Amid the AI computing power boom, the liquid cooling industry is receiving multiple clear signals for development!!

Driven by continuous iteration of large models, demand for computing power has exploded. Power consumption of AI chips keeps rising, and the construction of high-density intelligent computing clusters is accelerating, while traditional air cooling is gradually hitting its physical ceiling. Multiple clear industrial signals have emerged: liquid cooling is shifting from an alternative for computing center upgrades to essential infrastructure supporting stable operation of high-end AI computing power. A growth storyline driven by the resonance of technology, policies, markets and industrial chains is becoming increasingly distinct.

From the perspective of underlying technical signals, continuous breakthroughs in computing density force the iteration and upgrading of cooling routes. Power consumption of current mainstream AI chips keeps climbing. Single cabinet power consumption of 10,000-card-level intelligent computing clusters generally exceeds 120kW, while the theoretical upper limit of traditional air cooling is merely 20 to 30kW, making it unable to support the deployment of new-generation high-density computing power. Liquid media boasts far higher heat exchange efficiency than air, serving as a viable approach to break the cooling bottleneck. Major global computing platforms have established clear orientations: NVIDIA Rubin architecture explicitly removes air cooling adaptation and adopts full liquid cooling as standard equipment. Domestic super-node products such as Huawei Ascend and Inspur are advancing standardized liquid cooling adaptation simultaneously, and high-end AI training clusters are phasing out air cooling options. The technical landscape is stabilizing: cold-plate liquid cooling dominates the market thanks to strong compatibility and low retrofitting thresholds, suitable for newly-built computer rooms and reconstruction of existing computing centers. Immersion liquid cooling, featuring ultra-low PUE and recoverable waste heat, is accelerating pilot deployment for large-scale 10,000-card clusters. This forms a development pattern with large-scale adoption of cold-plate solutions in the short term and continuous penetration of immersion solutions in the long run.

Policy signals keep being released, and rigid energy consumption constraints accelerate the popularization of liquid cooling. Policies concerning high-quality development of computing infrastructure and green low-carbon transformation of data centers keep being implemented. PUE indicators for newly-built large and super-large intelligent computing centers are continuously tightened, and requirements for high-density AI computing projects are further stricter. Air cooling architectures can hardly meet energy efficiency control standards, which directly affects approval of project energy consumption indicators and acquisition of power quotas. Computing hub cities including Beijing, Shanghai, Guangzhou, Chengdu-Chongqing and the East-to-West Computing Transfer Project have rolled out supporting measures. They support the construction of liquid cooling computer rooms through project subsidies and computing power vouchers. Some regions have issued guiding requirements that newly-built high-density intelligent computing centers prioritize high-efficiency liquid cooling solutions. Meanwhile, industrial standards such as Technical Specification for Liquid Cooling Systems of Data Centers have been officially put into effect. Unified specifications for the design, construction, operation, maintenance and acceptance of liquid cooling computer rooms are gradually improved, while national standards for evaluation and energy efficiency assessment are being accelerated. These efforts continuously resolve standardization challenges in large-scale construction and remove institutional barriers for widespread industry promotion.

Inflection points in market demand have appeared, and industry penetration is entering a rapid upward channel. Statistics show that domestic penetration of liquid-cooled servers has steadily climbed from less than 3% in 2021 to 20% in 2025, expected to reach 37% in 2026 and break the critical threshold of 50% in 2027. Almost all newly-built 10,000-card-level AI intelligent computing centers plan liquid cooling solutions, and the market size maintains rapid growth. The domestic liquid-cooled data center market will exceed 23 billion yuan in 2026, with a 35%-40% compound annual growth rate over the next three years. The demand structure presents obvious stratification: supercomputing clusters for large model training prioritize immersion and whole-rack cold-plate solutions; cloud computing and general inference computing clusters mainly adopt cold-plate retrofitting; high-density computing scenarios for robots, autonomous driving and finance follow up with deployment. Economic viability has been gradually verified by the market. Although the upfront construction investment of liquid cooling is higher than air cooling, it can greatly reduce long-term refrigeration energy consumption. Large computing clusters generally achieve cost recovery within 2 to 3 years. Lower operating temperatures also reduce hardware failures and extend server service life, continuously boosting investment willingness.

Ecosystem signals of the industrial chain keep strengthening, coordination between upstream and downstream players speeds up, and the industry competition landscape is reshaped. Major domestic server manufacturers are fully advancing liquid cooling product layouts. Inspur, Sugon, Huawei and H3C keep launching standardized liquid-cooled servers and integrated intelligent computing cabinets. Upstream core component manufacturers for cold plates, CDUs, special coolants and sealing assemblies keep expanding production. Cross-border capital is accelerating entry into this track, and the localized supporting capacity of the industrial chain is continuously enhanced. The industry is overcoming previous pain points of heavy project customization and low delivery efficiency. Modular and standardized liquid cooling solutions are continuously launched, driving the industry to transform from customized engineering projects to mass product delivery.

Beyond opportunities, objective challenges remain for industrial development. Operation and maintenance specifications for liquid cooling systems still need wider popularization. Small and medium-sized enterprises face heavy initial investment pressure. Long-term operation and maintenance experience for immersion solutions needs accumulation, and there is still room for technological improvement in high-end media and precision components. In the long run, the dual trends of AI computing power expansion and green low-carbon development are irreversible. It has become a consensus in the industry that cooling capacity determines the upper limit of computing power. With continuous improvement of various mature conditions, the liquid cooling industry will embrace a sustained expansion cycle and become a key track supporting high-quality development of China's artificial intelligence industry and the construction of a green computing foundation.