转自：新华财经 步入展厅，各类前沿科技的应用成果跃然眼前，令人直观感受到媒体融合的创新活力。本次展览汇聚了 行业顶尖机构与企业，包括媒体融合生产技术与系统国家重点实验室、中国新闻技术工作者联合会等权 威机构，以及中国联通、华为、中兴、中国电子云、百度、北大方正、索贝数码等25家国内领先的媒体 融合应用软件与设备企业。它们带来的不仅是产品，更是对媒体业态深度变革的生动诠释。 十月的鄂尔多斯，秋意正浓，一场引领媒体行业风向的盛会——2025媒体融合发展鄂尔多斯年会于此启 幕。大会焦点之一的"媒体融合技术展"，以"智融·创变·未来"为主题，如同一扇未来之窗，向与会者全 景式展现了媒体融合技术的最新突破与前沿应用。这里不仅是技术的秀场，更是思想碰撞、预见行业未 来的重要平台。 展区现场，中国联通的展示区域以其独特的创新理念吸引大量参观者驻足关注。其创新打造的"基层通 智慧治理平台"，是本次展会"技术赋能社会治理"的典范之作。该平台以智能体应用为核心，精准赋能 党建引领下的基层治理，像一位智慧的"社区管家"，让基层工作更高效。其中"党群通"应用由新华社提 供端到端权威内容服务支持，构建形成集服务、宣传、教育于一体的一站 ...

Investment Rating - The report maintains a "Recommended" rating for the computer industry [1] Core Insights - The report addresses key issues such as the parameters of Huawei's next-generation Ascend processors, supernodes, and cluster solutions compared to NVIDIA, the core components of the Huawei Ascend ecosystem, and the changes in demand within the AI computing industry, highlighting how Huawei's Ascend benefits from the localization of AI chips [5][7] - Huawei's Ascend AI chips are positioned as the first globally to cover all scenarios, with the CloudMatrix 384 cluster solution based on the Ascend 910C processor. As of September 18, 2025, over 300 units of the Atlas 900 A3 SuperPoD supernode have been deployed [5][24] - The report emphasizes the growth potential of the AI computing industry, driven by increasing demand from cloud service providers (CSPs), sovereign nations, and AI inference, alongside the push for domestic chip production due to escalating U.S. sanctions [7][9] Summary by Sections Section 1: Ascend Chips - The Ascend processors are based on Huawei's Da Vinci architecture, optimized for matrix operations, and are the foundation for Ascend computing products, offering high computing power, energy efficiency, and flexibility [15] - The report outlines a roadmap for the Ascend processors, with annual releases expected to double computing power each year [5][17] Section 2: AI Computing Infrastructure - Huawei's AI computing solutions, including the Atlas series, provide a comprehensive infrastructure for AI applications across various scenarios, covering the entire process of deep learning inference and training [18] - The report highlights the deployment of the Atlas 900 A3 SuperPoD, which integrates 384 Ascend 910C NPUs and 192 Kunpeng CPUs, achieving a computing capacity of 300 PFLOPS [24][26] Section 3: Market Demand and Trends - The report notes that NVIDIA anticipates global AI capital expenditures to reach $3-4 trillion over the next five years, with significant growth expected in AI inference markets [7] - The localization of AI chips in China is accelerating, with domestic brands accounting for approximately 30% of the market share as of 2024 [7] Section 4: Competitive Positioning - Huawei's Ascend aims to become the second choice globally for AI computing power, competing directly with NVIDIA through its comprehensive software and hardware ecosystem [5][9] - The report identifies key players in the AI chip and server markets, including companies like Haiguang Information, Cambricon, and major server manufacturers [10]

Core Viewpoint - Huawei's Full Connect 2025 Conference showcased the Lingqu Unified Interconnection Protocol, announcing the launch of Ascend 950/960/970 and Kunpeng 950/960 over the next three years, highlighting the gradual enhancement of domestic AI computing chip capabilities amid US-China tensions [9][58]. Group 1: AI Computing Chip Development - The Ascend NPU roadmap includes the release of Ascend 950 (PR and DT versions) in 2026, followed by 960 in 2027 and 970 in 2028, with significant performance improvements [15][20]. - The Kunpeng CPU will see the launch of Kunpeng 950 in late 2026 and Kunpeng 960 in early 2028, supporting advanced computing needs [20][34]. - Domestic chip manufacturers like Haiguang and Cambricon are projecting substantial revenue growth, with Haiguang targeting a CAGR of 44% over three years [3][58]. Group 2: Advanced Manufacturing and Semiconductor Industry - The domestic lithography machine industry is focusing on complete machines and related components, with expectations for advanced process expansion by 2026 [3][59]. - The domestic semiconductor industry is expected to benefit from the acceleration of independent and controllable demands, particularly in advanced logic and storage production lines [3][62]. Group 3: Storage and Edge Computing - The demand for inference and edge computing storage is increasing, with significant growth expected in AI PCs, smartphones, and wearable devices by 2026 [4][58]. - Domestic manufacturers are enhancing their enterprise storage product lines, with companies like Jiangbolong and Baiwei Storage launching new enterprise-level solutions [4][58]. Group 4: Investment Recommendations - Investment opportunities are suggested in AI computing chips, high-end chip manufacturing, packaging, storage, and related equipment and materials [5].

Group 1 - The core viewpoint of the article highlights the introduction of the AI computing cluster solution Cloud Matrix 384 at the WAIC2025 conference, which utilizes a fully interconnected topology for efficient collaboration [1] - The Atlas 900 A3 SuperPoD features a super-node architecture with ultra-large bandwidth of 392GB/s, ultra-low latency of less than 1 microsecond, and exceptional performance of 300 PFLOPs, enhancing the training performance of models like LLaMA3 by over 2.5 times compared to traditional clusters [1] - The Ascend 910C outperforms NVIDIA's GB200 NVL72 in system-level BF16 computing power (300 PFLOPS), HBM capacity (49.2TB), and bandwidth (1229TB/s) [1] Group 2 - The semiconductor equipment ETF (159516) tracks the semiconductor materials and equipment index (931743), focusing on the materials and equipment sectors within the semiconductor industry, covering companies involved in manufacturing, processing, and testing [1] - The index constituents are primarily companies with significant technological advantages and core competitiveness in semiconductor materials and equipment, aiming to reflect the overall performance and industry development trends of listed companies in this segment [1] - Investors without stock accounts can consider the Guotai Zhongzheng Semiconductor Materials and Equipment Theme ETF Initiated Link A (019632) and Link C (019633) [1]

Group 1: Market Trends - AI computing hardware sector showed significant gains on July 29, driven by increased capital expenditure from overseas cloud vendors and a release of performance results [1] - Google's Q2 earnings report revealed an increase in annual capital expenditure to $85 billion, primarily for AI servers and data center construction, with cloud computing revenue reaching $13.6 billion, a 32% year-over-year increase [2] - The ongoing trade negotiations between China and the U.S. in Stockholm may lead to a reduction in trade conflicts, which could benefit domestic optical module companies [3] Group 2: Domestic Developments - At the recent World Artificial Intelligence Conference, companies like Huawei and ZTE showcased advancements in AI computing infrastructure, indicating steady improvements in domestic capabilities [2] - Huawei presented its Ascend 384 super node machine, while ZTE displayed a super node server designed for large-scale model training and inference [2] Group 3: Investment Opportunities - Investors are encouraged to consider communication ETFs (515880) and the ChiNext AI ETF (159388) for potential investment opportunities in the optical module sector [1][7] - Public funds have shown an increase in holdings in the communication sector, with a rise in the top ten holdings' proportion to 3.61%, marking a shift from underweight to overweight [3]

Core Insights - Huawei showcased its Ascend chips and AI system at the World AI Conference in Shanghai, highlighting its competitive position in the AI chip market [1][4] - Nvidia, despite previous hopes of re-entering the Chinese market with its H20 chips, did not participate in the conference, indicating potential challenges in its relationship with China [2][3] Group 1: Huawei's Developments - Huawei presented its new computing system, the "Atlas 900 A3 SuperPoD," which connects 384 Ascend chips for AI model training [4] - The Ascend chips, while individually less powerful than Nvidia's Blackwell chips, compensate by utilizing five times more chips in their system compared to Nvidia's GB200 [4] Group 2: Competitive Landscape - Nvidia's absence at the conference contrasts with Huawei's prominent display, emphasizing the latter's growing influence in the Chinese AI market [2][3] - Other Chinese companies, such as Moore Threads and Yunsilicon, also participated in the AI expo, indicating a competitive environment in the semiconductor industry [5]