股票简称：紫光股份 股票代码：000938 公告编号：2025－074 紫光股份有限公司 2025 年第六次临时股东会决议公告 本公司及董事会全体成员保证信息披露的内容真实、准确和完整，没有虚假记 载、误导性陈述或重大遗漏。 特别提示： 现场会议召开日期和时间：2025 年 12 月 19 日（星期五）下午 14 时 30 分 网络投票时间：2025 年 12 月 19 日。其中，通过深圳证券交易所交易系统进行投 票的具体时间为：2025 年 12 月 19 日上午 9:15-9:25，9:30-11:30 和下午 13:00-15:00 期间的任意时间；通过互联网投票系统进行网络投票的起止时间为 2025 年 12 月 19 日 9:15-15:00 期间的任意时间。 7、 会议出席情况：出席会议的股东及股东代表 1,756 名，代表股份数 890,173,199 股，占公司有表决权股份总数的 31.1241%。其中，出席现场会议的股东及股东代表 3 名，代表股份数 801,343,818 股，占公司有表决权股份总数的 28.0182%；参加网络 投票的股东 1,753 名，代表股份数 88,829,381 ...

Core Insights - The article discusses the launch of the H3C UniPoD S80000 super node product by Unisplendour Corporation's subsidiary, H3C, aimed at addressing the challenges of communication walls and computing power utilization in large model training and inference [1][5]. Group 1: Product Features and Innovations - The H3C UniPoD S80000 super node utilizes a "computing power × connectivity" technology concept, achieving full interconnection of GPUs through a Scale-up architecture, resulting in an 8-fold increase in inter-card bandwidth compared to traditional 8-card servers and an 80% improvement in single-card inference efficiency [1][5]. - The super node supports liquid cooling for high-density deployment and is compatible with multiple brands of GPUs, addressing the long-term stability requirements for large model training through software and hardware collaborative optimization [1][5][7]. Group 2: Market Context and Demand - As the market for high-performance computing power surges, driven by the increasing prevalence of high-parameter MoE large models like DeepSeek, the ability to efficiently train and infer large models becomes critical for gaining a competitive edge in the rapidly evolving AI landscape [2][3]. - The article highlights the importance of building robust and efficient AI computing infrastructure, with super node products emerging as a key focus in the current computing power sector [2][3]. Group 3: Technical Advantages - The article emphasizes that traditional cross-node communication methods lead to significant communication overhead, reducing computing power utilization. The Scale-up technology enables direct high-speed communication between GPUs, significantly enhancing GPU utilization and reducing idle time [3][4]. - In the inference phase, the super node's support for independent scaling of computing and storage resources allows for efficient resource allocation, particularly in scenarios requiring frequent access to KV Cache, thus minimizing resource waste and ensuring low latency [4][5]. Group 4: Stability and Reliability - The H3C UniPoD S80000 is designed with a focus on stability and maintainability, crucial for preventing training interruptions that could lead to resource waste and model performance degradation. The product incorporates collaborative optimization of software and hardware to ensure uninterrupted long-term training [7][8]. - The company is actively investing in optical interconnection technology to leverage the benefits of high speed, low latency, and low energy consumption while addressing the reliability issues associated with optical components [7][9]. Group 5: Future Outlook - H3C aims to continue developing super node products that support large-scale deployments of 1024 cards and above, enhancing the scale and efficiency of intelligent computing clusters [7][8]. - The company is committed to building a strong, diverse, and continuously evolving computing infrastructure to support the AI industry's growth and transformation [8][9].

Core Insights - The company is enhancing its product and solution capabilities through an "AI in ALL" strategy, focusing on building an integrated AI empowerment platform that includes a comprehensive digital infrastructure, cloud and intelligent platforms, and an AI training and application platform [2] Group 1: AI Empowerment Strategy - The company has adopted a "1+N" large model strategy to accelerate "AI for ALL" and is deeply engaging in industry-specific scenarios [2] - The intelligent platform has been adapted to over 20 types of GPUs and more than 60 large models, optimizing computing units to improve training and inference efficiency while lowering the barriers for enterprises to apply AI [2] Group 2: Industry-Specific Solutions - The company has launched the LinSeer ICT intelligent agent, which enhances the intelligence level of equipment inspection and operation and maintenance based on extensive ICT knowledge training [2] - A series of vertical industry AI solutions, including specialized models for medical, rail transportation, and education sectors, have been released and are already being applied in various scenarios [2] Group 3: Business Model and Services - The company provides one-stop computing power services through the "Turing Town" model, achieving a commercial closed loop from "construction, ecology, operation, empowerment" to "sales, service" [2] - The company will continue to strengthen its full-stack AI capabilities in both software and hardware to empower industry clients in their digital and intelligent transformation [2]

证券时报·数据宝统计，截至今日上午10:29，上证综指3876.98点，收于五日均线之上，涨跌幅0.17%， A股总成交额为7861.86亿元。到目前为止，今日有1647只A股价格突破了五日均线，其中乖离率较大的 个股有星辰科技、万隆光电、德艺文创等，乖离率分别为16.05%、13.66%、12.84%；拓新药业、紫光 股份、协昌科技等个股乖离率较小，刚刚站上五日均线。 | 300078 | 思创医 | 7.93 | 8.50 | 4.24 | 4.49 | 5.85 | | --- | --- | --- | --- | --- | --- | --- | | | 惠 | | | | | | | 002355 | 兴民智 通 | 10.04 | 8.41 | 5.92 | 6.25 | 5.54 | | 601566 | 九牧王 | 9.97 | 11.57 | 14.64 | 15.44 | 5.46 | 注：本文系新闻报道，不构成投资建议，股市有风险，投资需谨慎。 （文章来源：证券时报网） 12月18日突破五日均线个股乖离率排名 | 证券代 | 证券简 | 今日涨跌幅 | 今日换手率 | 五日均线 | ...

Investment Rating - The industry investment rating is "Buy" [3] Core Insights - The report highlights the potential of HBF (High Bandwidth Flash) technology to meet the memory capacity requirements of AI large models, which have been increasing due to the growth in model parameters and context lengths [13][20] - Companies with a database technology foundation are well-positioned to develop data infrastructure software based on HBF, with major players including Huawei and Alibaba, as well as independent firms like StarRing Technology and PingCAP [29][32] - The maturity of HBF technology is expected to drive the application of related data infrastructure software, particularly in AI inference tasks [33] Summary by Sections HBF Technology Development - HBF is anticipated to become the optimal solution for meeting the memory capacity demands of AI large models, as current HBM (High Bandwidth Memory) solutions are insufficient [20][23] - The first batch of HBF samples is expected to be delivered in the second half of 2026, with integrated AI inference hardware projected for early 2027 [23] Data Infrastructure Software Opportunities - The report identifies a growing need for data processing software that can handle the high memory requirements of various AI models, particularly large language models and video generation models [28][30] - Companies with existing database technology capabilities are likely to develop software optimized for HBF, enhancing their competitive edge in the market [29][32] Key Company Valuations and Financial Analysis - StarRing Technology (688031.SH) has a current price of CNY 97.99 with a rating of "Increase" [8] - Cambricon (688256.SH) is rated "Buy" with a target value of CNY 1480.92, reflecting strong growth potential [8] - Inspur Information (000977.SZ) is rated "Increase" with a current price of CNY 62.85 [8] - Unisplendour (000938.SZ) is rated "Buy" with a target value of CNY 39.18 [8]

Core Insights - The liquid cooling server industry is experiencing significant growth, driven by increased demand for high-density servers and efficient cooling solutions [1][2] - Major companies are actively investing in liquid cooling technologies, with substantial market opportunities projected in the coming years [2][3] Industry Developments - Xiechuang Data plans to invest up to 9 billion yuan to purchase servers from multiple suppliers, indicating accelerated infrastructure development in cloud computing [1] - The 5th International AIDC Liquid Cooling Supply Chain Summit will be held on December 18-19, focusing on technology exchange and market trends [1] Market Projections - UBS forecasts that the global direct liquid cooling market for data centers will grow from $1.138 billion in 2024 to $31.191 billion by 2030, with a compound annual growth rate of 51% [2] - The AI server liquid cooling market is expected to reach $23.746 billion, while the non-AI server liquid cooling market will grow to $7.446 billion [2] Company Highlights - Yingweike (002837) holds over 35% market share in liquid cooling technology and is a designated temperature control supplier for major clients like Tencent and Alibaba [2] - Shenling Environment (301018) is a core supplier for Huawei's data centers, with 63 liquid cooling patents and a PUE as low as 1.15 [2] - Highlan (300499) uniquely supplies both server and energy storage solutions, with a client base including ByteDance and Alibaba [3] Key Component Suppliers - Juhua (600160) is a leading domestic supplier of electronic-grade fluorinated liquids, aiming for a 15% global market share by 2025 [4] - Feirongda (300602) specializes in integrated cooling solutions and is a core supplier for Huawei servers [5] - Zhongshi Technology (300684) provides unique nano-carbon material cooling plates, with over 90% of its business in liquid cooling modules [5] System Integration Leaders - Inspur (000977) is a leading global liquid cooling server provider, with a projected market share of over 45% in 2024 [6] - Zhongke Shuguang (603019) collaborates on liquid cooling server development, introducing a high-density cooling solution for complex computing scenarios [6] - Unisplendour (000938) has a significant market presence with over 35% share in the operator market and expects liquid cooling orders to exceed 5 billion yuan in 2024 [6] Other Relevant Companies - Guangxun Technology (002281) is the only domestic company to achieve mass production of liquid cooling optical modules, holding a 15% global market share [7] - Changfei Optical Fiber (601869) applies optical fiber technology to liquid cooling, enhancing data transmission efficiency [8] - Ruijie Network (301165) offers a full range of liquid cooling products and has launched a new immersion liquid cooling switch [8]

Core Viewpoint - The strategic cooperation between Shanghai Data Exchange and Unisplendour Corporation's Xinhua San Group aims to address core challenges in the circulation of data elements, focusing on security, compliance, and ownership [1] Group 1 - The partnership will leverage Xinhua San's trusted data space solution to create a comprehensive system encompassing infrastructure, circulation mechanisms, and value realization [1] - The initiative is designed to provide practical technical support and ecological practices for the construction of a unified national data market [1] - The collaboration aims to facilitate efficient circulation and value release of data elements [1]

Core Viewpoint - Zhongyuan Securities report indicates that Unisplendour Corporation is actively acquiring the remaining equity of H3C, with overall fundamentals improving [1] Group 1: Acquisition and Ownership - The company will initiate a new round of equity acquisition for H3C, and after this acquisition, foreign investment will no longer hold shares in H3C [1] - As the company's ownership percentage in H3C increases, the contribution of H3C to the company's performance is continuously growing [1] Group 2: Growth Drivers - H3C's current growth drivers mainly come from domestic government and enterprise business as well as international business [1] Group 3: Financial Strategy - The company began the process of listing on the Hong Kong Stock Exchange in May, and successful listing is expected to significantly alleviate the financial cost pressure brought by the acquisition of H3C [1] - The company's server business and overseas operations are accelerating growth, while it is increasing its stake in H3C and conducting financing in Hong Kong stocks, indicating a positive overall development trend [1] Group 4: Investment Rating - The company is given an "Accumulate" investment rating [1]

Group 1 - The communication sector experienced a significant increase in valuation, with the communication index rising by 73.0% in 2025, ranking second among 31 primary industries in the Shenwan classification [1][15][24] - The growth in the communication industry is primarily driven by AI computing power, leading to accelerated technological iterations and scale growth across the entire AI computing power supply chain [1][15] - The report emphasizes the focus on the supply-demand imbalance in AI computing power and the evolution of new technologies, particularly the entry of low-orbit satellites into a prosperous cycle [1][2] Group 2 - AI computing power is entering a new stage, with structural changes expected in 2026, including a shift in demand from training to inference computing power [2][26] - The capital expenditure (Capex) investment structure is stabilizing among leading cloud providers, with second-tier cloud providers beginning to join the market [2][35] - The architecture of computing power networks is evolving, with scale-up and scale-across strategies contributing to increased network demand [2][42] Group 3 - The optical interconnection sector is facing supply-demand imbalances, particularly in optical modules, optical chips, and Faraday rotators, with a focus on technological trends such as 1.6T, silicon photonics, OCS, and CPO [3][51][54] - The liquid cooling market is at a historic turning point, with significant demand in North America, Southeast Asia, and China, particularly driven by the North American computing power market [3][11] - The IDC industry is expected to maintain an expansion trend driven by AI development, with an increase in AIDC penetration rates anticipated for 2026 [4][11] Group 4 - The report highlights the importance of breakthroughs in rocket capacity as a key variable for the commercial aerospace industry in China, with low-orbit satellite internet expected to enter an accelerated prosperity phase [1][11][24] - The commercial aerospace industry is a key focus area supported by national policies, with significant growth potential anticipated as satellite and computing power converge for space data processing [11][24][25] - The report suggests monitoring the success rates of various rocket models and their capacity planning as the industry matures [11][24]

AI Computing Power - AI computing power is entering a new stage, with a focus on structural changes by 2026, including a shift in demand from training to inference computing [2] - Capital expenditure (Capex) investment structure shows stability among leading cloud providers, with second-tier cloud providers joining the market [2] - The architecture of computing networks is evolving, with scale-up and scale-across strategies contributing to increased network demand [2] - Continuous upgrades in GPU technology and the rise of ASIC chips are accelerating the growth of domestic computing power [2] Optical Interconnection - The industry is facing a supply-demand imbalance, particularly in optical modules, optical chips, and Faraday rotators [3] - Key technological trends include 1.6T, silicon photonics, OCS, and CPO, with a focus on resolving supply-demand conflicts and the impact of technological upgrades on supplier dynamics and profitability by 2026 [3] - Major beneficiaries of technological upgrades in the optical module sector include leading companies like Zhongji Xuchuang and Xinyi Technology [3] - Supply constraints are noted for optical chip suppliers such as Yuanjie Technology and Shijia Photonics [3] Liquid Cooling - The server liquid cooling market is at a historic turning point, with significant demand in North America, Southeast Asia, and China, particularly in the North American computing market [4] - Current suppliers of liquid cooling solutions are primarily from Taiwan and the United States, with potential for domestic manufacturers to achieve breakthroughs [4] - Key focus on Invec [4] AIDC (Artificial Intelligence Data Center) - The IDC industry is expected to maintain an expansion trend driven by AI development, with an increase in AIDC penetration rates by 2026 [5] - Core focus areas include the international expansion of third-party IDC companies and the optimization of industry competition dynamics [5] - New REITs are anticipated to be released, enhancing companies' financing capabilities [5] Supernode Integrated Equipment - The supernode network requires high standards for complete cabinets, circuit boards, and connection chips, with strong supply chain dependencies [6] - Focus on connection chip suppliers, including switching chips and interface chips, as well as Ethernet switch manufacturers [6] Commercial Aerospace - The low Earth orbit satellite sector is entering an accelerated growth phase, supported by national policies and a strong upward trend in the commercial aerospace industry [7] - The successful launch of the Zhuque-3 rocket marks the beginning of a new era for China's commercial aerospace sector, which is expected to experience rapid growth [7] - Key variables include the success rate of various rocket payloads and capacity planning, with attention to satellite constellation bidding orders once capacity stabilizes [7]