行业研究丨深度报告丨通信设备Ⅲ [Table_Title] "液冷加速度"系列报告三： 液冷 0→1 后，从头部厂商表现再看行业变化 %% %% %% %% research.95579.com 1 丨证券研究报告丨 报告要点 [Table_Summary] 2025 年，是海外液冷 0→1 时刻，头部散热厂已实现液冷产品规模出货，拉动营收及盈利表现 亮眼。近期，多家散热大厂更新对外交流，我们整理了各家表述，以对液冷首年出货后，当前 的行业趋势和节奏判断做出参考。 分析师及联系人 [Table_Author] SAC：S0490517110002 SAC：S0490524100002 SFC：BUX641 请阅读最后评级说明和重要声明 2 / 20 %% %% %% %% research.95579.com 近期，各散热大厂均有相关更新对外交流，我们整理了各家表述，以对行业趋势和节奏判断做 出参考，可总结出几个主要关注点：1）2026 散热依然大有可为，头部散热大厂全面扩产，头 部厂商 AVC 指出订单能见度周期已开始拉长至 2028。2）2026 液冷放量主要动能仍来自于 NV 产品的全液冷需求。3）多家 ...

Schafer表示，云服务提供商的资本开支仍在持续上升，预计2026年全球云厂商资本开支将达到6500亿美元，明显高于2025年超过 4000亿美元的水平。同时，前沿大模型(LLM)规模仍以每年约10倍的速度增长，推理类token的需求增速也超过5倍，这进一步推高了对 高性能AI算力的需求。 在产品层面，Schafer指出，英伟达的机架级解决方案NVL72在单位功耗下的AI性能方面依然处于行业领先地位，而新一代Vera Rubin(VR200)平台正按计划推进，预计将在2026财年第三季度实现量产爬坡，随后更高端的VR300 Ultra有望在2027财年第三季度初 期推出。 Schafer进一步估算，Vera Rubin平台的平均售价有望比GB300高出40%至50%。作为参考，GB300单套售价约为350万美元。基于 此，Vera Rubin系列产品未来有望为英伟达带来约80亿美元的新增营收。 此外，随着中国市场重新纳入可服务范围，潜在可触达市场规模或高达500亿美元，这也可能进一步推升英伟达的总体可服务市场规模 (TAM)，目前已被估算至约4万亿美元。Schafer表示，从长期来看，英伟达仍是"最具通用性 ...

在产品层面，Schafer指出，英伟达的机架级解决方案NVL72在单位功耗下的AI性能方面依然处于行业 领先地位，而新一代Vera Rubin(VR200)平台正按计划推进，预计将在2026财年第三季度实现量产爬 坡，随后更高端的VR300 Ultra有望在2027财年第三季度初期推出。 Schafer表示，云服务提供商的资本开支仍在持续上升，预计2026年全球云厂商资本开支将达到6500亿 美元，明显高于2025年超过4000亿美元的水平。同时，前沿大模型(LLM)规模仍以每年约10倍的速度增 长，推理类token的需求增速也超过5倍，这进一步推高了对高性能AI算力的需求。 Schafer进一步估算，Vera Rubin平台的平均售价有望比GB300高出40%至50%。作为参考，GB300单套 售价约为350万美元。基于此，Vera Rubin系列产品未来有望为英伟达带来约80亿美元的新增营收。 智通财经APP获悉，美国芯片巨头英伟达(NVDA.US)即将公布第四财季业绩，投行奥本海默预计，公司 本次财报有望交出高于市场预期的成绩，相关营收上行空间或在20亿至30亿美元之间。Schafer重申对 英伟达的" ...

Group 1 - The core viewpoint of the article highlights that Nvidia's latest earnings report and shipment expectations for GB300/NVL72 are currently the main drivers of the market, with significant capital expenditure guidance igniting the global AI computing power supply chain [1] - The market is experiencing a shift in speculation from upstream chips to midstream manufacturing and downstream cooling/connection sectors, driven by the logic of "paying for certain incremental growth" [1] - The robotics sector is entering a critical phase of "distilling the genuine from the false," with companies in Tesla's core supply chain benefiting from a premium due to their association with strong industrialization entities [1] Group 2 - The market style is rapidly switching between "institutional investment in large-cap blue chips (such as AI and core T-chain)" and "retail speculation in thematic small caps (such as satellite communication and brain-computer interfaces)" [1] - Overall risk appetite is constrained by macro uncertainties (such as U.S. Treasury yields and tariffs), leading funds to favor sectors with strong fundamental data support (like capital expenditure and orders) [1] - The current market valuation is considered reasonably high, with an appropriate position of about 45% based on a composite of the Buffett and Graham indices [2] Group 3 - Recent data indicates a significant surge in the scale of the Hang Seng Technology Index ETF, which increased by 110.46 million in the past week, despite the overall lack of enthusiasm for the Hang Seng Technology sector [5] - The article notes that the Hong Kong stock market has potential for upward movement and certainty following the U.S. and A-share markets [4] - The article outlines various investment strategies, including value selection and valuation-based positioning, emphasizing a defensive approach in the current market environment [10][11]

Core Viewpoint - The evolution of AI large models towards trillion parameters, multi-modal capabilities, and intelligent agents is driving a transition in computing infrastructure towards "super-node" architecture, which significantly enhances training efficiency and inference throughput [1] Group 1: Technological Advancements - Traditional architectures are facing communication and energy consumption bottlenecks, necessitating a shift to super-node architecture [1] - NVL72 represents a solution that improves training efficiency and inference throughput through high bandwidth and low latency interconnections [1] Group 2: Market Opportunities - The demand for switching chips, optical modules, and high-speed connection modules is shifting from linear growth to exponential growth due to the architectural transformation [1] - There is significant room for improvement in domestic AI computing investments compared to overseas, making super-node architecture essential for catching up in domestic computing infrastructure [1] Group 3: Industry Recommendations - Companies are advised to focus on the value reassessment opportunities arising from increased interconnection density [1] - Recommended sectors include high-speed connection module manufacturers, switching interconnection manufacturers, optical module manufacturers, and AIDC and supporting manufacturers [1]

Core Insights - Bitdeer Technologies Group reported an increase in self-mining hashrate to 41.2 EH/s, surpassing its target of 40 EH/s, driven by the deployment of SEALMINER mining rigs [1][4][6] - The company mined 511 Bitcoins in October 2025, reflecting a 13% increase from September 2025 [4][7] - Bitdeer achieved an annual recurring revenue (ARR) of US$8 million from its AI cloud services, supported by strong customer demand for NVIDIA B200 systems [5][6] Mining Operations - The total proprietary hash rate deployed reached 41.3 EH/s in October 2025, up from 35.0 EH/s in September 2025 [2][7] - The company has 254,000 mining rigs under management, with 166,000 self-owned and 88,000 hosted [7] - The total hash rate under management increased to 55.5 EH/s, compared to 49.2 EH/s in September 2025 [7] SEALMINER Development - The SEALMINER A3 and A2 models are in final assembly, with the A3 model achieving a hashrate of 0.3 EH/s and the A2 model at 2.6 EH/s [2] - The first SEAL04 chip demonstrated power efficiency of approximately 6-7 J/TH, with mass production targeted for Q1 2026 [5][6] Infrastructure Updates - The company has completed construction of several data centers, including a 175 MW site in Tydal, Norway, and a 50 MW site in Oromia, Ethiopia, with 40 MW already energized [10][15] - Ongoing projects include a 221 MW site in Massillon, Ohio, expected to be fully energized by Q1 2026 [13][15] - The total global electrical capacity across all sites is 2,992 MW, with additional pipeline capacity of 1,381 MW [12][14] AI Cloud Services - Bitdeer deployed 584 GPUs with an 87% utilization rate, indicating strong demand for its AI cloud services [5][6] - The company is expanding its GPU infrastructure and has placed orders for NVIDIA's next-generation systems, expected to be delivered in December 2025 [5][6]

Summary of Key Points from the Investor Presentation on Greater China Technology Semiconductors Industry Overview - The focus is on the **Greater China Technology Semiconductors** industry, particularly in the context of **AI supply-chain updates** and **key opportunities in Asia** [1][2]. Core Insights and Arguments - **Investment Recommendations**: - **Overweight (OW)**: TSMC (Top Pick), Aspeed, Alchip, KYEC, ASE, FOCI, Himax, ASMPT, AllRing [11] - **Memory Stocks**: Winbond (Top Pick), GWC, Phison, Nanya Tech, APMemory, GigaDevice, Macronix [11] - **Underweight (EW/UW)**: MediaTek, UMC, ASMedia, Vanguard, WIN Semi [11] - **Market Dynamics**: - AI demand is expected to **reaccelerate** due to generative AI, impacting various verticals beyond the semiconductor industry [11]. - The **cannibalization effect** of AI on traditional semiconductor markets is noted, with a gradual recovery anticipated in the second half of 2025 [11]. - The **DeepSeek** technology is driving demand for AI inferencing, although concerns exist regarding the sufficiency of domestic GPU supply [11]. - **Long-term Demand Drivers**: - **Tech diffusion** and **tech deflation** are expected to stimulate demand for tech products, with a noted price elasticity effect [11]. Financial Metrics and Valuation Comparisons - **Valuation Metrics**: - TSMC's current price is **1,485.0 TWD** with a target of **1,688.0 TWD**, indicating a **14% upside** [12]. - UMC's current price is **44.9 TWD** with a target of **48.0 TWD**, indicating a **7% upside** [12]. - SMIC shows a significant downside with a target of **40.0 HKD**, representing a **-46% downside** [12]. - **Memory Sector Insights**: - Giga Device has a current price of **208.1 CNY** with a target of **255.0 CNY**, indicating a **23% upside** [12]. - Winbond's current price is **44.0 TWD** with a target of **50.0 TWD**, indicating a **14% upside** [12]. Additional Important Insights - **Market Trends**: - The semiconductor industry is experiencing a **prolonged downcycle** in mature node foundry and niche memory due to increased supply from China [11]. - The **historical correlation** between declining inventory days and rising semiconductor stock prices is highlighted, suggesting a potential positive outlook for the sector [11][68]. - **Future Projections**: - AI semiconductors are projected to account for approximately **34% of TSMC's revenue by 2027** [58]. - The **wafer demand** for TSMC's 2nm process is primarily driven by Apple, indicating strong customer reliance on TSMC for advanced technology [27]. - **Challenges**: - The **DDR4 shortage** is expected to persist into the second half of 2026, impacting supply dynamics [75]. - The **NAND flash market** is projected to face a double-digit percentage supply shortage, indicating ongoing supply chain challenges [75]. This summary encapsulates the critical insights and data points from the investor presentation, providing a comprehensive overview of the current state and future outlook of the Greater China Technology Semiconductors industry.

Core Insights - The article discusses the collaboration and solutions in the supernode field, highlighting the major players and their respective strategies in the market [3][4]. Supernode Collaboration and Solutions - Major CSP manufacturers are seeking customized server cabinet products from server suppliers, with a focus on NV solutions [4]. - Key supernode solutions in China include Tencent's ETH-X, NV's NVL72, Huawei's Ascend CM384, and Alibaba's Panjiu, which are either being promoted or have existing customers [4]. - ByteDance is planning an Ethernet innovation solution for large models, primarily based on Broadcom's Tomahawk, but it has not yet been promoted [4]. - Tencent's ETH-X collaborates with Broadcom and Amphenol, utilizing Tomahawk switches and PCIe switches for GPU traffic management [5]. - The main applications of these solutions differ: CM384 focuses on training and large model computation, while ETH-X is more inclined towards inference [5]. Market Share and Supplier Landscape - The supernode solutions have not yet captured a significant market share, with traditional AI servers dominated by Inspur, H3C, and others [6]. - From September 16, CSPs including BAT were restricted from purchasing NV compliant cards, leading to a shift towards domestic cards, which are expected to reach 30%-40% in the coming years [6]. - The overseas market share for major internet companies like Alibaba and Tencent remains small, with ByteDance's overseas to domestic ratio projected to improve [6]. Vendor Competition and Second-Tier Landscape - Inspur remains competitive in terms of cost and pricing, while the competition for second and third places among suppliers is less clear [8]. - The second-tier internet companies have smaller demands, and mainstream suppliers are not actively participating in this segment [9]. - The article notes that the domestic AI ecosystem is lagging behind international developments, with significant advancements expected by 2027 [9][10]. Procurement and Self-Developed Chips - Tencent and Alibaba have shown a preference for NV cards when available, with a current ratio of NV to domestic cards at 3:7 for Alibaba and 7:3 for ByteDance [10]. - The trend towards supernodes is driven by the need for increased computing power and reduced latency, with expectations for large-scale demand in the future [10]. Economic and Technical Aspects - The article highlights the profit margins for AI servers, with major manufacturers achieving higher gross margins compared to general servers [11]. - The introduction of software solutions is expected to enhance profitability, with significant profit increases anticipated from supernode implementations [11].

Core Viewpoint - The article provides a detailed comparison of Alibaba's super node with NVIDIA's NVL72 and Huawei's CM384, focusing on GPU count, interconnect technology, power consumption, and ecosystem compatibility. Group 1: GPU Count - Alibaba's super node, known as "Panjun," utilizes a configuration of 128 GPUs, with each of the 16 computing nodes containing 4 self-developed GPUs, totaling 16 x 4 x 2 = 128 GPUs [4] - In contrast, Huawei's CM384 includes 384 Ascend 910C chips, while NVIDIA's NVL72 consists of 72 GPUs [7] Group 2: Interconnect Technology - NVIDIA's NVL72 employs a cable tray interconnect method using NVLink proprietary protocol [8] - Huawei's CM384 also uses cable connections between multiple racks [10] - Alibaba's super node features an orthogonal interconnect without a backplane, allowing for direct connections between computing and switch nodes, reducing signal transmission loss [12][14] Group 3: Power and Optical Connections - NVIDIA's NVL72 uses copper for scale-up connections, while Huawei's CM384 employs optical interconnects, leading to higher costs and power consumption [15] - Alibaba's super node uses electrical interconnects for internal scale-up, with some connections made via PCB and copper cables, while optical interconnects are used between two ALink switches [18][19] Group 4: Parameter Comparison - Key performance metrics show that NVIDIA's GB200 NVL72 has a BF16 dense TFLOPS of 2,500, while Huawei's CM384 has 780, indicating a significant performance gap [21] - The HBM capacity for NVIDIA's GB200 is 192 GB compared to Huawei's 128 GB, and the scale-up bandwidth for NVIDIA is 7,200 Gb/s while Huawei's is 2,800 Gb/s [21] Group 5: Ecosystem Compatibility - Alibaba claims compatibility with multiple GPU/ASICs, provided they support the ALink protocol, which may pose challenges as major manufacturers are reluctant to adopt proprietary protocols [23] - Alibaba's GPUs are compatible with CUDA, providing a competitive advantage in the current market [24] Group 6: Supply Chain Insights - In the AI and general server integration market, Inspur holds a 33%-35% market share, while Huawei's share is 23% [33] - For liquid cooling, Haikang and Invec are key players, each holding 30%-40% of the market [35] - In the PCB sector, the number of layers has increased to 24-30, with low-loss materials making up over 60% of the composition, significantly increasing the value of single-card PCBs [36]

Core Insights - Intel announced a $5 billion investment in Nvidia, aiming to leverage both companies' strengths to develop custom data center and PC-related products [2] - Nvidia's CEO Jensen Huang highlighted the current limitations of x86 architecture products and the goal of integrating NVLink into Intel's data center CPUs to enable both Arm and x86 architecture offerings [2] - Huang acknowledged TSMC's significance in the semiconductor industry, indicating that while Intel has been a partner, TSMC remains a critical player for manufacturing capabilities [2] Group 1 - Intel's investment in Nvidia is approximately 355.31 billion RMB [2] - The collaboration focuses on introducing NVLink into Intel's data center CPUs [2] - Nvidia aims to create rack-level AI supercomputing by integrating x86 CPUs into the NVLink ecosystem [2] Group 2 - Huang emphasized that the x86 ecosystem currently cannot utilize NVL72 level products [2] - Both CEOs recognized TSMC as a world-class foundry and acknowledged their status as major clients [2] - The conversation between the companies indicates a productive partnership despite the manufacturing limitations at Intel [2]