Huachuang Securities: Optimistic about the growth trend of the liquid cooling industry chain, recommends focusing on two types of investment opportunities.

Huachuang Securities released a research report stating that with the accelerated development of AI, liquid cooling is expected to become the mainstream cooling solution for future data centers. Cold plate liquid cooling is a mature short-term mainstream solution, while immersion cooling is expected to meet the demands of supercomputing in the future. The firm is optimistic about the growth trend of the liquid cooling industry chain under the background of AI development, and recommends focusing on two types of investment opportunities: 1) NVIDIA industry chain: heat dissipation module and component suppliers; 2) emerging cross-border automotive zero enterprises. Huachuang Securities' main points are as follows: AI accelerates development, and liquid cooling is expected to become the mainstream cooling solution for future data centers. In the global context of AI acceleration development: 1) The release of large AI models drives the growth of data center construction demand. Global cloud giants continue to increase capital expenditures, and trillion-dollar investment plans are gradually opening up. 2) Model iterations drive the demand for computing power, further driving the increase in chip, server, and data center power density, while also increasing the overall heat dissipation power of the system. In order to ensure the security, stability, and lifespan of the system, the importance of system thermal management becomes more prominent. Compared to traditional air cooling, liquid cooling solutions have significant advantages such as lower energy consumption costs (air cooling PUE value is about 1.4-1.6, liquid cooling is about 1.05-1.2), high heat dissipation (the heat dissipation power density of liquid cooling in a 2MW data center is about 4-9 times that of air cooling), gradually replacing air cooling as the mainstream cooling solution for data centers. Cold plate liquid cooling is a mature short-term mainstream solution, while immersion cooling is expected to meet the demands of supercomputing in the future. Liquid cooling technology routes are diverse, mainly including non-contact (cold plate) and contact (immersion, spray) types. Cold plate liquid cooling is currently the mainstream liquid cooling solution due to its relatively mature technology, complete ecosystem, low cost of retrofitting cabinets and servers, low initial investment, and similar operation and maintenance modes, as well as data center weighing and air cooling scenarios. With the increase in chip power consumption in supercomputing centers, and the stricter requirements for energy efficiency PUE of data centers in major countries and regions under the background of carbon neutrality, immersion cooling is expected to become the main cooling solution for high-power density in intelligent computing centers in the future. The total value of NVIDIA GB series cold plate liquid cooling system is about 70,000-100,000 US dollars. The cold plate liquid cooling system consists of two parts: primary side (outdoor side) and secondary side (indoor side). The core components of the secondary side mainly include CDU, pipelines, valves, liquid cooling plates, quick couplings, and Manifolds. Taking the NVIDIA GB200 and GB300 solutions as examples, the value of the secondary side of cold plate liquid cooling is about 600-800 US dollars per kW. The total value of the liquid cooling system of the GB series products is about 70,000-100,000 US dollars. Optimistically, the global scale of cold plate liquid cooling is expected to exceed 10 billion US dollars by 2026. Based on NVIDIA's latest guidance, the firm estimates that the shipments of NVIDIA GPUs in 2025-2026 will be 9.2 million and 12.5 million respectively. With the launch of a new generation of system architectures, the proportion of liquid cooling will also increase. It is estimated that the demand for cold plate liquid cooling by NVIDIA GPUs in 2025-2026 will reach 6.9 billion and 17.3 billion US dollars, while the estimated demand for liquid cooling by other manufacturers' ASIC chips will be 0.5 and 1.2 billion US dollars. The liquid cooling industry chain pattern is clear, with automotive zero companies focusing on upstream components to expand into the industry. The liquid cooling industry chain mainly covers three aspects: upstream components and IT equipment, midstream solutions (integrators), and downstream application scenarios (data centers, AI computing power, electronic terminals, etc.). Automotive zero companies, such as Three Flowers, Yinlun, Lingyun, Chuanghuan, and Feilong, have entered the industry starting with automotive thermal management and pipeline components, and gradually entered the industry through research and development. Their products mainly include upstream liquid cooling components such as pumps, valves, pipelines, quick couplings, Manifolds, etc., and are expected to quickly open up the market with years of technical accumulation and precision manufacturing advantages. NVIDIA's AI chips continue to iterate and also have a profound impact on the development of liquid cooling solutions. At the 2025 GTC conference, NVIDIA unveiled its next-generation AI chips, with power consumption expected to exceed 2000W. The thermal power consumption of the Rubin GPU may even increase from the original expected 1.8kW to 2.3kW, exceeding the current cooling plate's heat dissipation load. Therefore, there is a high demand for cooling, and the "microchannel water-cooled plate (MLCP)" technology is expected to become the mainstream trend in the next generation of cooling. Risk warning: Data center construction is slower than expected, the development of domestic liquid cooling solutions is slower than expected, geopolitical risks, calculation errors due to incomplete consideration of factors, etc.