行 业 及 产 业 2025 年 10 月 10 日 摩尔线程：生态、架构、集群 看好 ——国产 ASIC 系列研究之 5 本期投资提示： 相关研究 《国产 ASIC：PD 分离和超节点——ASIC 系列研究之四》 2025/09/26 《国产算力趋势走强，沐曦领衔通用 GPU——国产 ASIC 系列研究之 3》 2025/09/24 证券分析师 黄忠煌 A0230519110001 huangzh@swsresearch.com 杨海晏 A0230518070003 yanghy@swsresearch.com 研究支持 曹峥 A0230525040002 caozheng@swsresearch.com 杨紫璇 A0230524070005 yangzx@swsresearch.com 王开元 A0230125030001 wangky@swsresearch.com 研 行 业 请务必仔细阅读正文之后的各项信息披露与声明 联系人 证 券 研 究 报 曹峥 (8621)23297818× caozheng@swsresearch.com 本研究报告仅通过邮件提供给 中庚基金 使用。1 究 / 行 业 深 ...

Group 1 - The electronic industry experienced a 3.51% increase from September 22 to September 26, ranking third among major sectors [2][3] - The electronic industry's price-to-earnings ratio stands at 72.55, with the highest valuations in the computer, defense, and electronic sectors [2][3] - Within the electronic sector, semiconductor equipment saw the largest increase of 15.56% during the same period [2][3] Group 2 - The storage chip market is entering a new price increase cycle, with major manufacturers like Samsung and SK Hynix reducing DDR4 production to focus on higher-margin products [3][4] - Flash memory prices have risen significantly, with NAND Flash wafer prices increasing by nearly 10% in September, and DRAM prices rising approximately 72% over the past six months [3][4] - Companies such as Micron and SanDisk have announced price hikes of 20%-30% and over 10%, respectively, indicating strong demand in the enterprise market [3][4] Group 3 - Alibaba announced significant upgrades to its AI infrastructure at the Cloud Summit, including the launch of the Qwen3-Max model with over a trillion parameters [6] - The new AI server, Panjiu 128, supports 128 AI computing chips and offers substantial improvements in performance and efficiency [6] - The successful IPO of Moore Threads marks a significant milestone in the domestic AI chip industry, with the company achieving technological breakthroughs in GPU architecture [7][8]

Core Viewpoint - The article discusses the rapid development and competitive landscape of domestic GPU companies in China, particularly focusing on the "Four Little Dragons" of the GPU sector: Moer Thread, Muxi Integration, Suiruan Technology, and Birun Technology. It highlights their technological advancements, market positioning, and the potential for investment opportunities in the sector [2][3][55]. Group 1: Company Overview - Moer Thread is set to undergo review for its IPO on the Sci-Tech Innovation Board, achieving a record timeline of just three months from acceptance to review. Post-IPO valuation is projected between 150 billion to 300 billion yuan, with optimistic estimates reaching 500 billion yuan [2]. - Muxi Integration, Suiruan Technology, and Birun Technology have also initiated or completed IPO counseling, with recent financing valuations ranging from 15 billion to 25 billion yuan [3]. Group 2: Technological Comparison - Moer Thread aims to create a full-function GPU capable of supporting AI computing, graphics rendering, scientific computing, and video encoding on a single chip. Its product line includes AI computing products, professional graphics acceleration products, and consumer-grade graphics cards [11][16]. - Muxi Integration focuses on general-purpose GPU solutions, with its main product, the Xiyun C series, supporting large model training and high-performance computing. It claims to have a memory capacity of 144GB HBM3e, surpassing NVIDIA's H20 [28][30]. - Suiruan Technology specializes in cloud-based AI computing infrastructure, with its flagship product, the L600, boasting a memory capacity of 144GB HBM3e, which is 1.5 times that of NVIDIA's H20 [52][54]. - Birun Technology's BR100 series claims to have peak computing power three times that of NVIDIA's A100, although its actual performance utilization is lower [42][44]. Group 3: Market Positioning and Competitive Landscape - The "Four Little Dragons" are positioned in the first tier of the domestic GPU market, each carving out differentiated competitive advantages in various niches such as full-function GPUs, high-performance computing, AI computing, and supercomputing clusters [55]. - Despite their advancements, these companies still face challenges in practical efficiency, software ecosystems, and supply chain optimization compared to established players like NVIDIA [55]. Group 4: Financing and Investment Dynamics - Moer Thread has experienced rapid financing, with its valuation increasing from under 1 billion yuan in the angel round to 24.62 billion yuan in the Pre-IPO round, making it one of the highest-valued unicorns in the domestic GPU sector [57]. - Muxi Integration has received significant government support, with over 60% of its funding coming from state-owned capital, participating in national strategic projects [59]. - Suiruan Technology has completed 11 rounds of financing, with Tencent as its largest institutional shareholder, holding a 20.49% stake [62]. - Birun Technology has seen continuous investment from state-owned funds, with its A round financing setting a record in the domestic chip design sector [65].

Core Viewpoint - DeepSeek has launched DeepSeek-V3.1, featuring comprehensive upgrades in mixed reasoning architecture, cognitive efficiency, and agent capabilities, with a notable focus on the UE8M0 FP8 technology designed for the next generation of domestic chips [1][3]. Company Insights - The announcement of UE8M0 FP8 technology has triggered a positive response in the capital market, with stocks of companies like Cambrian and Haiguang Information experiencing short-term increases [3]. - DeepSeek's introduction of UE8M0 FP8 technology signifies a strategic move into chip technology, reflecting the trend of software-defined hardware, where software stacks and algorithm optimization are crucial for enhancing hardware performance [5]. Industry Trends - The FP8 format is emerging as a new standard for AI training and inference, with major international players like NVIDIA and AMD also investing in this technology, indicating a potential shift in global technology direction [5]. - The FP8 technology optimizes data storage and transmission, significantly improving chip throughput and energy efficiency, thereby enhancing the competitiveness of domestic AI chips in both domestic and international markets [4][5]. - The balance between data precision and computational efficiency remains a core issue in the evolution of technology within the artificial intelligence and high-performance computing sectors [3].

Core Viewpoint - The release of DeepSeek's V3.1 and its mention of "UE8M0 FP8" has triggered a significant positive reaction in the A-share market, indicating a potential breakthrough for domestic chip manufacturers in the computing power sector [6][15]. Group 1: Market Reaction - Following the announcement, over 2800 stocks in the A-share market experienced an increase, with significant gains in computing power stocks and chip manufacturers like Cambrian and Zhongxin International [6][15]. - The market's response reflects optimism about the domestic chip industry's ability to compete on an international level, marking a rare moment of resonance within the domestic computing power industry [6][15]. Group 2: Technical Insights - "FP8" is a floating-point format that compresses data to 8 bits, aimed at reducing memory bandwidth pressure during AI model training and inference [8]. - The challenge for domestic chip manufacturers has been the limited native support for FP8, with only a few companies like Moore Threads offering GPUs that support this format [8][9]. - The introduction of "UE8M0" as a scaling factor in the MX format allows for efficient data management, significantly reducing bandwidth overhead by 75% compared to traditional methods [12][15]. Group 3: Industry Implications - DeepSeek's announcement signifies a shift in the domestic computing power ecosystem, moving away from reliance on NVIDIA and towards supporting local chip manufacturers [13][15]. - The successful implementation of UE8M0 FP8 indicates that domestic chip companies have made progress in optimizing floating-point formats, compiler optimizations, and training framework adaptations [13][15]. - Companies like Moore Threads and Cambrian are positioned to benefit from this development, as they have been in contact with DeepSeek and support FP8 natively [14][15].

Core Viewpoint - The release of DeepSeek's V3.1 has sparked significant excitement in the A-share market, indicating a potential breakthrough for domestic chip manufacturers in the computing power sector [3][4][17]. Industry Impact - The A-share market saw over 2800 stocks rise, with computing power stocks and chip manufacturers experiencing substantial gains, highlighting a collective optimism in the sector [3][4]. - DeepSeek's statement is viewed as a signal of a "technical shift" and a rare moment of resonance within the domestic computing power industry [4][13]. Technical Details - The "UE8M0 FP8" format introduced by DeepSeek is designed to enhance the performance of AI models by compressing data to 8 bits, which reduces memory bandwidth pressure during training and inference [5][11]. - The UE8M0 format allows for efficient scaling of data blocks, significantly improving the dynamic range and reducing the risk of information loss during processing [10][12]. - The architecture of existing domestic AI accelerators is evolving to support the MX format, which is essential for overcoming memory bandwidth limitations [12][14]. Competitive Landscape - The release of UE8M0 FP8 signifies that domestic chip manufacturers have achieved a comprehensive integration of floating-point formats, compiler optimizations, and training framework adaptations [14]. - Major domestic chip companies, including Moore Threads and Cambrian, are already in contact with DeepSeek and are positioned to benefit from this new technology [15][16]. Future Outlook - DeepSeek's initiative is seen as a step towards reducing reliance on foreign technology, potentially redefining the efficiency of large model computing power and enhancing the international standing of domestic chip manufacturers [17].

Core Viewpoint - The release of DeepSeek V3.1 and its mention of the next-generation domestic chip architecture has caused significant excitement in the AI industry, leading to a surge in stock prices of domestic chip companies like Cambricon, which saw an intraday increase of nearly 14% [4][29]. Group 1: DeepSeek V3.1 and UE8M0 FP8 - DeepSeek V3.1 utilizes the UE8M0 FP8 parameter precision, which is designed for the upcoming generation of domestic chips [35][38]. - UE8M0 FP8 is based on the MXFP8 format, which allows for a more efficient representation of floating-point numbers, enhancing performance while reducing bandwidth requirements [8][10][20]. - The MXFP8 format, defined by the Open Compute Project, allows for a significant increase in dynamic range while maintaining an 8-bit width, making it suitable for AI applications [8][11][20]. Group 2: Market Reaction and Implications - Following the announcement, the semiconductor ETF rose by 5.89%, indicating strong market interest in domestic chip stocks [4]. - Cambricon's market capitalization surged to over 494 billion yuan, making it the top stock on the STAR Market, reflecting investor optimism about the company's capabilities in supporting FP8 calculations [29][30]. - The adoption of UE8M0 FP8 by domestic chips is seen as a move towards reducing reliance on foreign computing power, enhancing the competitiveness of domestic AI solutions [33][34]. Group 3: Domestic Chip Manufacturers - Several domestic chip manufacturers, including Cambricon, Hygon, and Moore Threads, are expected to benefit from the integration of UE8M0 FP8, as their products are already aligned with this technology [30][32]. - The anticipated release of new chips that support native FP8 calculations, such as those from Huawei, is expected to further strengthen the domestic AI ecosystem [30][33]. - The collaboration between DeepSeek and various domestic chip manufacturers is likened to the historical "Wintel alliance," suggesting a potential for creating a robust ecosystem around domestic AI technologies [34].

Core Insights - The article highlights the emergence of domestic GPU companies, particularly Moores Threads, as they approach a new phase of technological and capital convergence, with the company preparing for an IPO on the Sci-Tech Innovation Board [1][10] - Moores Threads has established itself as a significant player in the semiconductor industry by focusing on self-developed "full-function GPUs," which are gradually reaching international standards in terms of computing power, versatility, and ecosystem compatibility [1][3] Industry Overview - The global GPU market is projected to reach 36,119.74 billion yuan by 2029, with China's GPU market expected to grow to 13,635.78 billion yuan, increasing its global market share from 30.8% in 2024 to 37.8% in 2029 [1] - The AI chip market in China is anticipated to surge from 1,425.37 billion yuan in 2024 to 13,367.92 billion yuan by 2029, with a compound annual growth rate of 53.7% [8] Company Development - Moores Threads has developed the MUSA architecture, which is the first domestic architecture to support AI computing, graphics acceleration, and physical simulation on a single chip, marking a significant technological breakthrough [4][6] - The company has launched several products, including the MTT S80 consumer graphics card and the MTT S5000 AI computing card, which have shown competitive performance against international counterparts [4][5] Financial Performance - Moores Threads' revenue has increased from 45.84 million yuan in 2022 to 432 million yuan in 2024, reflecting a compound annual growth rate of 208% [9] - The company's AI computing business accounts for 77.6% of its revenue, driven by the demand for large model training and GPU cloud services [9] Strategic Positioning - The company has a strong focus on R&D, with an investment of 1.359 billion yuan in 2024, resulting in a research expense ratio of 309.88% [6] - Moores Threads has secured 450 patents, including 442 domestic patents, which positions it well within the GPU intellectual property landscape [6] Market Opportunities - The company is well-positioned to capitalize on the growing demand for GPUs in various sectors, including AI, digital twins, autonomous driving, and virtual reality [8][11] - The ongoing IPO process is expected to provide Moores Threads with additional capital and resources to support its technological advancements and market expansion [10][11]

Group 1: Core Insights - The company, Moer Thread, has submitted its IPO application to the Sci-Tech Innovation Board, marking its entry into the capital market and revealing its growth trajectory and core competitiveness [1] - The latest prospectus showcases Moer Thread's technological accumulation and commercialization potential in the full-function GPU field, highlighting its all-stack technical barriers and impressive growth curve [1][2] Group 2: Technological Advantages - Moer Thread has built a comprehensive technical moat from chip to cluster, with its MUSA architecture supporting AI computing acceleration, graphics rendering, physical simulation, scientific computing, and ultra-high-definition video encoding [2][3] - The MUSA architecture signifies Moer Thread's technological maturity in the domestic full-function GPU sector, with its flagship AI product, MTT S5000, aiming to compete with NVIDIA's H100 in performance [3][4] Group 3: Market Performance - From 2022 to 2024, Moer Thread's revenue surged from 46.09 million to 438.46 million, achieving a compound annual growth rate (CAGR) of 208.44%, reflecting its strong market competitiveness [6] - Despite being in a loss-making position, the company has narrowed its net loss from -1,839.55 million in 2022 to -1,491.94 million in 2024, indicating a 19% reduction in losses [6] Group 4: Strategic Positioning - The company emphasizes strategic losses as a means to build competitive barriers, similar to Tesla's approach, where initial losses are viewed as investments in future market dominance [7][10] - Moer Thread's R&D investment efficiency has improved significantly, with revenue per unit of R&D expenditure increasing from 0.041 to 0.323, a rise of approximately 7.88 times [7] Group 5: Future Prospects - The company plans to raise 8 billion for R&D of three new-generation chips, targeting improvements in computing efficiency, AI computing capabilities, and the development of a self-controlled ecosystem [10] - Moer Thread's path indicates a potential transition from a technology investment phase to a value return phase, with expectations of reaching a J-curve inflection point for profitability [8][10] Group 6: Conclusion - Moer Thread is positioned to break through in the GPU sector, leveraging rapid architectural iterations and a robust patent portfolio to establish a technological moat [11] - The company aims to convert its technological advantages into commercial barriers, potentially leading to a new growth phase [11]

Group 1: Core Insights - The report highlights the trend of domestic GPU companies, specifically Muxi and Moer Thread, receiving approval for their IPO applications on the Sci-Tech Innovation Board, indicating a clear trend towards autonomous computing power in AI [3][4] - Muxi's product matrix includes the Xisi N series for intelligent computing inference, the Xiyun C series for training and general computing, and the Xicai G series for graphics rendering, with cumulative GPU sales exceeding 25,000 units as of Q1 2025 [7][8] - Moer Thread has launched a comprehensive product matrix covering AI intelligent computing, professional graphics acceleration, and desktop graphics acceleration, with cumulative revenue from AI intelligent computing products reaching 336 million [33][34] Group 2: Muxi Company Overview - Muxi's self-developed architecture is compatible with CUDA, and the company has successfully implemented a thousand-card cluster, with plans for a ten-thousand-card cluster [7][13] - The flagship product, Xiyun C500, is set for mass production in February 2024, while the C600 and C700 are expected to follow in 2026 [8][10] - Muxi's management team has a strong background in AMD GPGPU development, contributing to a robust software and hardware team [10][11] Group 3: Moer Thread Company Overview - Moer Thread's product matrix includes four generations of GPU architectures, with the MTT S4000 and MTT S5000 set to enhance performance through innovations like FP8 precision support [33][34] - The company has developed a high-bandwidth interconnect technology, supporting thousands of interconnected cards, with the KUAE2 cluster expected to launch by the end of 2024 [33][34] - Moer Thread's market strategy focuses on both consumer and enterprise markets, aiming to cultivate a developer ecosystem while strengthening vertical solutions in key industries [33][34] Group 4: Market Strategy and Industry Outlook - Muxi's market strategy is defined by a "1+6+X" model, targeting six key industries including education, finance, transportation, energy, healthcare, and entertainment, while also exploring long-tail markets [26][28] - The report anticipates that AI computing will drive the localization of core supply chain segments, with several companies identified as potential beneficiaries of this trend [3][4] - The report emphasizes the importance of domestic supply chain stability and technological advancement in the semiconductor industry, particularly in the context of Muxi's and Moer Thread's growth [26][29]