11月25日晚间，摩尔线程智能科技（北京）股份有限公司（以下简称"摩尔线程"，688795.SH）披露 《网下初步配售结果及网上中签结果公告》。 网下初步配售结果显示，网下有效申购股数合计704.06亿股，获配股初步配售股数总额为3920万股。其 中，公募基金、社保基金、养老金、年金基金、银行理财产品、保险资金、保险资产管理产品、合格境 外投资者资金获配股初步配售股数约3858.96万股，占网下最终发行比例的98.44%，其他类型投资者占 比1.56%。 本次参与摩尔线程战略配售的投资者阵容强大，主要包括四类机构：第一类是参与科创板跟投的保荐人 相关子公司——中信证券投资有限公司。根据科创板跟投机制安排，保荐机构子公司以自有资金参与战 略配售，是对发行人价值判断和未来成长潜力的重要背书，也体现出保荐机构对摩尔线程技术实力、商 业模式及长期发展空间的高度认可。 依托"AI+图形"双线驱动的技术布局，摩尔线程构建起坚实的核心壁垒，成为国内极少数同时覆盖B端 与C端市场的全功能GPU企业，凸显出极强的行业稀缺性与市场竞争力。截至2025年6月，摩尔线程已 获得授权专利514项，授权数量位居国内GPU企业前列，展现出 ...

Core Insights - AI SoC is optimized for AI tasks, providing efficient parallel computing capabilities, particularly suitable for machine learning and deep learning applications [1][5] - The global AI SoC shipment volume is projected to grow from 949 million units in 2020 to 1.565 billion units in 2024, achieving a compound annual growth rate (CAGR) of 13.3% [1][11] - The demand for edge AI SoC is driven by advancements in large language models and multimodal models, leading to a surge in edge-side inference requirements [1][11] AI SoC Industry Overview - AI SoC integrates specialized functional modules for AI computation on top of traditional SoC, distinguishing itself with AI accelerators like NPUs [1][5] - The AI SoC industry consists of upstream software suppliers, IP providers, raw material suppliers, and equipment suppliers, with a focus on design, manufacturing, and testing in the midstream [1][5][6] Market Trends - The visual AI SoC segment is growing, with shipments expected to reach 246 million units in 2024, a year-on-year increase of 52.9%, accounting for 15.7% of total AI SoC shipments [1][13] - In China, the AI SoC market is projected to reach approximately 280 million units by 2024, supported by government policies promoting AI integration across various sectors [1][14] Future Projections - AI SoC shipment volumes are expected to continue growing, with projections indicating over 4.5 billion units by 2030 due to increasing demands for high-performance computing and AI inference capabilities [1][16] - The product categories within the AI SoC market are anticipated to diversify, focusing on AI processing capabilities and addressing varying requirements across different devices [1][16]

Group 1 - The core point of the article highlights that Moer Thread has become the fastest company to pass the review for the Sci-Tech Innovation Board, achieving this in just 88 days from acceptance to approval [1][2] - The company plans to raise 8 billion yuan through its IPO, which will be fully allocated to the development of AI training and inference chips, graphics GPUs, and AI SoCs [2] - The market is intrigued by Moer Thread's potential as a "domestic Nvidia," but there are concerns regarding its fundamentals and valuation in the face of competition from established players like Nvidia [2][3] Group 2 - Moer Thread, established in 2020, is one of the few companies in China with full GPU development capabilities, showing significant advantages in general computing and graphics rendering [3] - The company has faced scrutiny from regulators regarding five major risks, including continuous losses, fluctuating gross margins, potential intellectual property risks, customer concentration, and high inventory levels [3] - The global GPU market is characterized by a dominant player, Nvidia, which holds approximately 90% of the AI chip market, while the domestic market is fragmented [3] Group 3 - Moer Thread's revenue has seen substantial growth, increasing from 46 million yuan in 2022 to 1.24 billion yuan in 2023, and projected to reach 4.38 billion yuan in 2024, reflecting a compound annual growth rate of 208% [6] - The company's net losses have narrowed from 1.84 billion yuan in 2022 to 1.49 billion yuan in 2024, with a significant reduction in losses to 271 million yuan in the first half of 2025 [7] - The gross margin has improved significantly, rising to over 70%, which is higher than the average of 38% for semiconductor companies on the Sci-Tech Innovation Board [7] Group 4 - Despite the improvements, Moer Thread still has accumulated losses of approximately 1.48 billion yuan and does not expect to achieve profitability until at least 2027 [8] - The company has a manageable short-term debt pressure with cash reserves of 4.9 billion yuan by the end of 2024 [9] - Moer Thread has successfully launched four generations of GPU architectures and has expanded its product matrix to cover AI computing, cloud computing, and personal computing applications [9]

Core Viewpoint - The rapid approval of Moer Technology's IPO on the Sci-Tech Innovation Board marks a significant milestone, being the fastest in the board's history, with a fundraising target of 8 billion yuan for AI chip development [1][3]. Company Overview - Moer Technology, established in 2020, is one of the few companies in China with full GPU development capabilities, excelling in general computing and graphics rendering to meet the evolving demands of AI models [4]. - The company has applied for an IPO on the Sci-Tech Innovation Board and has faced regulatory scrutiny regarding its financial health and market position [5]. Financial Performance - Moer Technology's revenue has shown remarkable growth, increasing from 46 million yuan in 2022 to 1.24 billion yuan in 2023, and projected to reach 4.38 billion yuan in 2024, reflecting a compound annual growth rate of 208% [8]. - The company's net profit loss has narrowed from 1.84 billion yuan in 2022 to 1.49 billion yuan in 2024, with a significant reduction to 271 million yuan in the first half of 2025 [8]. - The gross margin has improved significantly, rising to over 70%, compared to the average of 38% for semiconductor companies on the Sci-Tech Innovation Board [8]. Market Position and Competition - The global GPU market is dominated by Nvidia, which holds approximately 90% of the AI chip market share, while the domestic market is characterized by a fragmented landscape [5][6]. - Moer Technology acknowledges the gap in R&D strength and technology compared to Nvidia, but sees significant potential in its "full-function" positioning within the domestic market [6]. Product Development and Strategy - The company has successfully launched four generations of GPU architectures and has developed a product matrix covering AI computing, cloud computing, and personal computing applications [10]. - The revenue from AI computing products is expected to constitute a significant portion of the company's income, with projections of 336 million yuan in 2024 and 665 million yuan in the first half of 2025, indicating a strategic shift towards high-margin, high-performance segments [10].

Summary of Conference Call Records Industry Overview - The domestic computing chip industry is experiencing a long-term investment opportunity driven by domestic substitution and market demand growth, with companies like Cambricon, Huawei, and Haiguang showing development potential alongside self-developed ASICs from major firms [1][2][3] Key Points and Arguments - **Revenue Growth and Financial Performance**: - Moer Thread's IPO raised 8 billion, with projected revenues of 438 million in 2024, while Muxi raised 4 billion with projected revenues of 743 million in 2024. Both companies have high gross margins despite negative net profits due to upfront semiconductor design and R&D costs [1][3][4] - **Competitive Positioning**: - Moer Thread's latest Pinghu architecture chip has an IP 32 computing power close to NVIDIA's H20, with an interconnect bandwidth of 800GB per second and memory capacity of 80GB, indicating strong competitiveness against NVIDIA's high-end chips [1][5] - **Policy Impact**: - Recent policies from the National Cyberspace Administration of China require companies like Alibaba and ByteDance to stop testing and purchasing NVIDIA's RTX PRO6,000D, indicating a push for domestic processors that have reached or surpassed NVIDIA's performance levels [1][6] - **Market Growth Projections**: - The domestic computing chip market is expected to see significant growth, with early estimates predicting a market size of 200 billion, potentially reaching 800 billion by 2027 according to NVIDIA [3][8][9] - **Supply Chain and Production Capacity**: - By 2026, domestic computing chip manufacturers are expected to scale up production significantly due to resolved supply chain issues and capacity releases, with major internet companies beginning to test and adopt domestic chips [1][7] Additional Important Insights - **Product Development and Competitiveness**: - Moer Thread has enhanced AI training and inference capabilities in its products, with significant improvements in parameters and performance, indicating a positive trend in product strength and supply chain progress [5][11] - **Market Dynamics**: - The coexistence of self-developed ASICs and traditional chip manufacturers like Huawei and Haiguang is feasible due to the large market size and low domestic penetration rates, providing ample growth opportunities [2][10] - **End-Side Computing Trends**: - Companies like Rockchip are positioned well in the end-side computing sector, with recent product launches showing better performance and price competitiveness compared to Qualcomm and NVIDIA [12][13][15] - **3D Stacking Technology**: - 3D stacking technology is becoming increasingly important for high-bandwidth hardware in model training and inference, with companies like Zhaoyi holding a strong position in this area [17]

Investment Rating - The report does not explicitly state an investment rating for MediaTek, but it highlights solid performance and growth potential in AI SoC and ASIC businesses, suggesting a positive outlook for the company. Core Insights - MediaTek reported 2Q25 revenue of NTD 150.4 billion, slightly below market consensus of NTD 151.2 billion, representing a 1.9% QoQ decrease and an 18.1% YoY increase. EPS reached NTD 17.5, exceeding market expectations. The gross margin was 49.1%, up 1 percentage point QoQ, aided by a one-off contribution from a long-term supply agreement [1][5][9]. - The ASIC business is a significant growth driver, with a total addressable market (TAM) exceeding USD 40 billion. MediaTek is expanding its ASIC R&D resources and expects substantial revenue contributions starting in 2026, with a target of reaching USD 1 billion in revenue by the end of 2026 [2][6][10]. - MediaTek is actively investing in 2nm process technology, with the first 2nm chip expected to tape out in September 2023, which will enhance performance and power efficiency for both SoC and ASIC businesses [3][7][10]. - Collaboration with NVIDIA continues to deepen, with the GB10 chip entering mass production in Q3, which is expected to contribute to revenue growth [4][8][10]. - For 3Q25, MediaTek expects revenue between NTD 130.1 billion and NTD 140 billion, a QoQ decrease of 7% to 13%, but a YoY increase of 10% to 18%. The gross margin is projected at 47% [9][10]. Summary by Sections Financial Performance - 2Q25 revenue: NTD 150.4 billion, -1.9% QoQ, +18.1% YoY [1][5] - EPS: NTD 17.5, exceeding expectations [1][5] - Gross margin: 49.1%, +1 ppt QoQ [1][5] ASIC Business - Strong performance in enterprise ASIC sector with TAM over USD 40 billion [2][6] - Rapid expansion of ASIC R&D resources [2][6] - Expected revenue contributions starting in 2026, targeting USD 1 billion by end of 2026 [2][6][10] 2nm Process Development - Active investment in 2nm process technology [3][7] - First 2nm chip expected to tape out in September 2023 [3][7] Collaboration with NVIDIA - Ongoing collaboration in chip business and automotive cockpit chips [4][8] - GB10 chip to enter mass production in Q3 [4][8] Future Guidance - 3Q25 revenue guidance: NTD 130.1 billion to NTD 140 billion [9][10] - Expected gross margin: 47% [9][10]

Investment Rating - The report indicates a positive investment outlook for the integrated machine and AI SoC sectors, particularly in the context of AI model deployment and local data security needs [3][11]. Core Insights - The demand for integrated machines has surged following the release of the Deep Seek AI model, with significant interest from government and enterprise sectors due to their data security requirements [3]. - The integrated machine is designed for AI model applications, effectively shortening deployment cycles and lowering barriers to entry, with a projected demand of 70,000 units in the government and enterprise sectors by 2025, translating to a market size of 54 billion yuan [3][11]. - The Deep Seek distilled small parameter models demonstrate excellent performance on terminal devices, achieving real-time question answering with a 1.5B model and suitable for text summarization and image description with 7B/8B models, requiring a minimum of 4-5GB of memory under INT4 quantization [3][5]. - Domestic computing power is expected to play a crucial role in the integrated machine sector, aligning well with mainstream downstream demands, although challenges in software-hardware collaboration remain [3][7]. - Quantization techniques are highlighted as a means to reduce AI hardware costs by converting model parameters from 16-bit floating-point to 8-bit integers, thus decreasing model size and computational complexity [3][8][9]. - The report notes a significant reduction in AI inference costs, which is driving the trend towards edge computing, with lightweight AI hardware gaining advantages in both edge and cloud environments [3][15]. - In the smart automotive sector, companies are integrating AI technologies to enhance smart cockpit functionalities, with BYD leading the way in adopting new AI chips in its vehicles [3][17]. - The domestic automotive chip sector is making strides, with companies like Yikatong collaborating with Volkswagen to export new SOC chips, indicating a growing acceptance of domestic chips in international markets [3][20]. Summary by Sections Integrated Machines - Integrated machines are tailored for AI model applications, providing a plug-and-play computing solution that meets high data security requirements for sectors like government and finance [4][10]. - The projected demand for integrated machines in the Chinese server market is expected to reach approximately 70,000 units by 2025, with a market potential of 54 billion yuan [11]. AI Model Performance - The Deep Seek distilled models are effective in reducing hardware resource requirements while maintaining performance, making them suitable for various applications [5][16]. Domestic Computing Power - The report emphasizes the importance of domestic computing power in the integrated machine sector, with a need to overcome challenges related to precision support in AI chips [7]. Cost Reduction Techniques - Techniques such as fixed-point quantization are crucial for lowering AI hardware costs and improving overall efficiency [8][9]. Smart Automotive Sector - The integration of AI in smart vehicles is on the rise, with significant advancements in automotive chip technology and collaborations with international partners [17][20].