present, a large number of newly built computing power centers in China are located in the western region, and many of these areas are themselves water-scarce.

In Ulanqab, Inner Mongolia, the operating ultra-large computing power center Xinghe Base project does not use groundwater for cooling, but instead uses urban reclaimed water and builds its own sewage treatment project; hundreds of tons of water are charged into the system once during the initial construction phase, and thereafter circulate in a closed internal loop for liquid cooling and heat dissipation. The project also adopts a "waterless liquid cooling" solution, replacing cooling towers with dry coolers, greatly reducing evaporation water consumption and achieving water-saving and low-carbon operation of intelligent computing centers in arid regions.

Chen Cong, HVAC product expert at Envision AI Intelligent Computing Center, introduced that the cooling solution must first consider extremely low water consumption, or even zero-water cooling forms; this solution's water consumption is only 5% of the annual water consumption of traditional cooling towers, and in cooler climate regions, it is expected to achieve zero water consumption throughout the year. He explained that dry coolers do not require evaporative cooling, but instead exchange heat between external air and an internal circulating medium, greatly reducing water consumption compared with traditional cooling towers. Taking a 100-megawatt computing power center as an example, if cooling towers were still used for heat dissipation, annual water consumption would be about 700,000 to 800,000 cubic meters; therefore, from the top-level design and planning stage, low-water or even waterless cooling forms need to be considered to meet the layout and development needs of computing power projects.

Industry estimates show that water evaporation has a strong heat absorption capacity; when 1 cubic meter of water completely evaporates, it can theoretically carry away about 2.5 million kilojoules of heat; under certain conditions, using water evaporation for heat dissipation can reduce the electricity demand for mechanical refrigeration, but water resource consumption must also be counted. Water and electricity are like the two ends of a seesaw: using more water can save electricity, and using more electricity can save water.

Zhengzhou, Henan: "Adapting to water conditions" empowers liquid-cooled computing power and makes good use of existing water resources

In Zhengzhou, Henan, the idea of relying on the city's existing water bodies for heat dissipation has already been implemented. The core node of the National Supercomputing Internet relies on the existing landscape water body of Beilong Lake to build a phase-change liquid cooling heat dissipation system cooled by lake water, and the municipal tap water consumption for computing operations is almost zero.

Inside the computer room, the 100,000-card computing power supercluster equipment releases a large amount of heat; in the core heat exchange room on the second floor, heat exchange equipment completes heat transfer between different water bodies. Zhang Peng, senior vice president of Sugon Liquid Cooling, introduced that the computer room building is hexagonal, with a water outlet on each side; the heated high-temperature water passes through the artificial landscape lake and a 1.2-kilometer heat dissipation channel, and finally returns to the lake body, with the return water temperature rising by no more than 1 degree Celsius. The entire system uses lake water cooling combined with phase-change immersion liquid cooling technology, completing heat dissipation through three layers of water circulation and two heat exchanges.

Industry estimates show that such a scale can save more than 1 million tons of municipal tap water each year, and can also save tens of millions or even hundreds of millions of yuan in cooling tower costs. Zhang Peng introduced that computing infrastructure construction should reduce the municipal tap water consumption of computing power centers as much as possible; in regions with natural resource conditions such as rivers, lakes, and seas, building with existing water resources will be relatively more economical.

Industry observation: From competing on computing power scale to balancing computing-water coordination

Industry insiders say that competition in the computing power industry is shifting from simply comparing rack numbers and computing power scale to competition in comprehensive capabilities such as the share of green electricity, water consumption indicators, land use, and operating costs behind each unit of computing power, and adapting to local conditions is becoming an important issue in the layout and green transformation of computing infrastructure. Zhang Chen, deputy general manager of the Market Development Division of China Power Engineering Consulting Group Co., Ltd., believes that whether to build a data center and how large to build it should not depend only on electricity prices and land, but also on whether the water account can be balanced; for water-scarce areas, blind expansion of water-cooled, high-water-consumption computer rooms should be strictly controlled.

At the same time, some existing projects within the computing power industry are still relatively extensive in water source intake and metering supervision, and have not effectively formed value feedback to local development. Industry insiders suggest improving the mechanisms for paid water intake and centralized metering, and simultaneously improving supporting construction of water conservancy, reclaimed water recycling, and closed-loop circulation facilities.

Reporters/Ping Fan, An Yu news_keyword_pub,stock,bj920808 iframe below main text iframe below main text General statement General statement QR code at end start QR code at end start

QR code at end start QR code at end start Editor name and work code Editor name and work code