Investment Rating - The report maintains a "Recommended" rating for the computer industry [1][44] Core Insights - The supernode has become the new norm for AI infrastructure, characterized by its technical complexity and rapid iteration, which builds a wide moat for OEM manufacturers and drives their profitability [6][44] - Domestic CSP capital expenditure outlook is optimistic, with significant growth in capacity and orders for wafer fabs and computing rental companies [7][35] - The report highlights the increasing demand for AI capabilities in China, with domestic models surpassing U.S. models in usage [7][35] Summary by Sections Recent Performance - The computer industry has shown a relative performance of -13.7% over 1 month, -5.5% over 3 months, and +2.7% over 12 months, compared to the CSI 300 index which is at -4.6%, -3.4%, and +14.7% respectively [4] Investment Highlights - Supernodes are designed for building large-scale AI computing clusters, integrating multiple GPUs or AI chips into a unified system for high bandwidth and low latency [6][10] - The supernode architecture is not merely hardware assembly but a cohesive system that allows for collaborative computing, enhancing efficiency significantly [10][15] - Major companies like NVIDIA, AMD, Huawei, and Sugon are continuously launching related products, indicating a robust market for supernodes [19][30] Domestic CSP AI Capital Expenditure Outlook - The overall capital expenditure for computing power in China is in a catch-up phase, with optimistic projections for 2026 [7][35] - Demand-side advantages include a large user base and diverse application scenarios, with domestic models leading in usage [35][40] Complexity and Profitability of Supernode Solutions - Supernodes offer advantages over traditional GPU clusters in terms of communication latency, computing density, and total cost of ownership [8][41] - The high technical complexity and rapid iteration of supernode systems create a significant barrier to entry, enhancing the profitability of capable OEM manufacturers [41][42] Investment Strategy - The report suggests that supernode OEM manufacturers will be the primary beneficiaries in the context of optimistic capital expenditure outlooks and the international expansion of domestic tokens [44] - Key companies mentioned include Sugon, Inspur, and Huawei in the server/supernode OEM space, as well as various AI chip and cloud computing firms [44]

中金研究 Abstract 摘要 需求：端云共振，共筑AI硬件发展美好蓝图。 我们认为，在云侧，在中国AI大模型综合实力逐渐接近全球领先、Agent应用的普及放大tokens消耗、以及 模型出海等因素推动下，中国云端AI算力硬件市场需求有望迎来快速成长；在端侧，中国AI小模型智能水平已进入可商用区，OpenClaw的"火出圈"也证 明Agent能力可以开始嵌入本地终端，端侧AI硬件产业链布局深厚的中国企业有望在未来3-5年的创新周期中受益。同时，结合政企端场景偏好本地化方 案、消费者端用户基数全球领先的特征，我们认为本土市场有望继续反哺AI端侧硬件需求，飞轮效应凸显。 供给：云侧AI硬件全链条技术攻关，算力网络规划清晰；端侧AI硬件产业链布局完备。 在云侧我们看到，当下AI芯片、服务器集群开始缩小与海外领先 水平差距，新型计算架构的产品不断涌现，供应链技术瓶颈问题也正在逐步被解决。伴随着算力资源从国家层面的统筹规划进一步清晰，及"东数西算"工 程深入推进，我们认为国产AI算力硬件系统性竞争力有望进一步增强。端侧来看，中国拥有最完整的端侧AI产业链，新技术一旦传导至AI端侧硬件规格 侧，易形成供给侧快速响应与需 ...

Core Viewpoint - The article emphasizes the transformation in the domestic semiconductor industry, particularly focusing on the launch of the scaleX40 super node by Zhongke Shuguang, which aims to bridge the gap between small servers and large super nodes, making high-end computing more accessible to enterprise users [2][4]. Group 1: Product Features and Market Positioning - The scaleX40 super node is designed with 40 AI acceleration cards, providing over 28 PFLOPS (FP8) of total computing power and more than 5TB of HBM memory, targeting the "sweet spot" of computing power for small and medium-sized enterprises [4][5]. - The product features a wireless cable box design that eliminates the need for complex cabling, reducing deployment difficulties and operational burdens, thus making it easier for companies with limited IT capabilities to deploy [5][6]. - The scaleX40 is based on a standard 19-inch box design, compatible with mainstream cabinets, which allows for standardized product delivery and significantly accelerates the equipment deployment cycle [5][6]. Group 2: System-Level Efficiency and User Value - The product integrates a storage system optimized for AI and a comprehensive development, scheduling, and operation platform, addressing the issue of insufficient storage capacity while enhancing training and inference efficiency [6]. - The shift in focus from merely competing on parameters to system-level efficiency marks a new phase for domestic super nodes, emphasizing user value and practical application [6][7]. - The goal of making super nodes affordable and usable for a broader range of users is highlighted, with Zhongke Shuguang's call to make super nodes a standard in China's computing landscape [7].

证券代码：603019 证券简称：中科曙光 公告编号：2026-014 曙光信息产业股份有限公司 2026年第一次临时股东会决议公告 一、 会议召开和出席情况 (三) 出席会议的普通股股东和恢复表决权的优先股股东及其持有股份情况： | 1、出席会议的股东和代理人人数 | 3,823 | | --- | --- | | 2、出席会议的股东所持有表决权的股份总数（股） | 359,006,489 | | 3、出席会议的股东所持有表决权股份数占公司有表决权股 | | | 份总数的比例（%） | 24.5502 | (四) 表决方式是否符合《公司法》及《公司章程》的规定，会议主持情况等。 本次股东会由公司董事会召集，经公司过半数董事推举，由董事崔梓杰先生 主持，会议采用现场投票与网络投票相结合的表决方式，会议的通知、召开、表 决方式符合《公司法》及《公司章程》的规定。 (五) 公司董事和董事会秘书的列席情况 (一) 股东会召开的时间：2026 年 3 月 25 日 (二) 股东会召开的地点：北京市海淀区中关村软件园 36 号楼会议室 本公司董事会及全体董事保证本公告内容不存在任何虚假记载、误导性陈述 或者重大遗漏，并 ...

北京市中伦律师事务所 关于曙光信息产业股份有限公司 2026 年第一次临时股东会的 法律意见书 二〇二六年三月 北京市中伦律师事务所 关于曙光信息产业股份有限公司 2026 年第一次临时股东会的 法律意见书 致：曙光信息产业股份有限公司 北京市中伦律师事务所（以下简称"本所"）接受曙光信息产业股份有限公 司（以下简称"公司"）委托，指派本所律师对公司 2026 年第一次临时股东会 进行见证，并出具《北京市中伦律师事务所关于曙光信息产业股份有限公司 2026 年第一次临时股东会的法律意见书》（以下简称"法律意见书"）。 为出具本法律意见书，本所律师见证了公司 2026 年第一次临时股东会，并 根据《中华人民共和国证券法》（下称《证券法》）、《中华人民共和国公司法》 （下称《公司法》）、《上市公司股东会规则》（下称《股东会规则》）、《律 师事务所从事证券法律业务管理办法》等法律、法规和规范性文件的要求以及《曙 光信息产业股份有限公司章程》（下称《公司章程》），按照律师行业公认的业 务标准、道德规范和勤勉尽责精神，查阅了本所律师认为出具本法律意见书所必 须查阅的文件，对公司本次股东会的召集和召开程序、出席会议人员资 ...

Core Viewpoint - The article discusses the upcoming launch of a new generation of super nodes by Zhongke Shuguang, which is expected to address the rapidly growing demand for AI agents and all-scenario computing power [1][3]. Group 1: Market Demand and Trends - The demand for computing power is shifting from a few tech giants to thousands of SMEs, research institutions, and traditional enterprises, which require accessible and cost-effective computing solutions [3][5]. - The super node market has seen significant activity over the past two years, but few products have achieved scalable delivery, particularly for ordinary enterprises [3][5]. - The core consumers of computing power are evolving, necessitating a transition from "extreme parameters" to "affordable applicability" in super node products [5]. Group 2: Technological Evolution - The competition in AI computing power is moving from single-card performance to cluster collaboration efficiency, interconnectivity, and system stability [5][6]. - Traditional server equipment struggles with training large models, while high-end super nodes are too costly for mid-tier markets, creating a gap that needs to be filled [5]. Group 3: Policy and Infrastructure - National policies are increasingly emphasizing the need for affordable and green computing power, which is driving the construction of computing infrastructure towards value-oriented solutions [5]. - There is a growing market space for practical technologies that can lower the barriers to computing power usage [5]. Group 4: Product Launch Expectations - The new product from Zhongke Shuguang is anticipated to achieve breakthroughs in deployment capabilities, potentially enhancing efficiency and cost-effectiveness for non-top-tier AI companies [6]. - The ability to quickly deploy and maintain environmental friendliness will be key highlights of the new super node product [6]. - The upcoming launch is seen as a pivotal moment in making high-end computing power more accessible to ordinary enterprises, which could reshape the competitive landscape in AI infrastructure [6].

Core Viewpoint - The issuance of convertible bonds has significantly accelerated this year, driven by new refinancing policies and increased corporate financing needs, indicating a positive shift in the market dynamics [1][4]. Group 1: Convertible Bond Issuance Trends - As of March 23, 2023, 20 A-share listed companies have announced plans to issue convertible bonds, with a total issuance scale exceeding 310 billion yuan, representing a 230% increase compared to the same period in 2025 [1][4]. - Notable companies leading in planned issuance include Zhongke Shuguang, Zhongchuang Zhiling, and Shentong Express, with Shentong Express planning to raise up to 30 billion yuan for logistics upgrades [3][4]. - The acceleration in convertible bond issuance is attributed to both policy relaxation and a resurgence in corporate financing demand, which may help alleviate long-standing supply-demand imbalances in the market [1][4]. Group 2: Policy Changes and Market Impact - The recent refinancing policy changes, effective from February 2026, aim to support high-quality listed companies and enhance the inclusivity and convenience of refinancing mechanisms, particularly for innovative SMEs [4][5]. - The new regulations favor "light asset, high R&D" companies and shorten the interval for rolling refinancing, encouraging firms to seek financing for growth and industrial upgrades [5]. Group 3: Approval Process and Future Expectations - The process for issuing convertible bonds has seen a reduction in time from board proposal to shareholder approval, dropping from an average of nearly 100 days in 2023 to approximately 175 days in 2026 [7]. - As of now, 12 convertible bonds have been issued this year, totaling 92.16 billion yuan, with expectations for 916.6 billion yuan in convertible bond issuance by the end of the year based on current approval rates [7]. Group 4: Market Conditions and Valuation - Despite the increase in issuance, the convertible bond market may still face supply constraints due to ongoing redemption pressures and the potential for strong redemption clauses to be triggered [8]. - The market anticipates that 70 convertible bonds, with a total outstanding scale of over 830 billion yuan, will mature in 2026, indicating a potential market contraction [8]. - The defensive attributes of convertible bonds are expected to remain crucial as the market navigates volatility, with a focus on low-risk investments amid broader market uncertainties [11].

Investment Rating - The investment rating for the computer industry is "Recommended" [2] Core Insights - MiniMax released its new flagship model M2.7, achieving a breakthrough with "self-evolution" capabilities, which allows the model to participate deeply in its training and optimization processes, resulting in approximately a 30% performance improvement on internal evaluation sets [6][14][17] - The price increases by AI companies, including Zhizhu and Alibaba Cloud, are driven by the explosive global demand for AI, leading to increased costs in hardware procurement and cloud services [18][19] - The report suggests that domestic AI manufacturers are entering an innovation cycle, with efficiency improvements expected to accelerate AI applications in various sectors, benefiting related application ecosystem and solution companies [20] Summary by Sections 1. Recent Developments - MiniMax's M2.7 model can execute over 100 cycles of "analyze-improve-validate," significantly enhancing its capabilities in software engineering and project delivery [6][14] - Zhizhu's API prices increased by 50% and 20% in two instances, while Alibaba Cloud announced price hikes for AI computing and storage services, marking a shift from discounting to a new pricing cycle in the cloud computing industry [18][19] 2. Market Performance - The computer index fell by 4.74% from March 16 to March 20, 2026, with notable stock performances among various companies [9] 3. Investment Recommendations - The report highlights key companies to watch in AI infrastructure and applications, including Haiguang Information, Cambricon, and Zhaoyi Technology, among others [20][21]

Investment Rating - The report maintains a "Recommended" rating for the computer industry [1] Core Insights - The demand for AI inference is continuously growing, driven by a significant increase in tokens usage, with a reported increase from 1.62 trillion tokens in March 2025 to 18 trillion tokens in March 2026, representing a growth of approximately 1011% [6][11] - Cloud service providers are raising prices for AI computing products due to rising hardware costs and increased demand for AI services, with price hikes ranging from 5% to 34% for various services [8][33] - The report highlights that the expansion of AI capabilities is expected to lead to a substantial increase in the number of active agents and token consumption, with annual token consumption projected to grow from 0.0005 PetaTokens in 2025 to 152,667 PetaTokens by 2030, reflecting a compound annual growth rate of 3418% [9][38] Summary by Sections Recent Trends - The computer industry has shown a relative performance decline of -10.1% over the past month, while the Shanghai Composite Index has remained stable [5] Price Adjustments - Major cloud providers like Alibaba Cloud and Baidu Cloud are increasing prices for AI computing services due to rising hardware procurement costs, with specific increases of up to 34% for Alibaba's AI computing services and 30% for Baidu's [8][27] Token Consumption Growth - The report indicates a significant rise in token consumption, with OpenClaw being a major contributor, achieving a monthly token call volume of 13.4 trillion as of March 2026 [17] - The share of domestic models in token consumption is increasing, with domestic models accounting for approximately 53.4% of the top models' total token calls as of March 2026 [14] Future Outlook - The report anticipates that the demand for AI computing and tokens will continue to rise, benefiting cloud service providers and related upstream and downstream companies [10][44] - The ongoing increase in hardware costs and the demand for AI services suggest that price adjustments in the cloud computing sector may persist [9][33]

- The CSI Cloud Computing 50 Index selects 50 listed companies involved in providing IaaS, PaaS, SaaS services, and hardware for cloud computing as index samples to reflect the overall performance of the cloud computing industry[3][27][28] - The index is based on December 31, 2014, with a base point of 1000 and was officially released on June 12, 2020[27][28] - The sample selection process includes filtering securities based on daily average trading volume, revenue growth rate, and market capitalization over specific periods, ensuring the top 50 securities are included[28] - The index is adjusted semi-annually, with changes implemented on the next trading day after the second Friday of June and December each year[28] - The index has demonstrated superior long-term and short-term performance compared to similar cloud computing indices and broad-based indices like CSI 500, with an annualized return of 13.44% since inception and a Sharpe ratio of 0.36[29][32] - From March 6, 2023, to March 6, 2026, the index achieved a return of 138.94%, significantly outperforming the CSI 500 index's return of 30.83% during the same period[29][32] - The index exhibits high liquidity, with a daily average trading volume of 29 billion shares and a daily average trading value of 1176 billion yuan as of March 6, 2026[32][34] - The index covers large, medium, and small-cap stocks with balanced weight distribution, with the weighted average free float market capitalization of its components being approximately 975 billion yuan[35][36] - The index is heavily weighted in industries such as data centers, application software, and communication system equipment, which collectively account for over 50% of the index's weight[38][39] - The index demonstrates strong profitability, with ROE at 8.83%, sales net profit margin at 7.07%, and revenue growth rate at 25.35% as of Q3 2025, outperforming similar indices[41][43][45] - The index's PE ratio (TTM) is 69.30, and its PB ratio (LF) is 8.14, indicating moderate valuation levels relative to its historical range[47][49][50] - The top 10 constituents of the index include companies from sectors such as communication equipment, electronic components, data centers, and software development, with weights ranging from 3.51% to 8.40%[51][52]