Core Insights - NVIDIA has introduced two silicon photonics CPO switches at the GTC conference to enhance the interconnect speed and energy efficiency of GPU clusters, making CPO a focal point in the industry [1] - The evolution of optical interconnect technology is crucial, with a roadmap from pluggable optical modules to 3D CPO, significantly increasing single-chip bandwidth [1][3] - The demand for computing power is growing globally, necessitating advancements in optical interconnect products to address this challenge [1][2] Group 1: Optical Interconnect Technology - The transition from traditional network interconnects to NVIDIA's GB200 NVL72 supernode can increase throughput by over three times compared to conventional methods [2] - Domestic AI chip and server manufacturers are increasingly adopting the supernode concept, indicating a shift in industry trends [2] - The two main paths for expanding supernode scale are using high-density cabinets or multiple cabinets with direct optical interconnect capabilities [2][3] Group 2: Challenges and Solutions - Current solutions face bandwidth and resource wastage issues, leading to network congestion, highlighting the need for revolutionary interconnect systems [3] - The proposed evolution of optical interconnect technology includes a shift towards 3D co-packaged optics, which could enhance interconnect bandwidth by 1-2 orders of magnitude within five years [3] - The complexity of connecting multiple GPUs necessitates advanced scheduling systems for efficient network management [3] Group 3: Innovations and Developments - At the 2025 WAIC, the company launched the LightSphere X distributed OCS all-optical interconnect chip and supernode solution, demonstrating its application with partners [4] - LightSphere X is recognized as the first domestic solution for optical interconnect and GPU supernodes, winning the SAIL Award for its innovation [4][5] - The technology allows for flexible scaling of supernodes, reducing deployment costs and enabling dynamic adjustments based on computing needs [5] - Performance metrics indicate that the unit interconnect cost is only 31% of that of the NVL72, with a significant increase in model computing efficiency [5]

Core Insights - The recent WAIC showcased the rising trend of supernodes, with multiple companies, including Huawei and Shanghai Yidian, presenting their supernode solutions, indicating a growing interest in high-performance computing [1][2][4] Group 1: Supernode Technology - Supernodes are designed to address the challenges of large-scale computing clusters by integrating computing resources to enhance efficiency and support models with trillions of parameters [1][2] - The technology allows for improved performance even when individual chip manufacturing processes are limited, marking a significant trend in the industry [1][5] - Supernodes can be developed through two main approaches: scale-out (horizontal expansion) and scale-up (vertical expansion), optimizing communication bandwidth and latency within the nodes [3][4] Group 2: Market Dynamics - The share of domestic AI chips in AI servers is increasing, with projections indicating a drop in reliance on foreign chips from 63% to 49% this year [6] - Companies like Nvidia are still focusing on the Chinese market, indicating the competitive landscape remains intense [6] - Domestic manufacturers are exploring alternative strategies to compete with established players like Nvidia, including optimizing for specific applications such as AI inference [6][8] Group 3: Innovation in Chip Design - Some domestic chip manufacturers are adopting sparse computing techniques, which require less stringent manufacturing processes, allowing for broader applicability in various scenarios [7] - Companies are focusing on edge computing and AI inference, aiming to reduce costs and improve efficiency in specific applications [8] - The introduction of new chips, such as the Homa M50, highlights the industry's shift towards innovative solutions that leverage emerging technologies like in-memory computing [8]

（原标题：【WAIC2025】 AI算力创新竞速，国产化实践走出超节点等新路） 在世博展览馆内，为应用提供底座能力的AI芯片、服务器、智算中心等厂商，展示了在芯片底层架构的自主研发，软件、整机的国产化适配，以 及针对应用场景的解决方案等方面的创新尝试。 AI算力创新 在H1核心技术馆内，阶跃星辰、月之暗面、智谱等大模型厂商的展台人流攒动。与这些模型厂商展台错落有致搭配的是沐曦、无问芯穹、摩尔线 程、燧原科技等算力厂商的展台。 伴随模型的迭代演进，人们对算力的需求也呈指数级增长。 正处于上市辅导备案中的沐曦展示了基于曦云C500系列芯片的服务器以及解决方案。展台人员介绍，服务器从编译、驱动到互联等全链路均已实 现国产化。 沐曦还首次展出了曦云C600通用GPU——一颗基于国产供应链设计、制造，自主可控的训推一体自研芯片。 经济观察报记者在展台未发现C600的性能、参数等内容。展台人员介绍，该芯片配置了业内前沿的显存，能强力支撑大模型训练、推理期间的海 量数据吞吐，主要用于云端AI训练与推理、通用计算、AI for Science等计算任务。 国内端边大模型AI芯片企业后摩智能是第二次参加WAIC，首发并展示了自 ...

Core Viewpoint - The article discusses the evolving landscape of the semiconductor industry in China, highlighting the significant role of investment firms like Wehao Chuangxin and the impact of recent market trends and IPO activities on the sector [1][2][3]. Group 1: Industry Trends - The semiconductor industry is undergoing a transformation, with a notable increase in IPO activities in Hong Kong, where 43 companies successfully listed in the first half of 2025, a 43% increase from the previous year [2]. - The market for semiconductor investments is shifting, with a focus on high-tech barriers and specialized projects as the industry matures and faces new challenges [8][17]. - The rise of AI and the need for integrated hardware solutions are creating new opportunities for companies that can deliver complete systems rather than just individual chips [46][50]. Group 2: Investment Insights - Investment logic in the semiconductor sector is centered around two key questions: the market potential of a product and the feasibility of reducing costs to a profitable level [9][66]. - The current investment environment is characterized by a cautious approach, as many market funds are retreating from semiconductor projects due to long lead times and uncertain returns [21][22]. - Despite challenges, there are still opportunities in areas such as AI-related sensors, smart terminals, and critical components that impact manufacturing processes [8][60]. Group 3: Company Case Studies - Wehao Chuangxin, backed by Weir Shares, has played a crucial role in the semiconductor ecosystem, facilitating significant mergers and acquisitions, including the notable acquisition of OmniVision by Weir Shares [2][39]. - The article highlights the importance of understanding the internal logic of emerging technologies, such as the trend towards "super nodes" in AI, which require a comprehensive approach to hardware and software integration [7][46]. - The shift in focus from general-purpose GPUs to specialized applications reflects the competitive landscape where companies must adapt to survive [40][55].

Group 1: Market Trends and Opportunities - The A-share market is experiencing increased opportunities, with the Shanghai Composite Index rising and reaching above 3600 points, driven by investor enthusiasm [1][7] - The recent performance of the semiconductor chip sector has been strong, indicating a positive trend in technology-related investments [1] - The "Digging Gold Competition" is ongoing, providing a platform for participants to engage in simulated trading and share insights on market trends and investment strategies [1][2] Group 2: Huawei's Ascend 384 Super Node - Huawei showcased its Ascend 384 super node at the 2025 World Artificial Intelligence Conference, attracting significant attention alongside other Chinese companies' super node solutions [2] - The Ascend 384 super node claims to achieve 67% higher total computing power, 107% higher network interconnect bandwidth, and 113% higher memory bandwidth compared to NVIDIA's NVL72 super node [3] - Analysts suggest that super nodes represent an efficient, scalable, and standardized computing cluster architecture necessary for the era of large models, influenced by chip performance and geopolitical factors [3] Group 3: Chikungunya Fever and Market Reactions - Chikungunya fever, caused by the chikungunya virus and transmitted by mosquitoes, has garnered market attention due to its symptoms, which include high fever and joint pain [5][6] - Companies such as Rainbow Group, Runben Co., and Renhe Pharmaceutical have indicated they possess products related to mosquito repellent and pain relief, responding to investor inquiries about chikungunya-related products [6] - Some participants in the "Digging Gold Competition" view the chikungunya fever topic as a speculative investment, suggesting that ordinary investors should consider smaller positions while focusing on stocks with growth potential [6] Group 4: Fund Predictions and Market Sentiment - A public fund's internal prediction of the Shanghai Composite Index reaching 10,000 points has sparked discussions, although some experts express skepticism about the accuracy of such forecasts [7] - Market analysts emphasize the importance of following trends and maintaining positions above the 5-day moving average, with a critical resistance level at 3700 points that could attract more buying interest if surpassed [7]

Core Insights - The report by Shenwan Hongyuan highlights a significant shift in computing power demand from single-point solutions to system-level integration, driven by the explosive growth of model parameters [1] - Two core dimensions for expanding computing power are identified as Scale-up and Scale-out, which will reshape the computing power industry chain and create investment opportunities [1] Group 1: Scale-up and Scale-out - Scale-up refers to increasing the number of GPUs within a single node, moving beyond traditional single-server limitations to a "super node" era, enabling full interconnectivity of GPUs [1][2] - Scale-out focuses on increasing the number of nodes, allowing for elastic expansion to support loosely coupled tasks like data parallelism, with essential differences in protocol stacks, hardware, and fault tolerance mechanisms [1][2] Group 2: Industry Trends and Mergers - Major chip manufacturers like NVIDIA, Broadcom, Huawei, and Haiguang are expected to deepen their focus on the Scale-up domain, while Ethernet technologies will concentrate on Scale-out [2] - Haiguang Information's planned merger with Zhongke Shuguang reflects the trend of vertical integration in the AI chip sector, aiming to enhance capabilities across communication, storage, and software [3] Group 3: Market Dynamics and Opportunities - AI chip manufacturers are not expected to enter the foundry business, as seen with AMD's divestment of its foundry operations post-acquisition of ZT System [4] - The industry chain may further differentiate into card design foundry suppliers and cabinet foundry suppliers, with card design capabilities becoming a key differentiator for value capture [4] - Companies to watch in this evolving landscape include Haiguang Information, Zhongke Shuguang, Inspur Information, Unisplendour, Digital China, Lenovo Group, and Huaqin Technology [4]

Investment Rating - The report maintains a "Positive" investment rating for the supernode industry, driven by the explosive growth of model parameters and the shift in computing power demand from single points to system-level integration [3]. Core Insights - The supernode trend is characterized by a dual expansion of high-density single-cabinet and multi-cabinet interconnection, balancing communication protocols and engineering costs [4][5]. - Domestic supernode solutions, represented by Huawei's CloudMatrix 384, achieve a breakthrough in computing power scale, surpassing single-card performance limitations [4][5]. - The industrialization of supernodes will reshape the computing power industry chain, creating investment opportunities in server integration, optical communication, and liquid cooling penetration [4][5][6]. - Current market perceptions underestimate the cost-performance advantages of domestic solutions in inference scenarios and overlook the transformative impact of computing network architecture on the industry chain [4][7]. Summary by Sections 1. Supernode: New Trends in AI Computing Networks - The growth of large model parameters and architectural changes necessitate understanding the two dimensions of computing power expansion: Scale-up and Scale-out [15]. - Scale-up focuses on tightly coupled hardware, while Scale-out emphasizes elastic expansion to support loosely coupled tasks [15][18]. 2. Huawei's Response to Supernode Challenges - Huawei's CloudMatrix 384 represents a domestic paradigm for cross-cabinet supernodes, achieving a computing power scale 1.7 times that of NVIDIA's NVL72 [4][5][6]. - The design of supernodes must balance model training and inference performance with engineering costs, particularly in multi-GPU inference scenarios [69][77]. 3. Impact on the Industry Chain - The industrialization of supernodes will lead to a more refined division of labor across the computing power industry chain, with significant implications for server integration and optical communication [6][4]. - The demand for optical modules driven by Huawei's CloudMatrix is expected to reach a ratio of 1:18 compared to GPU demand [6]. 4. Key Company Valuations - The report suggests focusing on companies involved in optical communication, network devices, data center supply chains, copper connections, and AI chip and server suppliers [5][6].

Core Viewpoint - The article discusses the emergence and significance of Super Nodes in addressing the increasing computational demands of AI, highlighting their advantages over traditional server architectures in terms of efficiency and performance [4][10][46]. Group 1: Definition and Characteristics of Super Nodes - Super Nodes are defined as highly efficient structures that integrate numerous high-speed computing chips to meet the growing computational needs of AI tasks [6][10]. - Key features of Super Nodes include extreme computing density, powerful internal interconnects using technologies like NVLink, and deep optimization for AI workloads [10][16]. Group 2: Evolution and Historical Context - The concept of Super Nodes evolved from earlier data center designs focused on resource pooling and space efficiency, with significant advancements driven by the rise of GPUs and their parallel computing capabilities [12][13]. - The transition to Super Nodes is marked by the need for high-speed interconnects to facilitate massive data exchanges between GPUs during model parallelism [14][21]. Group 3: Advantages of Super Nodes - Super Nodes offer superior deployment and operational efficiency, leading to cost savings [23]. - They also provide lower energy consumption and higher energy efficiency, with potential for reduced operational costs through advanced cooling technologies [24][30]. Group 4: Technical Challenges - Super Nodes face several technical challenges, including power supply systems capable of handling high wattage demands, advanced cooling solutions to manage heat dissipation, and efficient network systems to ensure high-speed data transfer [31][32][30]. Group 5: Current Trends and Future Directions - The industry is moving towards centralized power supply systems and higher voltage direct current (DC) solutions to improve efficiency [33][40]. - Next-generation cooling solutions, such as liquid cooling and innovative thermal management techniques, are being developed to support the increasing power density of Super Nodes [41][45]. Group 6: Market Leaders and Innovations - NVIDIA's GB200 NVL72 is highlighted as a leading example of Super Node technology, showcasing high integration and efficiency [37][38]. - Huawei's CloudMatrix 384 represents a strategic approach to achieving competitive performance through large-scale chip deployment and advanced interconnect systems [40].

Core Viewpoint - Huawei's CloudMatrix 384 super node claims to rival Nvidia's NVL72, but there are discrepancies in the hardware descriptions and capabilities between CloudMatrix and the UB-Mesh paper, suggesting they may represent different hardware forms [1][2][8]. Group 1: CloudMatrix vs. UB-Mesh - CloudMatrix is described as a commercial 384 NPU scale-up super node, while UB-Mesh outlines a plan for an 8000 NPU scale-up super node [8]. - The UB-Mesh paper indicates a different architecture for the next generation of NPUs, potentially enhancing capabilities beyond the current 910C model [10][11]. - There are significant differences in the number of NPUs per rack, with CloudMatrix having 32 NPUs per rack compared to UB-Mesh's 64 NPUs per rack [1]. Group 2: Technical Analysis - CloudMatrix's total power consumption is estimated at 500KW, significantly higher than NVL72's 145KW, raising questions about its energy efficiency [2]. - The analysis of optical fiber requirements for CloudMatrix suggests that Huawei's vertical integration may mitigate costs and power consumption concerns associated with fiber optics [3][4]. - The UB-Mesh paper proposes a multi-rack structure using electrical connections within racks and optical connections between racks, which could optimize deployment and reduce complexity [9]. Group 3: Market Implications - The competitive landscape may shift if Huawei successfully develops a robust AI hardware ecosystem, potentially challenging Nvidia's dominance in the market [11]. - The ongoing development of AI infrastructure in China could lead to a new competitive environment, especially with the emergence of products like DeepSeek [11][12]. - The perception of optical modules and their cost-effectiveness may evolve, similar to the trajectory of laser radar technology in the automotive industry [6].