2025年08月24日 证券研究报告 | 数据中心互联技术专题四： CSP云厂AI军备竞赛加速，智算中心架构快速发展 国信通信·行业专题报告 行业研究 · 行业专题 通信 投资评级：优于大市（维持评级） 证券分析师：袁文翀 联系人：赵屿 021-60375411 021-61761068 yuanwenchong@guosen.com.cn zhaoyu6@guosen.com.cn S0980523110003 请务必阅读正文之后的免责声明及其项下所有内容 投资摘要 CSP互联网云厂AI军备竞赛进入2.0时代，智算中心互联技术发展快速迭代。自2023年，ChatGPT3.5 点燃 "大模型革命"起，AI发展万众 瞩目，各大科技公司纷纷投入大模型研发并加大智算中心建设。根据CSP厂商的Capex指引，预计2025年，海外亚马逊、谷歌 、微软 、 Meta四家厂商合计Capex增至3610亿美元，同比增幅超58%；国内字节、腾讯、阿里Capex有望超过3600亿元。本轮AI浪潮前期，英伟达作为 AI芯片领军企业，其AI芯片供不应求；随着CSP云厂持续加大智算中心投入，具备更高性价比的自研ASIC算力芯片成为AI ...

证券研究报告 | 2025年08月23日 金融工程日报 沪指单边上行站上 3800 点，本周累计涨幅 3.49%创年内最佳 核心观点 金融工程日报 市场表现：20250822 市场全线上涨，规模指数中上证 50 指数表现较好，板 块指数中科创 50 指数表现较好，风格指数中中证 500 成长指数表现较好。 电子、通信、计算机、非银、国防军工行业表现较好，银行、石油石化、农 林牧渔、钢铁、交通运输行业表现较差。服务器、万得元宇宙、炒股软件、 AIPC、ASIC 芯片等概念表现较好，黄酒、宠物经济、家纺、西藏振兴、培 育钻石等概念表现较差。 市场情绪：20250822 收盘时有 81 只股票涨停，有 4 只股票跌停。昨日涨 停股票今日高开高走，收盘收益为 1.94%，昨日跌停股票今日收盘收益为 -0.57%。今日封板率 84%，较前日提升 25%，连板率 25%，较前日提升 13%， 封板率创近一个月新高。 市场资金流向：截至 20250821 两融余额为 21468 亿元，其中融资余额 21320 亿元，融券余额 148 亿元。两融余额占流通市值比重为 2.3%，两融 交易占市场成交额比重为 11.0%。 折溢 ...

Core Insights - The upgrade of NVIDIA's architecture is driving the development of liquid cooling, with the GB300 liquid cooling system covering over 80% of high-power components, and the Rubin architecture expected to achieve 100% liquid cooling by 2027 [1][4] - The ASIC cooling segment shows high gross margin potential, with a surge in ASIC chip shipments anticipated in 2026, as major tech companies like Google, Meta, and Amazon plan to release large quantities of ASIC chips [4][5] - The liquid cooling industry is expected to give rise to new industry leaders, with significant growth potential for Chinese companies in this sector due to their strong manufacturing and materials science foundations [5][6] Group 1: Liquid Cooling Development - The GB300 liquid cooling system utilizes a Direct-to-Chip Liquid Cooling (DLC) architecture, allowing for precise thermal conduction by cooling liquid through microchannel cold plates directly attached to high-power components [2][3] - The GPU liquid cooling market is projected to reach 80 billion yuan by 2026, driven by the demand for liquid cooling solutions in high-performance computing environments [3][4] - The first batch of GB300 shipments will still use the Bianca architecture, while future iterations will adopt independent liquid cooling plate designs to enhance cooling efficiency [4][38] Group 2: ASIC Chip Market - ASIC products are primarily designed and developed in collaboration with clients, leading to more flexible pricing and higher gross margins, making them a key focus for future industry growth [4][5] - Google has fully adopted liquid cooling solutions for its TPU clusters, achieving a GW-level operational scale with high availability [4][5] - Major companies are expected to release significant quantities of ASIC chips, with Google projected to ship 1.5 to 2 million TPUs by 2025, and Meta planning to release 1 to 1.5 million high-performance AI ASIC chips between 2025 and 2026 [4][5][56] Group 3: Industry Leaders and Opportunities - Companies like Qiyi Technology and Shuguang Data Creation are positioned to become leaders in the liquid cooling market, with successful deployments and innovative solutions [5][6] - The liquid cooling market is anticipated to expand significantly, with NVIDIA's GB200 and GB300 architectures driving increased adoption and market penetration [4][47] - The integration of liquid cooling solutions in data centers is expected to enhance overall system performance and efficiency, creating new opportunities for growth in the sector [5][6]

【交流纪要】 近期，PCB概念反复活跃，消息面上，AI服务器对PCB产品的性能要求日益提高，进一步推动了HDI市场需求的大幅增长。根据Prismark报告，预计在 2024至2029年间，HDI市场的年均复合增长率将达到6.4%，预计到2029年全球市场规模将增至170.37亿美元。同时，人气公司于周五盘后发布了业绩大增 的公告。生益电子(688183.SH)公告称，2025年上半年实现营业收入37.69亿元，同比增长91%；净利润5.31亿元，同比增长452%。公司拟每10股派发3元 现金红利，合计拟派发现金红利2.47亿元。 除此之外，PCB行业催化不断。此前，在近期的谷歌第二财季报告会中，公司披露其2025年Q2资本支出达224.46亿美元，环比增长30.5%，同比激增 70.2%，为历史新高。其中，约三分之二用于AI训练/推理所需的高端服务器（TPU/GPU），其余投向数据中心与全球骨干网络扩建。公司将全年资本开 支指引由此前的750亿美元上调至850亿美元，并预计2026年仍将维持高投入节奏。 国盛证券电子团队认为，随着GB200服务器下半年进入放量期，GB300出货也将接棒启动，谷歌、亚马逊与Me ...

Market Overview - The A-share market saw a strong performance on August 15, with major indices rising, including the Shanghai Composite Index up by 0.47% and the ChiNext Index up by 2.14% [1][2] - The total trading volume reached 1.33 trillion yuan, with a predicted total of 2.12 trillion yuan, indicating a decrease of 181.6 billion yuan [2] Brokerage Sector Performance - The brokerage sector experienced a significant inflow of nearly 12 billion yuan, with a strong increase of 3.57% in the sector index [2][4] - Notable performers included Changcheng Securities and Tianfeng Securities, both hitting the daily limit, while Zhongyin Securities and Dongfang Caifu rose nearly 7% [4][6] Fund Inflows and Stock Movements - The brokerage sector led the market with a net inflow of 11.979 billion yuan, followed by the electric power and industrial machinery sectors [3] - The overall market saw 4,460 stocks rise, with the brokerage sector being a key contributor to this upward movement [2][3] AI Industry Activity - The AI industry chain stocks showed strong activity, with companies like Chuanhuan Technology and Jintian Co. hitting the daily limit, reflecting ongoing advancements in AI technology [8][9] - The release of significant AI models, such as OpenAI's GPT-5, is expected to drive demand for computing power and related components, benefiting the ASIC industry [9] Company Earnings Forecasts - Changcheng Securities projected a net profit of 1.335 to 1.407 billion yuan for the first half of 2025, representing a year-on-year growth of 85% to 95% [6][7] - The overall performance of listed brokerages is expected to show significant improvement due to a recovering capital market and increased trading activity [7]

Market Overview - The A-share market showed a mixed performance today, with major indices mostly closing higher. The Shanghai Composite Index rose by 0.50% to close at 3665.92 points, while the Shenzhen Component increased by 0.53% to 11351.63 points. The Sci-Tech 50 index saw a significant rise of 1.91%, closing at 1069.81 points, and the ChiNext index gained 1.24%, ending at 2409.40 points. Overall, there were 3162 stocks that declined, while 2083 stocks advanced, with 172 stocks remaining flat. The total trading volume in the Shanghai and Shenzhen markets reached 18815 billion yuan, an increase of 545.47 billion yuan compared to the previous trading day [1][2]. Industry Performance - Among the primary industries, telecommunications, electronics, coal, household appliances, real estate, and non-bank financials led the gains, while sectors such as defense, steel, building materials, food and beverage, non-ferrous metals, and social services experienced declines [1][2]. Sector Highlights - The top-performing concept sectors today included optical modules (CPO), GPU, optical chips, servers, photolithography machines, ASIC chips, AI computing power, industrial gases, East Data West Computing, optical communications, semiconductor materials, stock trading software, semiconductor industry, consumer electronics OEM, and the National Big Fund [2].

Market Overview - The A-share market saw most major indices decline today, with the Shanghai Composite Index falling by 0.33% to close at 3593.66 points, and the Shenzhen Component Index decreasing by 0.22% to 11168.14 points. The ChiNext Index dropped by 0.23% to 2340.06 points, while the Sci-Tech 50 Index increased by 2.07% to 1054.20 points. Overall, 2724 stocks declined, 2532 stocks rose, and 158 stocks remained flat, with total trading volume in the Shanghai and Shenzhen markets amounting to 12189 billion yuan, a decrease of 6258.16 billion yuan from the previous trading day [1][2]. Industry Performance - Among the primary industries, electronic, computer, real estate, light manufacturing, textile and apparel, and media sectors led the gains, while construction decoration, building materials, food and beverage, coal, comprehensive, and steel industries experienced declines [1][2]. Sector Highlights - The top-performing concept sectors today included GPU, Kimi, multimodal models, ChatGPT, photolithography machines, intelligent agents, servers, selected rare metals, AIGC, artificial intelligence, machine vision, ASIC chips, selected semiconductors, Xiaohongshu platform, and Pinduoduo partners [2].

Core Insights - Nvidia reached a historic market capitalization of $3.92 trillion, surpassing Apple's previous record, driven by advancements in AI computing power [1] - The new Blackwell platform is expected to significantly boost Nvidia's data center revenue, contributing nearly 70% in Q1 2025 [1] - However, Nvidia faces challenges from key clients like OpenAI, which is testing Google's TPU chips for its products, indicating a potential shift away from Nvidia's GPUs [3][7] Group 1: Market Position and Competition - Nvidia's dominance is threatened by the rise of ASIC chips, which are being developed by major cloud players like Google, Amazon, and Meta, as well as startups like Cerebras and Graphcore [6][11] - The UALink alliance, formed by AMD, Intel, Google, Meta, Microsoft, and AWS, aims to create an open interconnect standard that could reduce reliance on Nvidia's GPU technology [6][12] - Google's TPU series, particularly the seventh generation Ironwood, is designed for inference tasks and is reported to outperform Nvidia's Blackwell in terms of performance per watt [7][9] Group 2: Dependency and Pricing Challenges - Nvidia's revenue is heavily reliant on a few large clients, with approximately 88% coming from data center operations, raising concerns about over-dependence [12] - The high cost of Nvidia's latest products, such as the $3 million GB200 NVL72 server, may deter potential customers, pushing them to seek more cost-effective alternatives [13][16] - As AI inference becomes mainstream, the demand for cost-effectiveness is increasing, favoring specialized ASIC chips over Nvidia's high-performance GPUs [16] Group 3: Ecosystem and Strategic Responses - Nvidia's NVLink Fusion architecture aims to create a more open ecosystem, but it still requires third-party chips to connect to Nvidia's products, indicating a defensive strategy [4][11] - The rise of open ecosystems like UALink and alternatives to Nvidia's CUDA programming environment may limit Nvidia's ability to maintain its competitive edge [12][17] - Despite being a leader in AI chip technology, Nvidia's position is increasingly challenged by a decentralized market driven by customer preferences for flexibility and cost [17]

Core Viewpoint - 嘉楠科技 is terminating its non-core AI semiconductor business to focus on its core operations, which include Bitcoin mining machine sales, self-mining, and consumer mining products. This strategic shift is expected to significantly reduce operational costs in the coming months [1][2]. Group 1: Business Strategy - The decision to end the AI semiconductor business is primarily due to its low revenue of approximately $900,000 for FY 2024, which accounts for 15% of the company's total operational expenses, thus dragging down overall efficiency [2]. - Post-strategic adjustment, resources will be concentrated on mining equipment sales, self-mining, and consumer-grade mining machines, where the company has established a clear competitive advantage [2]. Group 2: Mining Operations - 嘉楠科技's mining business is set to launch the Avalon Q mining machine in 2025, which will be compatible with global voltage standards, and the next-generation A16 series is expected to achieve an energy efficiency ratio of 15J/TH, nearing industry-leading levels [2]. - The company has a global hash rate of 8.15 EH/s in North America, with a low average electricity cost of $0.044 per kWh, and overseas projects in Ethiopia are operating with over 95% uptime [2]. Group 3: Financial Performance - In Q4 FY 2024, 嘉楠科技 reported a 33.2% reduction in net losses and achieved positive EBITDA, indicating improved financial health [2]. - The Avalon miner series, particularly the A14 product line, saw significant revenue growth following the Bitcoin halving event in April 2024, with demand for the A15 and upcoming A16 series continuing to rise [3]. Group 4: Company Milestones - 嘉楠科技 was founded in 2013 and released the world's first blockchain computing device based on ASIC chips, marking the beginning of the ASIC era in the industry [4]. - The company underwent a name change in 2015 and achieved significant milestones in chip production, including the mass production of 28nm and 16nm chips [5][6].

Core Viewpoint - The AI ASIC industry is experiencing significant growth opportunities due to its advantages over general-purpose chips like GPUs and CPUs in terms of computing power, efficiency, and cost per unit of computing power, which are driving demand growth [1][2]. Group 1: Market Dynamics - The ASIC market is catalyzed by its ability to adapt to various business scenarios and customer needs, providing flexibility in business layout [1]. - Major overseas CSP companies are launching self-developed chips, with Google releasing its latest TPU chip, Trillium, which has more than doubled its computing power compared to the previous generation [2]. - AWS has introduced the Trainium series, with the latest Trainium2 chip's performance being four times that of its predecessor, showcasing advancements in AI chip technology [2]. Group 2: Company Performance - Leading AI ASIC design companies like Broadcom and Marvell are benefiting from the surge in AI demand, with Broadcom projecting a 220% year-on-year increase in AI business revenue to $12.2 billion for fiscal year 2024 [3]. - Broadcom anticipates that its AI business revenue will reach between $60 billion to $90 billion by 2027, indicating strong market confidence [3]. - Marvell expects its AI business revenue to grow from $550 million to $2.5 billion between fiscal years 2024 and 2026, reflecting accelerated growth [3].