Core Insights - The article emphasizes the growing demand for switch chips driven by the expansion of AI chip Scale up clusters, with significant opportunities for domestic manufacturers in the high-speed chip market [1][2]. Group 1: Market Trends - The total scale of China's switch chips is projected to reach 257 billion, 356 billion, and 475 billion yuan in 2025, 2026, and 2027 respectively, with year-on-year growth rates of 61%, 39%, and 33% [1]. - The current low domestic production rate of switch chips, especially in the high-end market, presents a substantial opportunity for domestic manufacturers to increase market share as they continue to make breakthroughs in high-speed rates [1]. Group 2: Technological Developments - The evolution of large language models (LLMs) is leading to the development of various strategies such as data parallelism and tensor parallelism to address the challenges of model size, necessitating high bandwidth and low latency networks [1]. - The Scale up network's bandwidth significantly exceeds that of Scale out networks, making it the preferred technical solution in the industry [1]. Group 3: Domestic Innovations - Domestic AI companies are launching their own super-node products equipped with Scale up switch nodes to enhance communication efficiency [2]. - Companies like Huawei, Baidu, and ZTE are developing super-node solutions that support high interconnectivity, with ZTE's super-node server achieving GPU communication bandwidth of 400GB/s to 1.6T/s [2]. Group 4: Protocol Landscape - In the Scale up switch domain, Ethernet, PCIe, and private protocols (such as NVLink and UB) are expected to coexist, while Ethernet is likely to dominate the Scale out domain due to its open ecosystem and cost advantages [3].

Group 1 - The core viewpoint is that domestic manufacturers are expected to gradually increase their market share in high-end switching chips due to continuous breakthroughs and increased overall AI spending, with projected market sizes for 2025, 2026, and 2027 being 257 billion, 356 billion, and 475 billion yuan respectively, representing year-on-year growth rates of 61%, 39%, and 33% [1] - The current overall domestic substitution rate of switching chips is low, especially in the high-end chip market, where companies like Broadcom, Marvell, and NVIDIA dominate, indicating significant room for domestic chip replacement [1] Group 2 - The evolution of large models and the expansion of Scale up clusters are identified as important trends, with large language model parameters evolving from hundreds of billions to trillions and beyond, employing various strategies to address the limitations of model size [2] - The communication requirements for tensor and expert parallelism are stringent, making high-bandwidth, low-latency Scale up networks the mainstream technical solution in the industry [2] Group 3 - The ongoing upgrade of overseas AI chips to Scale up sizes is driving new demand for switching chips, with current GPU Scale up interconnects reaching dozens of cards and evolving towards hundreds, while AI custom chip interconnects are expanding from dozens to thousands [3] - Domestic AI companies are launching their own supernode products equipped with Scale up switching nodes, with Huawei's Ascend supporting interconnects of 384 chips and Baidu's Kunlun supporting 32/64 card interconnects [3] - Various domestic manufacturers, including ZTE and H3C, are providing foundational engineering capabilities for domestic chips to transition to supernodes, with ZTE's supernode server achieving GPU communication bandwidths of 400GB/s to 1.6T/s [3] - In the Scale up switching domain, Ethernet, PCIe, and private protocols (such as NVLink and UB) are expected to coexist, while Ethernet is anticipated to dominate the Scale out domain due to its open ecosystem and cost advantages [3]

Core Viewpoint - The article emphasizes the significant growth potential in China's switch chip market driven by increased AI spending and the trend towards Scale up technology, with projected market sizes of 257 billion, 356 billion, and 475 billion yuan for the years 2025, 2026, and 2027 respectively, reflecting year-on-year growth rates of 61%, 39%, and 33% [1] Group 1: Market Dynamics - The overall domestic localization rate of switch chips is currently low, particularly in the high-end chip market, where companies like Broadcom, Marvell, and NVIDIA dominate, indicating a substantial space for domestic manufacturers to replace imports [1] - The continuous evolution of large models and the expansion of Scale up clusters are identified as key trends, with large language model (LLM) parameters evolving from hundreds of billions to trillions and beyond, necessitating advanced strategies to address the challenges of model size [1] Group 2: Technological Developments - The article notes that the Scale up network bandwidth significantly exceeds that of Scale out networks, making it the mainstream technical solution for the industry, particularly in addressing the high bandwidth and low latency requirements of tensor and expert parallelism [1] - Overseas AI chips are upgrading their Scale up capabilities, with current GPU interconnects reaching dozens of cards and evolving towards hundreds, while AI custom chips are moving from dozens to thousands of cards, leading to increased demand for Scale up switch chips to enhance communication efficiency [2] Group 3: Domestic Innovations - Domestic AI companies have launched their own super node products equipped with Scale up switch nodes, with notable examples including Huawei's Ascend achieving interconnects of 384 chips and Baidu's Kunlun supporting 32/64 card interconnects [2] - Companies like ZTE, H3C, and Super Fusion have also introduced their super node solutions, providing foundational engineering capabilities for domestic chips to transition to super nodes, with ZTE's super node server supporting GPU communication bandwidths ranging from 400GB/s to 1.6T/s [2] Group 4: Protocol Perspectives - In the Scale up switch domain, Ethernet, PCIe, and proprietary protocols (such as NVLink and UB) are expected to coexist, while in the Scale out domain, Ethernet is anticipated to dominate due to its open ecosystem and cost advantages, with InfiniBand capturing a portion of the market [2]

Group 1 - The core viewpoint of the article is that New H3C Group, a subsidiary of New Ziguang Group, has launched a storage server called UniServer R3350G7, aimed at establishing a data foundation for the AI era [1] Group 2 - The UniServer R3350G7 is designed to meet the growing demands of data storage and processing in the context of artificial intelligence [1]

Investment Rating - The report maintains a "Positive" investment rating for the industry [12] Core Insights - The cooling methods for data centers are transitioning from air cooling to liquid cooling due to the increasing power consumption of chips and single cabinets. Liquid cooling technologies offer significant advantages such as higher cooling efficiency, lower energy consumption, and reduced total cost of ownership (TCO) [6][19] - The report highlights the growth of liquid cooling solutions, particularly single-phase cold plate cooling, which currently accounts for over 90% of liquid cooling applications in data centers. The technology is maturing, and small-scale commercial applications are being promoted [6][19] - The report suggests that various cooling media will emerge as opportunities in the liquid cooling market, with a focus on companies involved in the production of these cooling liquids [10][52] Summary by Sections Data Center Cooling Transition - The report discusses the shift to liquid cooling as a response to the rising power density of chips, particularly in AI servers, which are pushing the limits of traditional air cooling methods. The maximum power density for traditional air-cooled cabinets is around 20 kW, while future designs may reach up to 600 kW [7][25][29] Cooling Media Opportunities - Different cooling media are being explored for single-phase and two-phase cooling solutions. For single-phase cooling, common media include ethylene glycol solutions and propylene glycol solutions, while two-phase solutions primarily use refrigerants and low-boiling point fluids [8][41] - The report emphasizes the need for cooling liquids that meet specific technical requirements, such as excellent insulation properties, low surface tension, and high thermal conductivity [9][42] Investment Recommendations - The report recommends focusing on companies involved in the liquid cooling media industry, including: - Juhua Co., Ltd. (fluorinated liquids + refrigerants) - Sinoma Science & Technology (fluorinated liquids) - Yonghe Technology (fluorinated liquids + refrigerants) - Haohua Technology (fluorinated liquids + refrigerants) - Sanmei Co., Ltd. (refrigerants) - Xin'an Chemical (silicone oil) - Xingfa Group (silicone oil) [10][52]

Core Viewpoint - ZTE Corporation (00763) continues its strong performance, with A-shares hitting the daily limit and H-shares rising over 14% during trading, reflecting positive market sentiment towards the company and its positioning in the AI sector [1] Company Performance - As of the latest update, H-shares of ZTE are up 7.34%, priced at 31.9 HKD, with a trading volume of 2.041 billion HKD [1] Industry Developments - On July 26, Huawei showcased its Ascend 384 Super Node at the 2025 World Artificial Intelligence Conference, which features a high-bandwidth, low-latency interconnection among 384 NPUs, achieving a single-card inference throughput of 2300 Tokens/s, surpassing NVIDIA's H100/H800 [1] - Other major players like Tencent and Alibaba are promoting their respective ETH-X and ALS solutions, while companies such as Inspur, ZTE, H3C, Huqin, and Super Fusion are also accelerating the deployment of super node solutions [1] Market Insights - Zheshang Securities notes that the progress of domestic super nodes has exceeded expectations, with the core growth area being the switching network [1] - The penetration rate of domestic super nodes is only one year behind that of overseas markets, indicating a strong growth potential [1] - The ScaleUp switching network is identified as a critical growth segment for super node solutions, with expectations for the domestic ScaleUp market size to more than double [1] - ZTE is positioned at the core of the domestic super node solution market, with a significant gap in market perception [1] - The accelerating demand for ScaleUp switching is driving the localization of switching chips [1]

Core Insights - The Ninth Future Network Development Conference was held in Nanjing, focusing on new network communication technologies driven by low-altitude smart networks, intelligent cloud networks, and cybersecurity [1][3] - The conference showcased three major technological achievements from the Zijinshan Laboratory, including an integrated scheduling platform, a network large model with multi-agent systems, and an AI-based design toolchain for baseband circuits, which are expected to pave new paths for future network development in China [1][4] Group 1: Conference Overview - The conference was organized by the Nanjing Municipal Government and aimed to enhance technical exchanges and industry cooperation among experts, researchers, and telecom operators [1][3] - This year marks the tenth anniversary of the Future Network Experimental Facility (CENI) in Jiangsu, which serves as a significant national infrastructure for communication and information technology [3] Group 2: Technological Innovations - The integrated scheduling platform enables nationwide cross-domain mobilization and distribution of data, computing power, and models, addressing challenges in the "East Data West Computing" project [4] - The network large model and multi-agent system enhance decision-making capabilities in large-scale heterogeneous networks, achieving significant efficiency improvements over traditional methods [4] - The AI-based design toolchain for baseband circuits addresses the need for customized designs in diverse 5G/6G scenarios, streamlining the design process [5] Group 3: Industry Collaboration and Future Directions - Nanjing aims to become a leading city in 6G development and has established partnerships with major internet companies to foster innovation in the future network sector [8][9] - A strategic cooperation agreement was signed between Nanjing and Lhasa to promote the "East Data West Computing" initiative, forming a collaborative ecosystem of over 50 computing power enterprises [9] - The city is committed to enhancing its role as a global hub for future network innovation, focusing on integrating technological and industrial advancements [9]

Core Insights - Qianhe Capital recently conducted research on a listed company, focusing on its core business and expansion into new products and clients [1] Company Overview - The company, Feiling Kesi, is concentrating on its main business while actively exploring new business opportunities and products [1] - Major clients include Xinhua San, S Client, and Ruijie Networks, with the top five clients accounting for 96.77% of the company's revenue [1] Financial Performance - Data center switch sales increased by 67% year-on-year, and multiple new product developments have been initiated [1] - Research and development investment amounted to 71.89 million, representing 9.96% of total revenue [1] - Overseas market sales reached 15.18 million, primarily through collaborations with clients in Japan and South Korea [1] - The production base in Haining, Zhejiang, generated revenue of 171 million, with a net profit of 2.86 million [1] Strategic Focus - The company has completed domestic replacement solutions for key products, covering both hardware and software [1] - It aims to enhance its internal capabilities and seize opportunities in the current market environment [1]

Core Viewpoint - The demand for cooling solutions is increasing due to the rising power density of AI chips and servers, with liquid cooling expected to replace air cooling as the mainstream solution by 2026, reaching a market size of $8.6 billion globally [2][3]. Group 1: Liquid Cooling Market Dynamics - The global AI liquid cooling market is projected to grow rapidly, with a forecasted market size of $8.6 billion by 2026, driven by the increasing power density of AI chips and servers [2][3]. - The power density of AI servers is expected to exceed 1MW per cabinet by 2029, highlighting the need for efficient cooling solutions [2][3]. - The liquid cooling market in China is anticipated to grow by 67% year-on-year in 2024, reaching $2.37 billion, with a CAGR of 46.8% from 2024 to 2029 [3][42]. Group 2: Advantages of Liquid Cooling - Liquid cooling systems offer higher cooling efficiency and reduced system power consumption compared to traditional air cooling, making them a preferred choice as power density increases [2][22]. - The cooling efficiency of liquid cooling is significantly higher than that of air cooling, with water's thermal conductivity being over 23 times greater than that of air [22]. - Liquid cooling allows for higher density deployment of computing units, reducing the physical space required and improving overall system performance [22][25]. Group 3: Industry Trends and Developments - Major cloud service providers are significantly increasing their capital expenditures, with the top four North American cloud companies expected to reach $366.1 billion in total capital expenditures by 2025, reflecting strong demand for AI infrastructure [7]. - Companies like Alibaba and ByteDance are ramping up investments in AI and cloud computing infrastructure, with Alibaba planning to invest over $50 billion in these areas over the next three years [8]. - The liquid cooling technology is being rapidly adopted by leading server manufacturers, with companies like Supermicro and HPE launching liquid-cooled server solutions [41][43]. Group 4: Liquid Cooling Technology Types - Liquid cooling can be categorized into indirect contact (cold plate cooling) and direct contact (immersion and spray cooling), with cold plate cooling being the most mature and widely adopted technology [3][30]. - Cold plate cooling systems are expected to lead the market due to their minimal impact on existing infrastructure and superior cooling performance compared to air cooling [3][30]. - Immersion cooling, while offering the best cooling performance, presents higher technical challenges and is still in the exploratory phase [3][30]. Group 5: Energy Efficiency and Cost Savings - Liquid cooling systems can significantly reduce energy costs in data centers, with potential savings in operational costs due to lower energy consumption compared to air cooling systems [25][29]. - The total cost of ownership (TCO) for liquid cooling solutions is expected to be lower than that of air cooling, as they can reduce the need for additional cooling equipment [25][29]. - As data center energy efficiency becomes a regulatory focus, liquid cooling solutions are positioned to meet stringent PUE requirements, enhancing overall energy utilization [17][20].

Market Overview - On August 20, the Shanghai Composite Index rose by 1.04%, the CSI 300 increased by 1.14%, the STAR Market 50 surged by 3.23%, the CSI 1000 climbed by 0.86%, the ChiNext Index grew by 0.23%, and the Hang Seng Index went up by 0.17% [4][6] - The best-performing sectors on August 20 were Beauty Care (+2.42%), Oil & Petrochemicals (+2.36%), Electronics (+2.32%), Automotive (+1.93%), and Food & Beverage (+1.39%). The worst-performing sectors were Pharmaceuticals & Biotechnology (-0.07%), Household Appliances (+0.12%), Real Estate (+0.16%), Environmental Protection (+0.2%), and Construction Decoration (+0.2%) [4][6] - The total trading volume of the A-share market on August 20 was 24,484.14 billion yuan, with a net outflow of 14.682 billion Hong Kong dollars from southbound funds [4][6] Important Recommendations - The report highlights Dou Shen Education (300010) as the first education company to implement AI virtual teachers across its entire product line, aiming to capture market share and establish a leading brand in the AI education era [7] - The company is expected to achieve significant market penetration due to its first-mover advantage and rapid iteration of AI education products, with a potential market space reaching hundreds of billions [7] - Revenue projections for Dou Shen Education are estimated at 1,228.96 million yuan in 2025, 2,047.78 million yuan in 2026, and 3,573.90 million yuan in 2027, with corresponding growth rates of 62.38%, 66.63%, and 74.53% [7] Key Insights - The macroeconomic report indicates a high probability of a "residential deposit migration" trend, similar to past occurrences in 2009 and 2014-2015, driven by factors such as declining deposit rates and a shift to a more accommodative monetary policy [8] - The bond market analysis suggests a long-term bullish outlook supported by economic recovery and a tightening credit environment, while short-term risks remain due to fluctuating risk preferences [9][10] - The communication industry report emphasizes the significance of supernode solutions in enhancing AI computing efficiency, with major companies like Huawei and ZTE accelerating their development [11][12] - The chemical industry report notes a surge in demand for fluorinated liquids driven by the growth of AI computing investments and the increasing adoption of immersion cooling technologies [13]