Core Insights - The article discusses the rapid adoption of liquid cooling technology in AI data centers, driven by the release of NVIDIA's GB200 NVL72 server, with penetration rates expected to rise from 14% in 2024 to 33% in 2025 [2][4]. Group 1: Liquid Cooling Technology Adoption - The introduction of NVIDIA's GB200/GB300 NVL72 systems, which have a thermal design power (TDP) of 130kW-140kW, necessitates the use of liquid-to-air (L2A) cooling technology due to the limitations of traditional air cooling systems [4]. - Liquid-to-liquid (L2L) cooling architecture is anticipated to gain traction starting in 2027, as new data centers are completed and AI chip power consumption increases, gradually replacing L2A technology [4]. Group 2: Market Expansion and Key Players - Major cloud service providers (CSPs) in North America are expanding AI infrastructure, with companies like Google and AWS implementing liquid cooling-compatible facilities in Europe and Asia [5]. - The demand for cooling modules, heat exchange systems, and peripheral components is expected to grow as liquid cooling penetration increases [5]. - Key suppliers of cold plates, a core component of contact heat exchange, include Cooler Master, AVC, BOYD, and Auras, with several expanding production capacity in Southeast Asia to meet demand from U.S. CSPs [5]. Group 3: Critical Components in Liquid Cooling Systems - The fluid distribution unit (CDU) is essential for heat transfer and coolant distribution, with the market currently favoring sidecar CDU configurations, led by Delta [6]. - Quick disconnects (QDs) are crucial for the reliability and safety of liquid cooling systems, with major manufacturers like CPC, Parker Hannifin, Danfoss, and Staubli involved in the NVIDIA GB200 project [6].
研报 | AI数据中心将规模化导入液冷散热技术,预估2025年渗透率逾30%