算力中心液冷电话会

Summary of the Conference Call on AI Computing Center Liquid Cooling Industry Overview - The conference focuses on the liquid cooling technology in the AI computing center industry, discussing energy efficiency standards and cooling methods used in data centers [1][2][3]. Key Points and Arguments 1. Energy Efficiency Standards: The industry defines energy efficiency using metrics such as PUE (Power Usage Effectiveness), CEUI (Carbon Emission Usage Index), and WUE (Water Usage Effectiveness). In China, PUE is the most clearly defined standard, with specific requirements for different regions [1][2]. - Northern regions: PUE ≤ 1.2 - Southern regions: PUE ≤ 1.25 - Other regions: PUE ≤ 1.3 - Beijing: PUE < 1.15 with a goal of 100% carbon neutrality within five years [1][2]. 2. Cooling Technology Selection: The choice of cooling technology depends on the maximum thermal design power (TDP) of the equipment and spatial constraints. High TDP chips may still use air cooling if space allows, while liquid cooling is preferred in confined spaces [2][3]. - Liquid cooling systems are about 25-30% more expensive than traditional air cooling systems, but this cost is relatively low compared to overall investment [2][3]. 3. Liquid Cooling Types: - Immersion Cooling: Higher costs due to expensive coolants and less mature supply chains, making cold plate cooling the mainstream choice [3][4]. - Direct-to-Chip Cooling: More effective for high-density heat sources, addressing uneven heat distribution issues [5][6]. 4. Cost Structure: - Cold plate cooling systems have a cost structure where the CPU costs approximately 1,000 to 1,500 RMB per kW, while immersion cooling systems incur higher costs primarily from the coolant [7][8]. - The cost of immersion cooling liquid is around 400,000 RMB per ton, with a standard tank requiring about 800 liters of coolant [7][8]. 5. Market Dynamics: The Chinese market typically employs a hybrid cooling model combining air and liquid cooling. The cost of single immersion cooling has decreased due to the use of synthetic oil substitutes [3][4]. 6. Key Players in the Liquid Cooling Supply Chain: - Manufacturers: Companies like Sugon and Inspur are involved in producing liquid cooling equipment, while specialized cooling suppliers like Shenyang and Yinvike focus on precision air conditioning [9][10]. - Component Suppliers: Companies like Gaolan and AVIC are entering the liquid cooling market by producing components previously used in other industries [9][10]. 7. Collaboration and Competition: The relationship between Taiwanese suppliers and domestic companies is characterized by complementary roles rather than direct competition. For instance, Nvidia collaborates with domestic firms like Midea for data center cooling solutions [10][11]. 8. Future Trends: The development of liquid cooling technology in China will increasingly rely on local chip manufacturing and liquid cooling integration, with companies like Sugon leading the supercomputing market [17]. Additional Important Information - The conference highlighted the importance of considering IT needs over cost when selecting cooling methods, emphasizing that infrastructure costs are a small fraction of total data center investments [2][3]. - The discussion also touched on the challenges of achieving complete silent operation in cooling systems, indicating that current solutions still rely on a mix of cooling methods [7][8]. This summary encapsulates the critical insights from the conference call, providing a comprehensive overview of the liquid cooling technology landscape in the AI computing center industry.