Group 1 - The Trump administration announced the formation of the "US Tech Force," a research team of 1,000 engineers and experts, inviting major tech companies like Apple, NVIDIA, Oracle, Google, Dell, Microsoft, and OpenAI to participate in advancing AI infrastructure and digital transformation in the US [1][2] - The initiative is seen as a strong endorsement of AI development in the US, aiming to mitigate concerns about an "AI bubble" and ensuring robust growth in the sector under US dominance [1] - Taiwanese server manufacturers, including Hon Hai, Quanta, and Wistron, are expected to benefit significantly from the upcoming large orders related to AI infrastructure, as they hold over 90% of the global market share in AI server manufacturing [1] Group 2 - Hon Hai's global market share in AI servers exceeds 40%, positioning it as a key partner for the US in AI development, with expectations of increased market share driven by upcoming product launches from NVIDIA [1] - Quanta is optimistic about the AI server business, with visibility of orders extending at least until the end of 2026, and plans to expand AI server production capacity to meet strong customer demand [2] - The "US Tech Force" aims to recruit 1,000 engineers and experts with annual salaries ranging from $150,000 to $200,000, reflecting a significant effort to modernize the US workforce in response to the rapid expansion of AI ecosystems, particularly in competition with China [2]

外资高盛引述鸿海管理阶层说法，AI服务器机柜的良率已达到令人满意的水准，目标是持续缩短生产 周期。随着客户对美国在地生产的需求增长，鸿海计划训练更多当地技术工人。另外，鸿海也需要确保 充足的电力供应，计划提高生产自动化，以缓解对技术工人需求持续增加的压力。 微软、Google、亚马逊AWS及Meta等美国科技巨头持续扩大AI布局，现在特朗普政府也加足马力冲刺 AI基础建设，使得美国当地设计及代工制造AI服务器全面成为市场趋势，鸿海、纬创（3231）、纬 颖、和硕等台湾代工大厂"美国制造"产能逐步到位，准备好大啖当地商机。 AI服务器代工龙头鸿海在北美布局相当深厚，其中，墨西哥厂2025年仍是集团最大AI服务器生产基 地，随着产能持续扩张，鸿海预期，到2026年，美国将成为最大的AI服务器生产基地。虽然美国的生 产成本较高，鸿海预期客户将协助分担额外成本。 业界分析，微软、Google、亚马逊AWS、Meta及甲骨文等美国云端服务大厂（CSP）相继宣布将斥资逾 千亿美元打造AI服务器运算中心，其中将有相当大比例落在美国设厂，加上美国政府将AI算力视为国 安，将顺势推动AI服务器代工制造地点扩大延伸到美国。 ...

大摩开门会：中国 AI 算力的供给及需求 251212 摘要 中国 AI 算力供不应求，尽管算法优化，AI 效率芯片仍紧缺。腾讯、阿里 巴巴等企业面临 GPU 供应限制，资本市场担忧芯片短缺，未来三年混合 云厂商资本支出预计增长至 4,500 亿人民币，但仍低于美国大厂水平。 H200 芯片在性能上领先，国内尚无替代品，需求量大。监管层或允许 企业采购 H200，以审批形式平衡模型玩家需求与国产芯片发展，放开 H200 供应将利好云厂商及下游模型和应用。 放开 H200 供应后，国内数据中心需求预计显著提高，过去需求受芯片 供给限制。推荐腾讯、阿里巴巴等超大型公司，以及 GDS、VNET 等数 据中心公司。 受 GPU 限制，中国云厂商倾向于在海外数据中心进行模型训练，国内数 据中心主要用于推理。台积电为中国市场生产芯片的计划将延续至 2026 上半年，每个 H200 中端单元成本约 1,300 美元。 中国需加强本土化生产能力，天生、阿里巴巴平头哥、百度昆仑等企业 开发专用集成电路（ASIC），调整设计以符合出口管制，减少对进口依 赖。中芯国际计划到 2027 年月产能达 3-4 万晶圆。 在 H200 板 ...

Core Viewpoint - The article discusses the AI industry chain, focusing on infrastructure, algorithms, and applications, while providing insights from recent reports by Morgan Stanley and JP Morgan regarding ODM manufacturers' performance and shipment forecasts. ODM Manufacturers' Performance and Shipment Analysis - Morgan Stanley ranks ODM manufacturers for GPU AI servers as Wistron > Hon Hai > Quanta [2][18] - Morgan Stanley's latest forecast for GB200/300 rack shipments is adjusted to 27,300 units, down from 28,000 units, primarily due to updates following Quanta's Q3 earnings call [2] - Quanta's management indicates a conservative outlook for AI revenue growth in Q1 2026, leading to a downward adjustment of their Q4 2025 rack shipment forecast from approximately 3,500 to 2,500 units [7] - Despite Quanta's adjustment, Wistron shows strong growth, leading to a slight increase in overall rack shipment forecasts for Q4 2025, from 8,000-8,500 to 13,500-14,000 units [7] Company-Specific Revenue Insights - Quanta reported November revenue of approximately NT$193 billion, with a month-on-month increase of 11% and a year-on-year increase of 36%, driven by GB200/300 rack shipments expected to reach 1,000-1,100 units [13] - Wistron achieved a record revenue of NT$281 billion in November, with a month-on-month increase of 52% and a year-on-year increase of 195%, attributed to significant increases in L10 computing tray shipments [14] - Hon Hai's November GB200 rack shipments remained stable at approximately 2,600 units, with expectations of a decline in December due to year-end holidays, maintaining a forecast of 7,200 units for Q4 2025 [15] 2026 Preliminary Outlook - The forecast for rack shipments in 2026 is challenging, but Morgan Stanley has adjusted its estimate to 70,000-80,000 units, up from 60,000-70,000 units, based on anticipated inventory carryover of approximately 2 million Blackwell chips [17] - Morgan Stanley maintains the ranking of ODM manufacturers as Wistron > Hon Hai > Quanta, noting that actual deliveries may be lower than predicted due to assembly and testing times for L11 racks not being included in the estimates [18]

经济观察网鸿海将向高雄项目投资159亿台币。 ...

Core Insights - The AI server hardware is undergoing a significant design upgrade driven by GPUs and ASICs, with new platforms from NVIDIA and AMD expected to enhance computing power and cabinet density by 2026 [1][4] - The demand for AI servers is surging, with NVIDIA's platform projected to see cabinet demand more than double from approximately 28,000 units in 2025 to at least 60,000 units in 2026 [2] - The transition to advanced cooling and power solutions is essential as GPU power consumption increases, leading to a complex engineering system for future AI data centers [4][5] AI Server Demand - The demand for AI servers continues to rise, with NVIDIA's platform expected to see cabinet demand increase significantly [2] - AMD's Helios server rack project is also making progress, further intensifying the market demand for advanced AI hardware [2] NVIDIA AI Server Roadmap - NVIDIA's upcoming AI server platforms include the GB300 and Vera Rubin series, with increasing power consumption from 700W for H100 to 3700W for the upcoming VR200 platform [4][22] - The GB200 platform is currently the core driver of the AI server market [3] Supply Chain and Production - Major ODM manufacturers like Hon Hai, Quanta, Wistron, and Wiwynn are key suppliers for NVIDIA's AI server cabinets, with Hon Hai leading in production [10] - Hon Hai's AI server cabinet shipments increased by 300% quarter-over-quarter, with expectations to exceed NT$1 trillion in revenue by 2025 [10] Revenue and Market Share - In November, Quanta, Wistron, and Wiwynn reported record monthly revenues, with significant year-over-year growth [12] - By 2025, Hon Hai is expected to capture over 40% of the AI server market share, with GB200 and GB300 platforms dominating shipments [12] Industry Upgrades - The shift to NVIDIA's GB300 and Vera Rubin platforms represents a new cycle in AI hardware, prompting a reevaluation of supply chain components [13] - The power supply and cooling systems are undergoing significant changes to accommodate the increasing power demands of AI workloads [14][16] Power Supply Innovations - NVIDIA's Kyber power supply strategy aims to redefine data center power architecture, with projected values for Rubin Ultra cabinets expected to exceed ten times that of current GB200 cabinets by 2027 [16] - The transition to 800V DC power solutions is critical for meeting the demands of modern AI data centers [14][16] Cooling Solutions - NVIDIA's cooling technology is evolving from air cooling to liquid cooling solutions to manage the rising thermal demands of high-performance GPUs [17][22] - The value of cooling components for AI servers is expected to increase significantly, with projections indicating a 17% rise for the next generation of cooling modules [22] PCB Demand Surge - The upgrade of AI servers is driving a surge in demand for high-end PCBs, with manufacturers increasing production capacity for advanced multi-layer boards [23][24] - The market for high-end PCBs is expected to grow significantly, with prices for advanced boards potentially doubling due to increased complexity [23] CSP Capital Expenditure - Major cloud service providers (CSPs) are ramping up capital expenditures to support the growing demand for AI infrastructure, with projected increases in spending for 2025 and 2026 [26][29] - The total capital expenditure for the eight major CSPs is expected to exceed $600 billion in 2026, reflecting strong long-term growth potential in AI infrastructure [26][29]

Core Insights - 2026 is identified as a critical window for AI server system upgrades, driven by significant design changes in GPU and ASIC technologies [1][4] - The demand for AI servers is expected to surge, with NVIDIA's platform projected to see cabinet demand more than double from approximately 28,000 units in 2025 to at least 60,000 units in 2026 [2][26] - The overall cost of AI servers is anticipated to rise significantly due to advancements in power supply, cooling solutions, and PCB requirements [5][26] Group 1: AI Server Hardware Upgrades - NVIDIA is set to launch the GB300, Vera Rubin platform, and Kyber architecture in 2026, enhancing computational power and cabinet density [1][4] - The GPU power design is evolving, with TDP increasing from 700W for H100 to 3700W for VR200 NVL44 CPX by late 2026, necessitating a shift to liquid cooling solutions [4][23] - The transition to more efficient power systems is underway, moving from 12V VRM to 48V DC bus systems to reduce conversion losses [4][14] Group 2: Market Demand and Supply Chain Dynamics - ODM manufacturers like Hon Hai, Quanta, Wistron, and Wiwynn are ramping up production, with Hon Hai's AI server cabinet shipments increasing by 300% quarter-over-quarter [10][12] - In November, Quanta and Wistron reported record monthly revenues, with Wistron showing a remarkable 194.6% year-over-year growth [12] - The market share for AI server cabinets in 2025 is projected to see Hon Hai holding over 52%, with Quanta and Wistron at approximately 19% and 21%, respectively [13] Group 3: Power and Cooling Solutions - NVIDIA's Kyber project aims to redefine power supply architecture for AI data centers, with a target to produce new power solutions by the end of 2026 [15][17] - The cooling technology is evolving from air cooling to liquid cooling, with the GB300 adopting a full cold plate liquid cooling solution to handle up to 1400W [18][23] - The cost of cooling components is expected to rise, with the total value of cooling components for the next-generation Vera Rubin platform projected to increase by 17% [23] Group 4: PCB and Component Upgrades - The demand for high-end PCBs is surging, with the number of layers and material quality increasing significantly due to the enhanced functionality of AI servers [24][25] - The global PCB market is expected to grow, with high-end HDI boards and multi-layer boards seeing demand increases of 14.2% and 18.5%, respectively [25] - The price of PCBs is anticipated to double with each upgrade cycle, reflecting the growing complexity and performance requirements of AI hardware [25] Group 5: Capital Expenditure Trends - Major cloud service providers (CSPs) are increasing their capital expenditures, with a projected total of over $600 billion in 2026, reflecting a 40% year-over-year growth [26][29] - CSPs like Google, Meta, and Amazon are significantly raising their capital expenditure forecasts for 2025, indicating strong demand for AI infrastructure [29] - The ongoing investment from CSPs provides a solid foundation for the rising costs associated with AI server upgrades [26][29]

2025 年 12 月 10 日 晨会纪要 研究所： 证券分析师： 余春生 S0350513090001 yucs@ghzq.com.cn [Table_Title] 晨会纪要 ——2025 年第 210 期 观点精粹: 最新报告摘要 证券研究报告 1、最新报告摘要 1.1、资金供给端改善，解禁减持规模高增--策略周报 分析师：赵阳 S0350525100003 联系人：郭可凡 S0350124070038 资金供给端改善，解禁减持规模高增--策略周报 2027 年订单能见度提高，业绩有望持续高增--工业富联/消费电子(601138/212705) 公司点评 投资要点: 1.本周（2025/12/01-2025/12/05，下同）宏观资金面宽松，央行通过公开市场操作开展 7 天逆回购净回笼 8480 亿元，开展 10000 亿元 3 个月买断式逆回购操作，本周总计实现资金净投放 1520 亿元。资金价格方面，短 端利率下行，中端与长端利率上行，期限利差走阔。 2.股市资金供给端整体改善，权益基金发行回暖。融资余额小幅回升，但杠杆资金参与度有所下降，融资净流 入较多的行业为有色金属、电子，融资净流出较多的行业为 ...

2025 年 12 月 09 日 公司研究 评级：买入(维持) | 研究所： | | | | --- | --- | --- | | 证券分析师： | | 刘熹 S0350523040001 | | | | liux10@ghzq.com.cn | | 联系人 | ： | 唐锦珂 S0350125070014 | | | | tangjk@ghzq.com.cn | 最近一年走势 [Table_Title] 2027 年订单能见度提高，业绩有望持续高增 ——工业富联（601138）公司点评 | 相对沪深 300 | 表现 | | 2025/12/09 | | --- | --- | --- | --- | | 表现 | 1M | 3M | 12M | | 工业富联 | -6.5% | 39.4% | 213.3% | | 沪深 300 | -1.7% | 3.7% | 15.9% | | 市场数据 | | | 2025/12/09 | | 当前价格（元） | | | 67.99 | | 52 周价格区间（元） | | | 14.58-83.88 | | 总市值（百万） | | | 1,350,157.92 | ...

Core Viewpoint - Hon Hai's revenue for the first 11 months of the year reached NT$7.23 trillion, surpassing NT$7 trillion for the first time and exceeding the full-year revenue forecast of NT$6.86 trillion for 2024, indicating a record high for the year [1][4]. Revenue Performance - In November, Hon Hai reported revenue of NT$844.3 billion, a month-on-month decrease of 5.7% but a year-on-year increase of 25.5%, marking a record high for the month and the second highest in history [2][5]. - The "Components and Other Products" category showed significant growth due to increased demand for key components, while "Computer Terminal Products" and "Cloud Network Products" remained relatively stable [2][5]. - The "Consumer Electronics" category experienced a slight decline due to a high base effect from the previous month, while the overall revenue for the first 11 months increased by 16.6% year-on-year [2][5]. Future Outlook - For Q4, Hon Hai anticipates continued growth in AI server shipments and a peak season for ICT products, with visibility for the quarter being better than the previous month [6]. - The company aims to produce over 1,000 AI server cabinets weekly, with a target of exceeding 2,000 by the end of 2026, focusing on deepening its presence in the U.S. market [3][6]. - Hon Hai's chairman noted that the company can meet customer demands for both NVIDIA GPU servers and ASIC servers, with a market share exceeding 40% across various platforms [3][6].