格隆汇12月11日丨集泰股份(002909.SZ)在互动平台表示，公司研发的液冷硅油产品目前正与相关客户 进行合作测试，由于该产品验证周期相对较长，因此尚未实现规模化生产与市场投放。公司现有的生产 产线具备良好的柔性生产能力，能够适配未来可能的生产需求。公司将持续关注行业技术趋势与客户验 证反馈，稳步推进研发成果的转化进程。 ...

Company Overview - Feilong Automotive Parts Co., Ltd. has over 70 years of development history, focusing on thermal management system solutions [4] - The company has two main development phases: before 2017 focused on automotive thermal management components, and from 2017 onwards, expanding into civilian markets [4][5] Product Development - The company has established four R&D centers in Shanghai, Wuhu, Neixiang, and Xixia, with a focus on new energy thermal management components [4] - Current product offerings include electronic pumps and temperature control valves, with sales covering over 200 domestic and international clients [4][5] Subsidiary Information - Anhui Hangyi Technology Co., Ltd. was established in July 2025 with a registered capital of 50 million RMB, focusing on non-automotive liquid cooling pumps [5] - Hangyi Technology aims to become a leader in liquid cooling pump technology, offering solutions ranging from 8W to 37kW [5] Market Position and Strategy - The company is expanding its product applications in various sectors, including 5G base stations, AI cooling, and renewable energy [8][9] - The average price of electronic pumps in the civilian sector is currently higher than in the automotive sector, indicating a lucrative market opportunity [8] Customer Engagement - The company has established partnerships with over 40 leading firms in the server liquid cooling sector, including HP and Delta [9] - The company is actively developing multiple liquid cooling technologies to enhance its market position [10] Production Capacity and Global Expansion - The company has sufficient production capacity to meet rapid growth in the liquid cooling sector and is planning to expand capacity dynamically [11] - The first overseas smart factory in Thailand was completed in June 2025, focusing on engine thermal management products and new energy thermal management products [12] Risk Management - In response to U.S. tariff policies, the company is accelerating the construction of its Thailand production base and diversifying its international market strategy [12] Future Outlook - The company aims to leverage its core strengths in the automotive sector while expanding into high-growth areas such as commercial liquid cooling and energy storage systems [12]

Summary of Conference Call on Liquid Cooling Technology Industry Overview - The conference focuses on the liquid cooling technology industry, particularly related to NVIDIA's GB300 and upcoming Rubin series products [1][3][5]. Key Points and Arguments Liquid Cooling Technology Developments - NVIDIA's GB300 features a full liquid cooling plate design that covers chips, memory, and power supply, but faces limitations with single-direction cooling plates, indicating a need for dual-direction plates or microchannel technology by 2026 [1][3]. - Dual-direction plates enhance cooling capacity through phase change and efficient media, while microchannel technology improves conduction efficiency but faces challenges in processing and uniformity [1][3][9]. - Vertiv's experiments with NVIDIA's NV72 system show a significant reduction in liquid usage, from 700 liters to approximately 360 liters, by employing localized silent cooling [1][7]. Cost and Component Changes - The number of quick connectors in GB300 has increased from 108 pairs in GB200 to 252 pairs, leading to a 20% to 30% increase in total cooling plate costs, primarily due to quick connectors and cleaning processes [2][29]. - The overall cost of the GB300 liquid cooling plate has risen from $49,000 to $59,000, driven by increased complexity and component requirements [29]. Future Trends and Challenges - The upcoming Rubin Ultra project is expected to adopt NV576 architecture with a TDP of 1,400 watts, necessitating advanced cooling solutions [8]. - The development of microchannel technology and dual-direction plates will significantly impact the competitive landscape, potentially introducing new players into the market [20]. Supplier Landscape - Current leading suppliers for GB300 include Qihong and Shuanghong, with domestic manufacturers like Industrial Fulian entering NVIDIA's supply chain due to cost control and rapid response capabilities [1][19]. - Domestic suppliers are increasingly participating in the market, with companies like Zhonghang Optoelectronics and Fengguang beginning to sample or collaborate to meet the growing demand for quick connectors [24]. Material and Pump Innovations - The liquid cooling system's evolution includes a shift from traditional materials to advanced options like R1,233 refrigerant, which is more environmentally friendly and has a higher boiling point than R134 [14][15]. - The importance of pumps in liquid cooling systems is rising, especially with microchannel applications requiring high stability and smooth operation [16][33]. Emerging Technologies - New materials such as graphene and silicon carbide are being explored for their thermal management properties, although their widespread application remains limited [31][32]. - Domestic universities are actively researching microchannel technology, which may lead to significant advancements in the field [34]. Additional Important Insights - The liquid cooling technology market is evolving rapidly, with significant opportunities for domestic suppliers to close the gap with established international players [21][22]. - The ASIC server market may adopt liquid cooling solutions, with companies like Google potentially leading the way, although the urgency compared to NVIDIA's needs is still under evaluation [25][26].

随着AI芯片功耗迈入"千瓦时代"，一场围绕散热的革命正悄然加速，液冷正从一个"可选项"迅速变为数据中心的"必选项"。 综合瑞银与东吴证券近期发布的深度报告，市场似乎仍未完全消化这场变革的颠覆性。报告明确指出，市场普遍低估了向液冷过渡的速度 和规模，这并非一次普通的硬件升级，而是由AI芯片激烈竞争驱动的"即时、战略性必要之举"。 这一结构性转变不仅将催生一个高达311.91亿美元（约合人民币2205.23亿元）的超级周期，更在重塑全球供应链格局，特别是为国产厂商 打开了进入核心体系的大门。 "热墙"来袭：当风冷走到尽头，液冷即将迎来爆发 AI算力的爆炸式增长，正将芯片推向一道"热墙"。据瑞银及东吴证券报告梳理，AI加速器的热设计功耗（TDP）正以惊人的速度攀升。 英伟达的GPU产品路线图清晰地展示了这一趋势：从H100的700W，到Blackwell B200的1200W，再到预计明年推出的VR200的1800W- 2300W，以及2027年可能超过3600W的VR300，直至未来Feynman平台可能高达5000W-7000W的功耗。 芯片功耗的飙升直接导致机柜功率密度急剧攀升。一个GB200 NVL72机柜 ...

Group 1 - Liquid cooling technology is essential for addressing the heat dissipation pressure in data centers, offering advantages such as low energy consumption, high heat dissipation, low noise, and low total cost of ownership (TCO) [1][2] - The value of liquid cooling systems is expected to grow significantly alongside chip upgrades, with a projected increase of over 20% for rack liquid cooling modules in GB300-GB200 servers [1][2] - The market for ASIC liquid cooling systems is estimated to reach 35.3 billion yuan by 2026, while the market for NVIDIA liquid cooling systems is projected to reach 69.7 billion yuan [1][2] Group 2 - The domestic supply chain is accelerating its entry into the liquid cooling market, with NVIDIA opening its supplier list to allow more flexibility in supply chain composition [2] - The Rubin architecture requires new liquid cooling solutions due to its high thermal design power (TDP) of 2300W, necessitating alternatives to single-phase cooling plates [2] - Two viable solutions for the Rubin architecture include phase change cooling plates and microchannel plates, with the latter likely becoming the preferred option as power requirements increase [2] Group 3 - The demand for AI server computing power is rapidly increasing, driving the penetration rate of liquid cooling solutions, and leading to an explosive demand for efficient and energy-saving heat exchange solutions in data centers [3] - Companies to watch in this sector include Yingweike, Shenling Environment, Gaolan Co., Hongsheng Co., Zhongke Shuguang, and Jieban Technology [3]

Core Viewpoint - The current mainstream solution for liquid cooling is the cold plate system, while immersion cooling is expected to become a future development direction. The maturity of domestic liquid cooling systems is increasing, and domestic supply chains are likely to enter the NVIDIA ecosystem directly as demand for AI server computing power surges, driving the penetration rate of liquid cooling solutions [1][2]. Group 1: Liquid Cooling Technology - Liquid cooling technology is essential for addressing the heat dissipation pressure in data centers, offering advantages such as low energy consumption, high heat dissipation, low noise, and low total cost of ownership (TCO). It can also reduce the Power Usage Effectiveness (PUE) of data centers, meeting national requirements [1]. - As chip iterations increase power density, the heat dissipation requirements for corresponding chips are growing, making traditional air cooling insufficient and necessitating the introduction of liquid cooling solutions. Currently, cold plate systems dominate the market, while immersion cooling is anticipated to be a future trend [1]. - The liquid cooling system consists of an outdoor side (primary side) and a data center side (secondary side), with the primary side accounting for approximately 30% of the value, including chillers and circulation pipelines, while the secondary side accounts for about 70%, with core components such as CDU, manifold, quick connectors, and water pumps [1]. Group 2: Liquid Cooling Industry - The value of liquid cooling is expected to increase alongside chip upgrades. For example, the value of rack liquid cooling modules is projected to grow by over 20% with the transition to GB300-GB200 servers. By 2026, the market for ASIC liquid cooling systems is estimated to reach 35.3 billion yuan, while the market for NVIDIA liquid cooling systems is expected to reach 69.7 billion yuan [2]. - Domestic supply chains are accelerating their entry into the liquid cooling market. NVIDIA has opened its supplier list, allowing manufacturers to choose their supply chain components independently. This shift from a single certified CDU supplier to multiple suppliers enables domestic chains to indirectly enter the market. As domestic liquid cooling systems mature and end-user CSPs focus more on cost-performance ratios, domestic suppliers are likely to enter NVIDIA's ecosystem directly [2]. Group 3: Rubin Architecture Outlook - Single-phase cold plates are unsuitable for the Rubin architecture due to its thermal design power (TDP) of 2300W and a total cabinet power of approximately 200KW, exceeding the design limit of single-phase cold plates at 150KW per cabinet. Therefore, new liquid cooling solutions must be introduced [3]. - One feasible solution is the phase change cold plate, which utilizes the phase change of a liquid working fluid to absorb heat efficiently, suitable for scenarios exceeding 300KW per cabinet. Another option is the microchannel cold plate (MLCP), which features densely packed micro-scale cooling liquid channels, potentially becoming the preferred choice for the Rubin architecture as future iterations may require even higher power capacities [3].

2025年11月27日，联想(00992)正式宣布将其Neptune海神液冷技术扩展至全线数据中心产品组合，以应 对现代数据中心日益增长的计算需求与能耗压力。 就在其宣布这则消息的同一天，美国芝加哥商品交易所(CME)的交易系因数据中心冷却故障而瘫痪，全 球衍生品市场陷入了一场长达逾10小时的交易暂停。 这场巧合背后，是AI算力狂飙下传统数据中心散热体系的全面承压，功耗暴增与能耗失控正将全球数 字基建推向"热失控"边缘，一场大规模的技术迭代迫在眉睫。 数据更能说明困境。国际能源署(IEA)统计显示，全球数据中心年耗电量已占总用电量的1.8%，其中 38%的电力被用于冷却系统——这意味着每产生1度电的计算价值，就有近0.4度电被消耗在"对抗热 量"上。微软首席执行官萨提亚 纳德拉(Satya Nadella)近期在 BG2播客中表示，尽管市场对AI芯片的需 求持续高涨，但微软当前面临的并非算力过剩，而是数据中心的供电和物理空间已接近极限，导致大量 AI芯片只能滞留在库存中，无法"通电运行"。 联想瞄准液冷市场 在风冷逼近物理极限的背景下，液冷技术正成为行业转型的核心方向，而全球数据中心建设的热潮预示 着这一市场 ...

智通财经APP获悉，2025年11月27日，联想(00992)正式宣布将其Neptune海神液冷技术扩展至全线数据 中心产品组合，以应对现代数据中心日益增长的计算需求与能耗压力。 就在其宣布这则消息的同一天，美国芝加哥商品交易所(CME)的交易系因数据中心冷却故障而瘫痪，全 球衍生品市场陷入了一场长达逾10小时的交易暂停。 这场巧合背后，是AI算力狂飙下传统数据中心散热体系的全面承压，功耗暴增与能耗失控正将全球数 字基建推向"热失控"边缘，一场大规模的技术迭代迫在眉睫。 "冷却"失灵的代价 此次故障因位于伊利诺伊州奥罗拉的数据中心突然触发高温保护机制，服务器集群自动关机。据CME 事后声明，该数据中心故障根源是一台冷水机组突发机械故障，导致机房温度在90分钟内从24℃飙升至 48℃，远超电子元件安全阈值。 造成的代价也相当惨重，数万亿美元合约因此受到影响——亚洲时段的CME期货电子盘完全停摆，欧 洲交易商被迫转向流动性更低的场外市场，标普500指数期货主力合约在恢复交易后短时间内波动幅度 扩大至2.3%，创2025年以来单日最大盘中震荡。 CME宕机事件并非孤例，这不仅是一次交易中断，更是对金融市场基础设施可 ...

2025年11月27日， 联想正式宣布将其 Neptune海神液冷技术扩展至全线数据中心产品组合，以应对现代 数据中心日益增长的计算需求与能耗压力。 就在其宣布这则消息的同一天，美国芝加哥商品交易所(CME)的交易系因数据中心冷却故障而瘫痪，全 球衍生品市场陷入了一场长达逾10小时的交易暂停。 数据更能说明困境。国际能源署(IEA)统计显示，全球数据中心年耗电量已占总用电量的1.8%，其中 38%的电力被用于冷却系统 —— 这意味着每产生1度电的计算价值，就有近0.4度电被消耗在 "对抗热 量" 上。微软首席执行官萨提亚・纳德拉(Satya Nadella)近期在 BG2 播客中表示，尽管市场对 AI 芯片的 需求持续高涨，但微软当前面临的并非算力过剩，而是数据中心的供电和物理空间已接近极限，导致大 量 AI 芯片只能滞留在库存中，无法"通电运行"。 联想瞄准液冷市场 这场巧合背后，是AI算力狂飙下传统数据中心散热体系的全面承压，功耗暴增与能耗失控正将全球数 字基建推向"热失控"边缘，一场大规模的技术迭代迫在眉睫。 "冷却"失灵的代价 此次故障因位于伊利诺伊州奥罗拉的数据中心突然触发高温保护机制，服务器集群自 ...

紫光股份在互动平台表示，公司全面布局液冷技术，可提供冷板式、浸没式、喷淋式等多种技术路线， 致力于打造技术先进、性能卓越的全栈液冷解决方案，并推出了面向AI计算等领域的液冷整机柜产品 系列。公司具有从冷板、快接头、CDU到机柜、监控系统的全链条自研与整合能力，同时拥有服务大 型互联网公司的液冷基础设施交付案例和经验。公司在液冷技术上始终坚定液冷全栈解决方案路线，同 时将全栈技术能力与大规模交付的工程化经验深度融合，形成高效、可靠、可复制的系统级能力，两者 相辅相成，共同构成公司的核心竞争力。 ...