Core Insights - The article discusses the emergence of high-performance computing clusters, specifically the scaleX ultra-cluster developed by Sugon, which integrates 16 scaleX640 supernodes to achieve over 5 EFlops of computing power, marking a significant advancement in domestic AI computing infrastructure [4][5]. Group 1: Ultra-Cluster Development - The scaleX ultra-cluster is the world's first single-cabinet 640-card supernode, utilizing advanced technologies such as high-density blade servers and immersion cooling, resulting in a 20-fold increase in computing density and a PUE value as low as 1.04 [1][4]. - The scaleX ultra-cluster represents a shift from traditional scattered server deployments to a more integrated and efficient computing unit, showcasing the progress of domestic computing infrastructure from conceptual designs to tangible products [1][5]. Group 2: Demand for Computing Power - As mainstream AI models transition from hundreds of billions to trillions of parameters, the demand for computing power has surged, necessitating the development of EFLOPS-level and ten-thousand-card high-performance clusters as standard configurations for large models [2][3]. - The supernode architecture is becoming a preferred choice for new ten-thousand-card clusters due to its density and performance advantages, allowing for significant optimization in computing capabilities [3]. Group 3: Networking and Scalability - The scaleX ultra-cluster employs the scaleFabric high-speed network, which utilizes the first domestic 400G-class InfiniBand RDMA network cards, achieving 400 Gb/s bandwidth and under 1 microsecond communication latency, enhancing scalability to over 100,000 cards [7]. - The architecture allows for both Scale-up (vertical expansion) and Scale-out (horizontal expansion), addressing traditional communication bottlenecks and enabling the construction of large-scale intelligent computing clusters [6]. Group 4: Challenges and Considerations - The deployment of supernodes introduces systemic challenges, including heat dissipation from numerous chips, stability issues from mixed optical and copper interconnects, and reliability concerns from long-term operation of multiple components [8]. - As the scale of intelligent computing clusters expands, key challenges include ensuring scalability, reliability, and energy efficiency, necessitating breakthroughs in power supply technology and advanced software management for sustainable operation [8].

Core Insights - The launch of the scaleX万卡超集群 marks the first physical appearance of a domestic万卡级 AI cluster system in China, showcasing significant advancements in AI computing capabilities [1][3] - The scaleX640 super node, part of the scaleX万卡超集群, integrates 16 super nodes and achieves a total computing power exceeding 5 EFLOPS, highlighting the growing demand for high-performance computing in AI applications [3][5] - The industry is transitioning from traditional server architectures to super node designs, which offer higher density and performance, becoming the preferred architecture for new万卡级 clusters [2][5] Company Developments - 中科曙光's scaleX640 super node is recognized as the world's first single cabinet-level 640-card super node, emphasizing the company's leadership in high-density computing solutions [2][3] - The scaleX万卡超集群 utilizes the scaleFabric high-speed network, which can achieve 400Gb/s bandwidth and less than 1 microsecond communication latency, significantly enhancing inter-node communication efficiency [7][8] - The company is addressing challenges related to system cooling, stability, and reliability as it scales up its super node architecture to meet the increasing demands of AI workloads [6][8] Industry Trends - The demand for computing power is rapidly increasing as AI models evolve from hundreds of billions to trillions of parameters, necessitating the development of万卡级 and beyond computing clusters [1][5] - Major international players like Meta, Microsoft, and OpenAI are also investing in the construction of 100,000-card clusters, indicating a global trend towards larger-scale AI computing infrastructures [6] - The industry is facing critical challenges in scalability, reliability, and energy efficiency as computing centers grow from megawatt to gigawatt levels, necessitating innovative power supply technologies and advanced management software [8]

Core Viewpoint - Eagle Semiconductor is positioned as a leading player in the VCSEL (Vertical-Cavity Surface-Emitting Laser) market, having achieved significant milestones in high-end optical communication technology, particularly with the production of 100G VCSEL chips, breaking the monopoly of American companies in this field [5][11][12]. Group 1: Company Overview - Eagle Semiconductor, founded by Bian Difei, has over 50% of its workforce dedicated to R&D and is the only company in China capable of mass-producing 100G VCSEL chips for optical communication [5][11]. - The company has recently completed a B+ round financing of 700 million yuan, setting a record for domestic companies in this sector, with participation from major investors including CITIC Jinshi and Guoxin Fund [6][29]. - The founder's previous experience in the semiconductor industry, particularly in LED chips, has informed the company's strategy to adopt an IDM (Integrated Device Manufacturer) model, which is capital-intensive but necessary for long-term success [7][21]. Group 2: Market Dynamics - The VCSEL market is expected to grow significantly, with a compound annual growth rate (CAGR) of 22.2% in the optical communication sector from 2022 to 2027, driven by increasing demand for data centers and AI computing [11][12]. - The company strategically chose to enter the challenging optical communication market rather than the more accessible consumer electronics market, believing that tackling the hardest challenges first would provide a competitive advantage [32][33]. - The demand for optical interconnects is rising due to the increasing number of GPUs in data centers, which necessitates high-speed communication solutions [6][8]. Group 3: Competitive Landscape - Prior to Eagle Semiconductor's entry, the VCSEL market was dominated by American companies like Coherent and Lumentum, which held nearly 80% of the market share [11]. - The company aims to fill the gap in the domestic market for high-end VCSELs, which has been largely absent, and is focused on building a complete ecosystem for optical communication in China [28][41]. - Despite a 1-2 year technological gap compared to the U.S., the company is confident in its ability to catch up quickly due to the unique demands of the Chinese market [41][42]. Group 4: Future Outlook - The founder envisions a future where every household may have access to mini supernodes, with VCSEL technology playing a crucial role in the intelligent era of data processing and transmission [13][14]. - The company is also exploring opportunities in related fields such as laser radar and 3D sensing, indicating a broader vision for growth beyond just optical communication [43].

Core Viewpoint - Kunlun Core, a self-developed chip project under Baidu, is accelerating its path to IPO following its upcoming share reform [1][3]. Group 1: Company Overview - Kunlun Core was established in 2011 and began independent financing in 2021, with Ouyang Jian as CEO [1]. - Baidu is the controlling shareholder of Kunlun Core, which has completed six rounds of financing, the latest in July this year [1]. Group 2: Product Development - The main product of Kunlun Core is the P800 chip, set to launch in 2024, which has been validated within Baidu's operations [1]. - In the first half of this year, Baidu activated a 30,000-card cluster based on the P800 chip, achieving high cost-performance in training a multimodal model [1]. Group 3: Future Product Roadmap - Kunlun Core plans to release the M100 chip in 2026, optimized for large-scale inference scenarios, and the M300 chip in 2027, aimed at ultra-large multimodal model training and inference [1][3]. - Baidu will launch the "Tianchi 256 Super Node" in the first half of 2026, which supports 256 interconnected cards, enhancing total interconnect bandwidth by four times and improving performance by over 50% compared to the previous model [3]. - The "Tianchi 512 Super Node" is expected in the second half of 2026, supporting 512 interconnected cards and capable of training trillion-parameter models [3]. Group 4: Long-term Vision - According to Baidu's roadmap, Kunlun Core anticipates launching a new generation of N series chips by 2029 and achieving a million-card Kunlun Core single cluster by 2030 [4].

Investment Rating - The report assigns a rating of "Increase" for the company, indicating an expected relative increase of 5% to 15% compared to the CSI 300 index over the next six months [3][36]. Core Insights - The company is actively pursuing the acquisition of the remaining 19% stake in Xinhua San held by HPE, which reflects a positive outlook on its fundamentals [2]. - In 2025, the company's revenue growth accelerated, with a reported revenue of 77.32 billion yuan for the first three quarters, representing a year-on-year increase of 31.41%. However, the net profit attributable to the parent company declined by 11.24% [9]. - The growth momentum for Xinhua San is primarily driven by domestic government and enterprise business, as well as international business, with respective year-on-year growth rates of 62.55% and 83.99% in the first three quarters of 2025 [9]. - The company has increased its stake in Xinhua San to 81%, with Xinhua San contributing 77% of the company's revenue and 146% of its net profit in the first three quarters of 2025 [9]. - The company is expected to alleviate financial pressure from the acquisition of Xinhua San through its successful listing on the Hong Kong Stock Exchange [9]. Financial Data Summary - As of September 30, 2025, the company reported a net asset value per share of 5.06 yuan and a net profit margin of 9.70% [5]. - The company's revenue is projected to reach 101.06 billion yuan in 2025, with a growth rate of 27.88% [12]. - The earnings per share (EPS) for 2025 is estimated at 0.52 yuan, with corresponding price-to-earnings (P/E) ratios of 46.62, 32.07, and 24.41 for the years 2025, 2026, and 2027 respectively [12].

Investment Strategy Overview - The report highlights a significant investment opportunity in the electronic industry, driven by the dual forces of domestic cloud computing capabilities and the explosive growth of AI at the edge, marking a pivotal moment for the sector leading up to 2026 [1] Semiconductor Manufacturing - Capital expenditure in semiconductor manufacturing is set to reach new heights, with domestic fabs expected to experience a dual expansion in memory and advanced logic production in 2026, supporting a sustained high level of demand in the wafer foundry sector [2] - The semiconductor equipment sector is anticipated to witness a "β+α" resonance market, with a focus on industry leaders benefiting from expansion dividends and companies like Jingzhida and others that have a clear technology realization logic [2] Cloud Computing Chips - Global cloud service providers (CSPs) are increasing capital expenditures, with the combined capital expenditure of the four major overseas CSPs (Google, Amazon, Microsoft, Meta) reaching $97.9 billion in Q3 2025, a 10% quarter-on-quarter increase [5] - Domestic cloud computing is catching up, with significant growth potential as demand for computing power rises, particularly from leading firms like ByteDance [5][17] - Companies such as Cambricon and Haiguang Information are recommended for investment due to their expected performance release in the domestic computing power sector [5][17] Edge Computing Chips - The strategic importance of edge AI is rapidly increasing, with major tech companies integrating AI models into core products, enhancing the demand for System on Chip (SoC) manufacturers [7][36] - Companies like Amlogic and RichChip are positioned to benefit from the growing demand for edge AI applications, particularly in smart home devices [7][39] Storage Sector - The storage sector is experiencing a strong cyclical upturn, with DRAM and NAND indices showing significant increases of 101% and 79% respectively from September to November 2025 [5] - Major CSPs are increasing their procurement of storage products, leading to a sustained rise in storage prices and creating a "super cycle" in the industry [5] Analog Sector - The analog sector is seeing growth driven by increasing automotive demand, although price pressures are expected to persist [5] - The sector is poised for opportunities related to new AI applications as the industry evolves [5] Consumer Electronics - AI is driving a transformation in terminal interactions, with a notable shift in smartphone upgrades and the emergence of AR products [7] - The AR glasses market is expected to see significant growth in 2026, with major companies like Meta and Apple launching new products [7] PCB/CCL Market - The PCB/CCL market is set to benefit from increased capital expenditures by global cloud providers, with the AI PCB market projected to reach 60 billion yuan in 2026, a 229.8% year-on-year increase [7] - The introduction of low-loss materials and advanced architectures is expected to significantly enhance the value of PCB products [7]

Core Viewpoint - China Mobile has awarded a procurement project for a super node testing device to Huawei, with a bid of 135 million yuan, indicating a significant investment in advanced computing infrastructure [1] Group 1: Huawei's Innovations - Huawei's "Scale-up super large-scale super node computing platform" was recognized as its top invention, showcasing a new architecture that allows for flexible resource allocation among AI processors [1] - The Ascend 384 super node features 384 Ascend NPUs and 192 Kunpeng CPUs, achieving a total computing power of 300 Pflops, which is 1.7 times that of NVIDIA's NVL72 [2] - The total network bandwidth of the Ascend 384 super node reaches 269 TB/s, surpassing NVIDIA's NVL72 by 107% [2] Group 2: Industry Trends - The demand for computing power in large models is rapidly increasing, while traditional computing architectures face challenges such as low resource utilization and frequent failures [4] - Super nodes are emerging as a new norm in AI infrastructure, with several domestic manufacturers exploring super node systems [4] - Inspiring innovations include the launch of the world's first single-cabinet 640-card super node by Sugon, which enhances computing density by 20 times compared to traditional solutions [4][5] Group 3: Future Developments - Huawei plans to launch the Atlas 950 super node in Q4 2026, featuring 8192 Ascend 950DT chips and achieving FP8 computing power of 8 EFLOPS [12] - The Atlas 950 super node is projected to be significantly more powerful than NVIDIA's upcoming NVL144, with a total computing power 6.7 times greater and memory capacity 15 times larger [14] - Future plans include the Atlas 950 SuperCluster, which will consist of 64 Atlas 950 super nodes, achieving a total FP8 computing power of 524 EFLOPS, surpassing the current largest cluster, xAI Colossus [14]

Core Viewpoint - Jiangsu Maxinlin Aviation Technology Co., Ltd. is responding to an inquiry from the Shanghai Stock Exchange regarding its joint investment with related parties and plans to increase capital in a controlled enterprise, Photon Computing [9][10]. Group 1: Investment Details - The company plans to invest between 201 million and 311 million yuan in the joint venture, holding a 99.5% stake through the partnership with Bai Bing, who holds 0.5% [9][10]. - The investment is positioned as a financial investment aimed at enhancing operational capital and expanding market share for Photon Computing [10][11]. Group 2: Financial Implications - The funding for this investment will primarily come from the company's own funds and bank loans, which may increase the company's liabilities and financial expenses [7][42]. - The company has set up mechanisms such as performance commitments and buyback clauses to mitigate risks associated with the investment [45]. Group 3: Market and Competitive Landscape - The industry is experiencing intensified competition, particularly in the field of supernodes, which are critical for computing power and technological competitiveness [3][21]. - Photon Computing is positioned to benefit from the growing demand for intelligent computing centers, but faces risks from potential policy changes and market dynamics [2][20]. Group 4: Operational Synergies - The investment is expected to create synergies between the company's existing aviation component business and Photon Computing's operations, enhancing overall competitiveness in the computing market [37][39]. - The company aims to leverage Photon Computing's technology in optical interconnects to improve its own computing service offerings [38][44]. Group 5: Risk Management - The company has outlined various risks associated with the investment, including market competition, product innovation, and talent retention, which could impact Photon Computing's performance [3][4][5]. - A thorough due diligence process is in place to ensure compliance and mitigate potential risks during the investment phase [18][29].

在 AI 大模型的推动下，算力竞争正从"单卡比拼"走向"集群博弈"。越往前走，行业越清楚： 决定上限的，已经不再是一颗 GPU 的峰值算力，而是隐藏在背后，由芯粒（Chiplet）、封 装与互联共同构成的系统底座。 我们看到，算力架构正从单芯片走向多芯粒、超节点，如何在如此复杂的系统中，依然保持 高带宽互联、灵活扩展与高效调度，正成为新一代算力生态的关键命题。 在刚刚结束的 ICCAD 2025 上，半导体行业观察走进奇异摩尔展台，围绕"用芯粒互联重构 AI 算力生态"这一主题，与奇异摩尔联合创始人兼产品及解决方案副总裁祝俊东进行了深入 交流。 Q 您认为，互联是否是实现自主算力的关键环节？ 随着超节点架构兴起，GPU互联不仅是技术升级，更是系统模式的重塑。您如何看待 超节点架构带来的业务与系统挑战？奇异摩尔是如何应对和布局的？ 祝俊东： 超节点是计算中心领域这两年比较火的话题。过去大家对于超节点到底有没有用还存 在争议，但是我觉得从去年开始，这个争议已经基本平息了，大家都一致认为超节点是一个非常重 要的方向，并且现在无论是国内或国外，都已经开始大规模商业化。为什么会有超节点呢？这个技 术本身是有非常高的互联 ...

Core Viewpoint - The recent announcements from major Chinese tech companies like Huawei and Baidu regarding their AI chip development signify a shift in the domestic semiconductor landscape, aiming to fill the gap left by Nvidia and enhance the competitiveness of Chinese AI chips [3][4][5]. Group 1: AI Chip Development - Major Chinese companies are increasingly revealing their AI chip roadmaps, with Huawei planning to release four Ascend AI chips over the next three years, while Baidu is set to launch two Kunlun AI chips in the next two years [2][3]. - Huawei's detailed disclosure of technical parameters for its chips, including bandwidth, computing power, and memory, marks a significant change in the traditionally low-profile approach of Chinese chipmakers [2][7]. - The introduction of supernodes and clusters is seen as a critical strategy for overcoming the limitations of China's semiconductor manufacturing processes, which are currently capped at the 7nm node [10][12]. Group 2: Competitive Landscape - The competition in the global AI chip market is characterized as asymmetric, with Chinese chips lagging behind North American counterparts like Nvidia in various technical specifications, yet capable of leveraging networking capabilities to surpass them in performance [5][6]. - Huawei's Ascend series has been recognized as a formidable competitor, with its first chip released in 2018 and subsequent iterations showing significant performance improvements despite challenges posed by U.S. sanctions [6][8]. - Baidu's Kunlun chip, while still behind in performance compared to Nvidia's offerings, is focusing on cost-effectiveness and specific use cases, indicating a strategic approach to market entry [8][9]. Group 3: Market Dynamics - The domestic AI chip market is witnessing a shift towards inference tasks, with inference scenarios accounting for 42% of the GenAI IaaS service market, while training scenarios have decreased to 58% [14][15]. - The challenges of using domestic AI chips for large model training are acknowledged, with companies like Huawei and Baidu working to adapt their technologies to meet these demands [14][15]. - The push for self-developed chips by major cloud providers is seen as a way to reduce costs and improve performance, with companies like Kunlun seeking to penetrate external markets [16][17].