瑞银认为，当前科技巨头面临的核心问题是，在AI基础设施大规模投资的背景下，能否维持强劲的资本收益率。市场正从对科技巨头的集体乐观转向基 于基本面的差异化定价。那些能够在AI投资浪潮中保持强劲资本收益率、同时增长预期相对保守的公司，可能提供更好的风险调整后收益。 这些细微的修正 体现了当前市场对科技公司基本面变化的高度敏感性，直接影响了股价表现。 2025年以来，美股科技八巨头（英伟达、微软、苹果、谷歌、亚马逊、Meta、博通、特斯拉）股价表现呈现明显的两极分化：英伟达、博通、微软、 Meta跑赢标普500指数，而苹果、特斯拉、谷歌和亚马逊则明显落后。 瑞银认为，这一分化背后反映了市场对不同公司AI投资回报预期的重新评估。目前的关键在于识别哪些公司能在AI基础设施大规模投资的同时保持强劲 的资本收益率。 AI投资的双刃剑效应 美股科技巨头们正在经历一场前所未有的分化。 瑞银表示，市场分化的背后是投资者对各公司基本面预期的显著差异。 从6个月CFROI（现金流投资回报率）修正数据可以看出，博通获得了最积极的盈利预期修正，而特斯拉和苹果则面临负面修正。谷歌、Meta、微软和 英伟达的CFROI修正相对平稳，表明投资 ...

Summary of Key Points from the Conference Call Industry Overview - **Industry**: Global Technology, specifically focusing on AI Supply Chain and semiconductor industry - **Key Players**: TSMC, Nvidia, Broadcom, AMD, MediaTek, Alchip, Aspeed, Advantest, KYEC, and others Core Insights and Arguments 1. **Cloud Capex Growth**: Morgan Stanley expects 2026 cloud capex to increase by 31% year-over-year (Y/Y) to US$582 billion, significantly higher than the consensus estimate of 16% Y/Y growth [2][10][67] 2. **AI Server Capex**: The implied AI server capex is projected to grow approximately 70% Y/Y in 2026, driven by an increase in AI server capex mix [2][11] 3. **TSMC CoWoS Capacity**: TSMC's 2026 CoWoS capacity is estimated to remain unchanged at 93k wafers per month, which is expected to support the anticipated AI capex growth [1][4][27] 4. **Hyperscalers' Capex**: The top 11 global hyperscalers are tracking a cash capex of US$445 billion for 2025, reflecting a 56% Y/Y growth, with major contributions from Microsoft, Amazon, and Alphabet [9][68] 5. **AI Chip Demand**: Strong demand for AI semiconductors is expected to persist, with Morgan Stanley maintaining an "Overweight" rating on key semiconductor companies [2][11] Additional Important Insights 1. **China's AI Chip Supply**: There are ongoing developments in China's AI chip supply, with Nvidia placing new wafer orders at TSMC for H20 chips, adding 200k units to the previous production of 1 million units [8] 2. **Nvidia's CoWoS Consumption**: Nvidia's CoWoS consumption assumption has been raised from 580k to 595k units for 2026, indicating strong demand for their chips [22] 3. **Market Dynamics**: The report highlights the competitive landscape, noting that Huawei is developing its own AI chips to compete with Nvidia, showcasing the CM384 server rack prototype [8][41] 4. **Depreciation Trends**: Depreciation as a percentage of total expenses for data center customers is expected to rise, reflecting increased investments in data centers [60] 5. **AI Inference Demand**: Monthly tokens processed by major cloud service providers indicate a growing demand for AI inference, with Google processing over 980 trillion tokens in July 2025, doubling from May [15] Conclusion The conference call emphasizes a robust outlook for the AI semiconductor industry, driven by significant increases in cloud capex and AI server investments. TSMC's capacity plans and the competitive dynamics in the AI chip market, particularly with developments in China, are critical factors to monitor.

公众号记得加星标⭐️，第一时间看推送不会错过。 来源：内容 编译自彭博社等 。 人工智能数据中心规模越来越大的时代，让云计算公司面临一个问题：如何处理老旧、规模较小的数 据中心？ 博通公司提供了一个可能的答案，至少对于位置相对接近的运营而言是如此。该公司周一发布的新版 Jericho 网络芯片能够以更高的速度传输更大容量的数据。该公司核心交换事业部高级副总裁兼总经 理 Ram Velaga 表示，这意味着客户可以连接多个小型数据中心，创建一个用于开发或运行人工智能 模型的大型系统。 博通表示，Jericho4 产品能够连接多个数据中心的 100 多万个处理器，处理的信息量约为上一代产 品的四倍。 博通一直受益于对用于构建人工智能系统的设备的需求。其网络组件（例如路由器和交换机）可以引 导图形处理单元（GPU）之间的流量，而这些昂贵的芯片用于创建人工智能模型。 虽然博通的部分设备用于在同一机架或数据中心大楼内移动数据，但也需要能够跨楼宇甚至更远距离 工作的组件。此外，Velaga 表示，巨型 GPU 集群的功耗越来越高，无法将它们全部集中到同一位 置。 "当他们尝试构建一个包含 20 万个甚至 10 万个 GP ...

8月4日，博通公司宣布，专为新一代分布式AI基础设施设计的平台——Jericho4以太网结构路由器已正 式开始出货。该产品能够跨多个数据中心互连超百万个XPU处理器，并凭借突破性的带宽、安全性和无 损传输性能打破传统扩展限制。博通表示，结合Tomahawk 6、Tomahawk Ultra与Jericho4产品线，其为 高性能计算（HPC）和人工智能提供了完整的网络解决方案组合。 ...

博通核心交换事业部高级副总裁兼总经理Ram Velaga指出，Jericho4芯片具备连接超百万个处理器的能 力，数据处理量较前代产品提升约四倍。作为人工智能硬件生态链的关键供应商，博通的网络组件在 GPU集群中承担着数据路由与交换功能，其重要性随着AI算力需求增长而日益凸显。 Velaga特别提到，当前巨型GPU集群的功耗问题已成为技术瓶颈，单个实体建筑难以支撑超过300兆瓦 的电力需求，而分布式架构可有效缓解这一压力。 值得注意的是，企业正将数据中心向用户聚集的城市区域迁移，以缩短数据交互延迟。Velaga解释称， 在空间受限的都市环境中，整合多个小型数据中心更具现实可行性。相较于主要处理单数据中心内机架 间通信(通常覆盖1公里内)的Tomahawk系列产品，Jericho4的突破性优势在于其支持跨区域互联能力，最 远传输距离可达100公里以上。 据博通披露，Jericho4芯片已从即日起向云服务商及网络设备厂商等早期客户发货，相关产品集成与全 面部署预计将在九个月内完成。此次技术升级不仅为云计算企业提供了更灵活的基础设施扩展方案，也 折射出AI时代数据中心向"分布式+近用户端"转型的行业趋势。 这款芯片 ...

A new Broadcom chip will help power AI by linking up smaller data centers https://t.co/gTWSIF7d0E ...

Core Insights - Broadcom Inc. has launched the Jericho4 ethernet fabric router, designed for the next generation of distributed AI infrastructure, capable of interconnecting over one million XPUs across multiple data centers [1][2] - The Jericho4 router addresses the increasing infrastructure demands of AI models, which exceed the capabilities of a single data center, by providing high-bandwidth, secure, and lossless transport across regional distances [2][3] Product Features - Jericho4 can scale to 36,000 HyperPorts, each operating at 3.2 Tb/s, with features such as deep buffering and line-rate MACsec for enhanced security and performance over distances exceeding 100 km [3][4] - The router utilizes Broadcom's 3.2T HyperPort technology, consolidating four 800GE links into a single logical port, which improves utilization by up to 70% and streamlines traffic flow [4][5] - Jericho4 supports MACsec encryption on every port at full speed, ensuring data security without compromising performance, even under high traffic loads [5][6] Compliance and Interoperability - The Jericho4 router is compliant with specifications from the Ultra Ethernet Consortium (UEC), ensuring interoperability across open, standards-based Ethernet AI fabrics [6][9] - This compliance allows for seamless integration with a wide ecosystem of UEC-compliant NICs, switches, and software stacks, enhancing its utility in diverse networking environments [6][9] Market Position and Collaboration - Broadcom's Jericho4 is positioned to meet the growing demands of distributed AI workloads, with industry leaders expressing confidence in its capabilities to enhance scalability and efficiency in AI networking [11][12][13] - Collaborations with companies like Accton, Arista Networks, and DriveNets highlight the router's potential to support advanced AI infrastructure and improve energy efficiency in large-scale GPU clusters [11][12][13][14] Availability - Jericho4 is currently sampling to customers, indicating its readiness for market deployment [8]

Core Insights - Broadcom Inc. has launched the Jericho4 ethernet fabric router, designed for the next generation of distributed AI infrastructure, capable of interconnecting over one million XPUs across multiple data centers [1][2] - The Jericho4 router addresses the increasing infrastructure demands of AI models, which exceed the capabilities of a single data center, by providing high-bandwidth, secure, and lossless transport over regional distances [2][3] Product Features - Jericho4 can scale to 36,000 HyperPorts, each operating at 3.2 Tbps, featuring deep buffering and line-rate MACsec for enhanced security and performance over distances exceeding 100 km [3][4] - The router utilizes Broadcom's 3.2T HyperPort technology, consolidating four 800GE links into a single logical port, which improves utilization by up to 70% and streamlines traffic flow [4][5] - Jericho4 supports MACsec encryption on every port at full speed, ensuring data security without compromising performance, even under high traffic loads [5][6] Compliance and Interoperability - The Jericho4 is compliant with specifications from the Ultra Ethernet Consortium (UEC), ensuring interoperability across open, standards-based Ethernet AI fabrics [6][9] - This compliance allows for seamless integration with a wide ecosystem of UEC-compliant NICs, switches, and software stacks, enhancing its utility in diverse networking environments [6][9] Market Position and Collaboration - Broadcom's Jericho4 is positioned to meet the growing demands of distributed AI workloads, with industry leaders expressing confidence in its capabilities to enhance scalability and performance in AI networking [11][12][13] - Collaborations with companies like Accton, Arista Networks, and DriveNets highlight the router's potential to support next-generation AI infrastructure and improve energy efficiency [11][12][13][14]

Key Takeaways NVDA, AVGO and NOW are highlighted for their strong net income ratios and profitability metrics. Each firm showed higher-than-industry sales and income growth over the trailing 12-month period.All three stocks met strict screening criteria, narrowing 7,685 stocks down to a final group of 11.Investors should look for companies that generate strong returns after accounting for all operating and non-operating expenses. Therefore, it’s wise to invest in a profitable company rather than one that is ...

Investment Focus - The report highlights Amazon (AMZN US) as a top pick due to its leading position in the cloud industry with a 30% global market share, stable margin improvements, and strong demand for its T3 inference capabilities [1] - Alphabet (GOOGL US) is favored for its AI and advertising synergy, expected margin improvements in IaaS cloud services, and strong self-developed capabilities, although its stock price upside is currently limited [1] - Arista (ANET US) is recognized for its leadership in high-speed data center switches and expected revenue contributions from AI backend switch business, with a significant growth visibility [1] - Meituan (3690 HK) is noted for its strong cash flow generation ability and competitive cost structure, positioning it well in the face of industry competition [1] - Lenovo (992 HK) is highlighted for its record revenue in AI server business and significant growth potential compared to peers like Dell [2] - NVIDIA (NVDA US) is recognized for its strong financial performance and technological leadership in data center business, with a focus on emerging applications driving growth [2] - Tencent (700 HK) is expected to benefit significantly from AI advancements, with an upward revision in revenue and profit expectations for 2025 [2] - New Oriental (EDU US) is noted for its diverse revenue sources and strong brand recognition, supporting its high profit margins [3] - AIA (1299 HK) is favored for its steady growth in new business value and strong operational metrics, particularly in the ASEAN market [3] - Futu (FUTU US) is expected to see significant growth in paid user numbers and total AUM, supported by its low commission model and quality customer service [3] - The report emphasizes the potential of Chinese pharmaceutical companies like China Biologic Products (1177 HK) and Innovent Biologics (1801 HK) in their innovative drug pipelines and market leadership [4]