Core Viewpoint - NVIDIA is set to release its AI server, the NVIDIA DGX GB300NVL72, designed for the AI inference era, featuring enhanced computing power and a fully liquid-cooled rack design to meet the increasing thermal design power (TDP) requirements [1][2]. Group 1: AI Server Development - The NVIDIA DGX GB300NVL72 server will have a 1.5 times improvement in inference performance compared to the previous model, the DGX GB200 [2]. - The server's high chip computing capability leads to an increase in TDP, necessitating advanced cooling solutions [2]. Group 2: Liquid Cooling Technology - Various liquid cooling solutions exist, including cold plate, immersion, and spray cooling, with high efficiency relying heavily on fluorinated liquid materials [3]. - Liquid cooling can be categorized into indirect and direct cooling, as well as single-phase and two-phase cooling, with cold plate technology being the most mature and widely used [3]. Group 3: Demand for Electronic Fluorinated Liquids - The demand for electronic fluorinated liquids is expected to rise significantly, reaching 122,000 tons and 256,000 tons by 2025 and 2026, respectively, representing year-on-year growth of 150% and 110% [4]. - The global fluorinated liquid market is concentrated, with the top five manufacturers holding 80% of the market share; the exit of 3M will create structural opportunities for domestic fluorinated liquid manufacturers [4]. Group 4: Thermal Interface Materials (TIM) - The combination of efficient TIM materials, such as liquid metal, with two-phase cold plate cooling solutions offers significant advantages [5]. - Liquid metal TIM materials are ideal for high-end chips, compatible with various cooling methods, and are expected to become mainstream in the future [5]. Group 5: Investment Recommendations - The liquid cooling solution is anticipated to become the mainstream cooling method for AI servers, with a focus on upstream raw materials like fluorinated refrigerants and electronic fluorinated liquids, as well as efficient TIM materials [6].
AI服务器散热进入液冷时代,导热散热材料需求突显 | 投研报告