"The end of AI is computing power, and the end of computing power is electricity," a saying widely circulated. However, another key element supporting the operation of computing power—water—has long remained out of the spotlight. The more intense the computing power, the faster the electricity meters and water meters run; heat dissipation in data centers and water consumption in power generation are becoming two rigid constraints on the expansion of computing power.
On October 16, during an inspection tour in Ulanqab City, Inner Mongolia Autonomous Region, President Xi Jinping proposed that the computing power industry should "focus on 'compute-electricity synergy' and 'compute-water synergy'," properly construct and operate computing center projects, and actively promote the development of green computing power industries.
The four characters "compute-water synergy" entered public view for the first time, setting new coordinates for high-quality computing power development.
Why Emphasize Compute-Water Synergy?
In the past, when discussing the green transformation of data centers, the focus was almost entirely on electricity. PUE (Power Usage Effectiveness) was the core assessment metric, measuring "how much of each kilowatt-hour is used for computation." But data centers are also major water consumers. The industry average WUE (Water Usage Effectiveness) is approximately 1.8 to 1.9 liters/kWh. Traditional air-cooled server rooms consume nearly two liters of freshwater for every kilowatt-hour of electricity used; cooling systems themselves account for three to four tenths of data center energy consumption.
The explosion of AI computing power has greatly accelerated the trend toward high-density server rooms, amplifying heat dissipation pressure exponentially. According to estimates by the United Nations University Institute for Water, Environment and Health, global data center water consumption will double between 2025 and 2030, reaching an estimated 9.3 trillion liters by 2030. A report by the China Academy of Information and Communications Technology (CAICT) also pointed out that while the "absolute water consumption" of data centers may not be large, the "relative pressure on water resource carrying capacity" cannot be ignored.
Industry standards are also changing. Starting December 1, 2025, WUE will be officially included in the national data center management standard system, making "water saving" a mandatory option. The National Green Computing Facility Evaluation Indicator System released this September explicitly constructs evaluation criteria across four aspects: energy utilization level, carbon emission management level, water resource utilization level, and computing resource utilization level. Data center greenification must look at both green electricity and water usage.
The logic of industry competition is also shifting. A joint report by CAICT and Grundfos states: "In the AI era, core competitiveness has shifted from simply competing on computing scale to a comprehensive contest of synergistic utilization capabilities of computing, energy, and water resources."
Lü Tianwen, President of the Zhongguancun Association for Collaborative Innovation in Information Industry, stated that computing power belongs to a resource-intensive industry, with consumption elements concentrated in water, electricity, bandwidth, and other resources. Electricity is the primary factor, followed closely by water resources. At the current stage, most data centers still rely on water for cooling. The strategic value of water resources and matching issues need to be re-examined.
So-called compute-water synergy essentially means placing water resources on equal footing with electricity. From the initial planning of computing projects, water source conditions and water usage volumes must be simultaneously evaluated to achieve coordinated sustainable development between computing scale expansion and water resource carrying capacity. The computing power industry cannot just keep an "electricity ledger"; it must keep both "electricity" and "water" ledgers simultaneously.
In Lü Tianwen's view, the proposal of compute-water synergy will urge governments and enterprises at all levels to comprehensively consider local resource matching, build and layout data centers more rationally, and serve as a more comprehensive and detailed guide for the "Eastern Data Western Computing" policy.
How Does Ulanqab Respond?
Ulanqab is a typical sample for observing compute-water synergy: it has outstanding advantages in climate, green electricity, and latency, but its water resource shortcomings are equally prominent.
As computing power moves west, water becomes scarce in the West. By the end of June this year, the intelligent computing scale in western regions accounted for 32.6% of the national total, yet these regions generally face water shortages, creating a mismatch between computing layout and water resources.
Take Ulanqab as an example. Its annual average temperature is 4.3°C, naturally suitable for heat dissipation; the annual average wind speed at 70 meters height is 7.2 to 8.8 m/s, offering rich wind power resources; electricity prices are approximately CNY 0.32 to 0.36 per kWh; direct fiber optic cables to Beijing compress latency to 4.2 milliseconds, making it a nearly ideal location for data centers.
Currently, Ulanqab's computing power industry development ranks among the top nationwide. As of mid-October, Ulanqab's operational computing power scale reached 190,000 PFLOPS, a year-on-year increase of 112%; intelligent computing accounts for over 95%, and the comprehensive green electricity usage rate exceeds 85%. Cumulatively, 109 computing center projects have been signed, with a contracted rack scale exceeding7 million racks.
However, Ulanqab is located in a semi-arid region. The total annual water resources of the entire autonomous region are only 2% of the national total, and per capita water resources are less than one-quarter of the national average. The larger the computing scale, the greater the water pressure. This is not just Inner Mongolia's dilemma. Finding the optimal solution for the "Electricity-Water-Computing" combination under climate, electricity price, and water resource constraints is a common test for western hubs.
Currently, Inner Mongolia has launched measures regarding data center water usage. Based on the "Regulations on Groundwater Protection and Management of Inner Mongolia Autonomous Region," production water for computing center projects is strictly prohibited from using groundwater and must switch to reclaimed water; liquid cooling and air cooling are vigorously promoted to fully replace traditional high-water-consumption cooling methods.
For instance, UCloud's Ulanqab Intelligent Computing Center adopts an integrated natural cold source air-cooled chiller unit scheme. Compared to traditional water-cooled cooling tower schemes, the overall water saving rate exceeds 92%, achieving ultra-low WUE (Water Usage Efficiency). Century Internet's Ulanqab Cloud Computing Center has completed nine server room buildings, primarily adopting split-type fluorine pump stacked dual-cycle air conditioning systems. For nearly 10 months a year, outdoor cold air is directly used for heat dissipation; sprinkler-assisted cooling is activated only during the hottest days of summer, and all sprinkler water uses reclaimed water. In 2025, Century Internet's actual operational project WUE was 0.099 liters/kWh, far below the national Level 1 (best) standard of 0.8 liters/kWh.
However, Che Kai from China Unicom Digital Technology Company stated that besides water, there are two variables requiring continuous attention in Ulanqab's computing acceleration process. First, power reliability needs testing. The higher the proportion of green electricity, the greater the risk of supply continuity. Although demonstration projects signed in the last two years have concentrated load deployment, fluctuations in power supply during the operational phase and whether energy storage can provide stable and low-cost supply still require long-term verification. Second, there is a risk of structural surplus in computing power. Expansion is happening simultaneously in eight major hubs and non-hub cities; computing supply will be centrally released in 2027-2028, making local oversupply almost inevitable, which requires advance planning.
How to Implement Compute-Water Synergy?
"The core of compute-water synergy is determining computing based on water and laying out based on water volume, putting the water demand of computing facilities and regional water resource carrying capacity into the same framework for unified coordination. It no longer focuses solely on PUE and compute-electricity synergy but simultaneously calculates the water resource ledger, achieving bidirectional adaptation and dynamic balance between computing development and water resource protection," said Zhu Keli, Founder and President of the Guoyan New Economy Research Institute and Chief Expert in Intelligent Economy.
Computing power water consumption mainly comes from two links: first, server room cooling, where AI intelligent computing clusters have high thermal density, and traditional evaporative cooling continuously consumes freshwater; second, water consumption accompanying the power production process. Zhu Keli warned that currently, intelligent computing projects are clustering in the west, while many western hubs themselves are water-scarce areas. If computing power expands disorderly, conflicts with agriculture, pastoralism, and ecological water use are highly likely. Essentially, compute-water synergy practices ecology priority and green development, correcting the tendency of "emphasizing electricity and neglecting water" in computing construction, ensuring computing layouts adhere to watershed water resource bottom lines, avoiding water resource overdraft risks from the source, and guaranteeing long-term industrial sustainability.
For the entire industry, compute-water synergy must establish access mechanisms from the planning source: treat water resource carrying capacity assessment as a prerequisite threshold for computing projects, calculating the water ledger before launching projects; unify computing water usage monitoring standards, connect data links between water affairs, computing parks, and water conservancy departments, and establish a real-time water usage scheduling system; promote reclaimed water reuse, rainwater collection, and circulating water treatment, optimize cooling processes, and form a system supported by technology, institutions, and markets.
For cold and arid computing hubs like Ulanqab, implementation paths need to be more precise. Locally, priority should be given to using reclaimed water for computing cooling, strictly prohibiting groundwater extraction; fully utilize the cool climate endowment, vigorously promoting air cooling and fluorine pump natural cooling, extending the operating duration without evaporative cooling, and minimizing evaporative water consumption. Simultaneously, coordinate local sewage treatment and reclaimed water pipeline network construction, connecting urban sewage resource recovery and computing park water supply links; adhere to cluster intensive layout to avoid scattered small computing projects causing dispersed water consumption; dynamically control new computing scale based on reclaimed water surplus and watershed water diversion capabilities, avoiding blindly launching large computing projects beyond water resource carrying capacity.
Regarding WUE control paths, Zhu Keli stated that industrial water saving mainly relies on three main lines: cooling route selection, water circulation renovation, and unconventional water source substitution. Future optimization must move beyond single-point renovation ideas towards systematic governance. First, set differentiated WUE guidance targets according to local conditions, enforcing stricter standards in water-scarce areas rather than a simple nationwide uniform approach; second, improve unified statistical calibers, distinguishing municipal freshwater from reclaimed water, guiding the industry to prioritize unconventional water sources, de-emphasizing pure WUE numbers while strengthening freshwater conservation orientation; third, drive cooling technology iteration, developing low-water-consumption solutions such as immersion liquid cooling; fourth, incorporate WUE into green computing evaluation, energy consumption supervision, and project approval systems, assessing it alongside green electricity indicators to form long-term incentives and constraints, guiding computing operators to proactively save water.
Liquid Cooling Solutions Accelerate Implementation
Solving data center construction in water-scarce areas is the core issue of compute-water synergy. Lü Tianwen stated that the primary solution is to minimize water resource usage, eventually reducing it to almost zero for cooling. This requires vigorous development of the liquid cooling industry, especially immersion liquid cooling, which can significantly reduce water consumption, or even use almost no water for data center cooling.
Liquid cooling is mainly divided into cold plate liquid cooling and immersion liquid cooling. Cold plate liquid cooling can solve 60-70% of IT heat dissipation problems, while immersion cooling can solve almost100%, thereby maximally reducing water consumption. Lü Tianwen noted that in areas with good environmental conditions, external natural heat dissipation can solve most heat dissipation problems; some hot and humid areas can deploy centralized cold sources and cooling towers to holistically resolve heat dissipation. Of course, new technologies will emerge, but liquid cooling will definitely be one of the primary heat dissipation methods for future high-density data centers.
Lü Tianwen emphasized that looking at current technological maturity, cold plate liquid cooling is relatively suitable for Ulanqab's actual scenarios, while immersion liquid cooling will be more suitable in the future. The main reason for prioritizing cold plate liquid cooling is that server manufacturers need to cooperate with immersion liquid cooling by redeveloping servers, setting up production lines, etc. Therefore, immersion liquid cooling is not just about the cooling system; it requires viewing the complete industrial chain.
Che Kai also stated that Ulanqab's annual average temperature of 4.3°C allows for the use of natural cold sources for nearly 10 months a year, further amplifying the water-saving advantages of liquid cooling technology. However, compared to air cooling systems, liquid cooling technology has a more complex overall architecture, increasing system leakage, material corrosion, and operational complexity. At the current stage, supply chain and operational maturity lag behind air cooling. Looking solely at water-saving value, with medium-to-low rack density, air cooling plus indirect evaporative cooling can already achieve extreme WUE performance in cold and arid regions.
Is Computing Power Development Entering the 2.0 Era?
Che Kai stated that this inspection tour will affect the entire industry's development logic. The computing power industry will be formally redefined as resource-intensive infrastructure, implying that the industry is transitioning from rapid development to a stage of higher-quality development.
First, the siting logic is being reset. Over the past decade, data center siting was essentially cost arbitrage (low land prices, low electricity prices, government subsidies, and climate characteristics). Green electricity, compute-electricity synergy, and compute-water synergy transform electricity and water from costs into constraints. Future siting will no longer be about going where it is cheapest, but focusing on resource carrying capacity and environmental support capability.
Second, the security logic is upgrading. Requirements for green electricity proportions in data centers have driven the implementation of projects involving source-grid-load-storage and direct power supply. Meanwhile, the instability of green electricity and the scaling of energy storage bring certain safety hazards. Reliability, as a hidden cost of greenness, must be fully considered.
Third, "Eastern Data Western Computing" may enter the 2.0 stage. Competition among hubs shifts from competing on rack counts to competing on resource conversion efficiency. How many effective Tokens are produced per kilowatt-hour of green electricity may become the ultimate metric.
Cooling methods follow water resource carrying capacity, and water usage quotas follow computing output—only when this ledger balances can computing power "land successfully." The deep proposition of compute-water synergy perhaps lies not in equipping every data center with more water-efficient cooling systems, but in ensuring that water resource constraints truly enter the decision function of computing layout, weighed on the same table as land, electricity, and industrial policies.
Editor/Layout: Zhu Wenfeng
Proofreader: Mei Yaxin
Supervisor: Liu Qicheng