微软过去12个月新增超过2吉瓦数据中心容量，在全球70个地区运营超过400个数据中心，规模超过任何其他云服务提供 商。公司AI功能月活跃用户已超过8亿，客户采用Copilot的速度超过微软365套件中任何其他新产品，本季度实现了自推 出以来最大的席位增长。Azure云服务的高速增长很大程度上得益于企业迁移活动的持续活跃，新的AI工作流程和云原生 应用的持续扩展也推动了Azure增长。 硬·AI 作者 | 董 静 编辑 | 硬 AI 微软2025财年收官之际展现出强劲增长势头，AI与云业务持续激增，Azure云服务营收同比增长39%。AI 基础设施扩张加速，数据中心建设领跑竞争对手，Copilot产品线用户数突破1亿，云迁移业务成增长引 擎。 7月30日，微软公布2025财年四季度财报显示，营收达764亿美元，同比增长17%，Azure云服务营收同比 增长39%，全年营收首次突破750亿美元大关。微软云业务收入首次突破1680亿美元，同比增长23%。 据 见闻此前文章 ，微软2025财年四季度242亿美元的总资本支出创单季最高纪录，较前一季度增长28亿 美元、增幅接近13.1%，而且相比分析师预期的支出231. ...

Core Viewpoint - Meta's AI technology has significantly enhanced advertising economic benefits and improved user engagement and content quality, becoming a major growth engine for the overall business. Zuckerberg stated that Meta is now equipped to achieve "super intelligence," with substantial capital expenditure growth expected in 2026 [1][2][3]. Financial Performance - In Q2, Meta reported revenue of $47.52 billion, exceeding analyst expectations of $44.83 billion, with advertising revenue of $46.5 billion also surpassing forecasts. The Reality Labs division incurred a loss of $4.5 billion, which was better than market expectations. The company raised its 2025 capital expenditure lower limit from $64 billion to $66 billion, leading to a 10% increase in stock price post-announcement [2][3][4]. - The operating profit margin for the quarter was 43%, with a net income of $18.3 billion, translating to earnings per share of $7.14. Total expenses for the quarter were $27.1 billion, a 12% year-over-year increase [24][25]. AI Monetization and User Engagement - AI has become the core driver of Meta's current business growth, with a significant efficiency boost in the advertising system. The new AI-driven advertising recommendation model improved ad conversion rates by approximately 5% on Instagram and 3% on Facebook. The use of generative AI creative tools has also expanded, particularly among small advertisers with limited budgets [3][4][21]. - User engagement metrics improved, with Facebook's user time increasing by 5% and Instagram's by 6%. Video engagement saw a year-over-year increase of over 20% [4][28]. Investment in AI and Infrastructure - Meta plans to continue investing heavily in computing power and talent resources, establishing the "Meta Super Intelligence Lab" to develop next-generation models. The company is building multiple gigawatt-level computing clusters to provide "personal super intelligence" for billions of users [3][6][12]. - Capital expenditures for 2025 are projected to be between $66 billion and $72 billion, with expectations for significant growth in 2026 as well. The CFO emphasized that infrastructure costs will be the primary driver of expense growth in 2026, including depreciation and operational costs [5][6][38]. Talent Acquisition and Team Structure - The company is focusing on building a "small but elite" team of top talent in AI, with a particular emphasis on recruiting industry-leading experts. The structure of the team is designed to facilitate cutting-edge research in super intelligence [9][10][11]. - Employee compensation is expected to be the second-largest driver of expense growth in 2026, primarily due to investments in technical talent [11][36]. Future Outlook and Strategic Focus - Meta's future strategy includes enhancing the freshness of original content and improving the recommendation system to better match user interests. The company aims to leverage AI advancements to further improve user engagement and monetization efficiency [4][28][49]. - The company is optimistic about the long-term potential of AI technologies and their ability to reshape its systems and operations, with a focus on self-improvement capabilities in AI [42][48].

OpenAI今年前七个月实现收入翻番，年化收入达到120亿美元，相比2024年约40亿美元的收入水平实现大幅增长，有望 超越其2025年127亿美元的收入预期。ChatGPT周活跃用户从3月的5亿增加到7亿，快速增长的同时，OpenAI的现金消 耗也在加速，将2025年现金消耗预期上调至约80亿美元，较此前预期增加10亿美元。 硬·AI 编辑 | 硬 AI OpenAI收入增长超预期，用户规模持续扩张。 7月31日，据The Information报道，OpenAI年化收入已翻倍至120亿美元，同时其旗舰产品ChatGPT的 周活跃用户数突破7亿。 快速增长的同时，OpenAI的现金消耗也在加速。 该公司已将2025年现金消耗预期上调至约80亿美元，较 此前预期增加10亿美元。 其在服务器租赁方面的支出可能超过此前预计的140亿美元。 在融资方面，OpenAI正接近完成第二轮融资其中75亿美元融投资的承诺 ，红杉资本和Tiger Global Management等现有股东将各自投入数亿美元，其他参与方包括Dragoneer和Founders Fund等知名投资 机构。 01 前七个月实现收入翻番 周月活用户 ...

摩根士丹利认为，从CoWoS转向CoWoP在技术上仍面临重大挑战，对ABF基板的依赖短期内难以改变。技术转换的复杂 性和供应链重组风险使得短期内大规模采用CoWoP并不现实。不过，该行认为，不排除英伟达正在并行开发CoWoP技术 的可能性。 硬·AI 作者 | 董 静 编辑 | 硬 AI 尽管市场热炒芯片级晶圆板上封装（CoWoP）技术，但是英伟达下一代GPU产品Rubin Ultra采用该技术的 可能性较低。 7月30日，据追风交易台消息，摩根士丹利最新研究显示，英伟达的Rubin Ultra仍将沿用现有的ABF基板技 术，而非转向CoWoP方案。 大摩分析师认为，从CoWoS转向CoWoP在技术上仍面临重大挑战，对ABF基板的依赖短期内难以改变。研 报指出， 技术转换的复杂性和供应链重组风险使得短期内大规模采用CoWoP并不现实 。 不过，大摩认为， 尽管短期内不太可能大规模应用，不排除英伟达正在并行开发CoWoP技术的可能性 。 01 技术门槛过高： CoWoP面临制程难题 CoWoP技术要求PCB（印刷电路板）的线/间距(L/S)缩小至10/10微米以下，这与目前ABF基板的标准相 当。 大摩在研报中 ...

作者 | 赵雨荷 编辑 | 硬 AI 苹果人工智能团队正面临人才流失危机，该公司多模态AI研究员张博文于上周五离职，即将加入Meta公司新近成立的超 级智能团队。过去一个月内，已有四位核心AI研究员跳槽至Meta旗下新组建的"超级智能团队"。受关键成员离职影响，苹 果的AI基础模型团队陷入混乱，多名工程师正考虑另谋出路。随着苹果内部对是否继续开发自研模型或改用OpenAI、 Anthropic等第三方技术出现分歧，士气进一步受挫。外界担忧，这可能影响苹果在AI领域的长期竞争力。 硬·AI 相比之下，Meta正大举扩充AI团队，为硅谷AI人才开出巨额薪酬包，从苹果、OpenAI和Anthropic等公司 成功挖人。 媒体报道，苹果公司在AI研发工作又迎来一次重大打击，在一个月内已经失去了第四位人工智能研究员，继 续遭到Meta挖角。 知情人士对媒体透露，苹果公司关键多模态AI研究员张博文（Bowen Zhang）于上周五离职，即将加入 Meta公司新近成立的超级智能（superintelligence）团队。张博文原是苹果基础模型团队（AFM）的一 员，该团队负责构建苹果AI平台背后的核心技术。 据张博文领英页 ...

Core Viewpoint - The competition for TSMC's CoWoS capacity has intensified as major tech companies vie for a share in the AI semiconductor market, with NVIDIA projected to dominate the demand by 2026 [2][3][14]. Group 1: CoWoS Demand and Market Dynamics - Global demand for CoWoS is expected to reach 1 million wafers by 2026, representing a growth of 40% to 50% in the cloud AI semiconductor market [2][11]. - NVIDIA is forecasted to consume 595,000 wafers, accounting for approximately 60% of the total demand, solidifying its market leadership [3][5]. - Other key players like AMD, Broadcom, and Amazon are also aggressively securing capacity, with AMD expected to acquire 105,000 wafers (11% market share) and Broadcom 150,000 wafers (15% market share) [4][6][7]. Group 2: Capital Expenditure Trends - Major cloud service providers (CSPs) are increasing their capital expenditures, with Google raising its 2025 budget from $75 billion to $85 billion, indicating strong market growth signals [11][12]. - The monthly token processing volume on Google Cloud has surged from 480 trillion to 980 trillion, reflecting a doubling in demand [12]. Group 3: TSMC's Capacity Expansion - TSMC is ramping up its CoWoS capacity significantly, with projections indicating an increase from 32,000 wafers per month in 2024 to 93,000 wafers per month by the end of 2026 [14]. - AI-related revenue is expected to contribute 25% to TSMC's total revenue by 2025, positioning the company as a key beneficiary in the ongoing AI wave [14].

Core Viewpoint - The embodied intelligence industry is transitioning from demonstration to practical application, with the second half of the year being a critical period for delivering results [1] Group 1: Company Strategy and Market Position - Zhiyuan has secured a contract with China Mobile for 78 million, indicating strong market demand for its humanoid robots in service sectors [2] - The company aims to provide an integrated hardware and software experience, similar to Apple, rather than an open interface model like Android [10] - Zhiyuan's focus is on real-world data collection, emphasizing that synthetic data cannot fully capture the complexities of physical interactions [6] Group 2: Product Development and Supply Chain - The company anticipates several thousand units to be shipped this year, but faces challenges in the supply chain, particularly with core components like joints and reducers [4] - Zhiyuan is committed to a fully self-developed approach, integrating body, brain, and cognitive functions to create a closed-loop system for product development [5] - The company is exploring both open and real-world scenario data collection to enhance its data diversity and quality [7] Group 3: Market Trends and Future Outlook - The second half of the year is viewed as a window of opportunity for embodied intelligence, with expectations for significant market validation [2] - The company sees a vast potential market for embodied intelligence applications, predicting that specialized companies will emerge in various segments [2][10] - Zhiyuan is also considering entering the quadruped robot market to diversify its product offerings and better understand market needs [13] Group 4: Cost Management and ROI - The company believes that as industrial applications expand, manufacturing costs will decrease, making products more acceptable to clients [11] - Zhiyuan focuses on achieving a reasonable return on investment (ROI) rather than merely reducing costs [11] Group 5: Competitive Landscape - The entry of automotive companies into the embodied intelligence space is seen as a natural progression, but Zhiyuan remains focused on its core business [10] - The company acknowledges that while automotive firms have advantages in supply chain and management, the market for embodied intelligence is significantly larger than that for electric vehicles [10]

Core Viewpoint - The article discusses the launch of the new flagship model GLM-4.5 by Zhipu AI, which is designed for intelligent agent applications and has been released on HuggingFace and ModelScope platforms [2][3]. Group 1: Model Architecture and Performance - GLM-4.5 utilizes a mixture of experts (MoE) architecture with a total parameter count of 355 billion and 32 billion active parameters, while GLM-4.5-Air has 106 billion total parameters and 12 billion active parameters [4][6]. - The model integrates reasoning, coding, and intelligent agent capabilities, achieving a comprehensive performance ranking in the global top three, and is the leading domestic and open-source model [3][4]. - In comparative tests against models like Claude Code and Kimi-K2, GLM-4.5 demonstrated superior task completion and tool reliability, although it slightly lagged behind Claude-4-Sonnet in some dimensions [8]. Group 2: Cost and Efficiency - The API call pricing for GLM-4.5 is set at 0.8 yuan per million tokens for input and 2 yuan per million tokens for output, making it a cost-effective option [10]. - The high-speed version of the model supports a generation rate of up to 100 tokens per second, catering to high concurrency deployment needs [12]. Group 3: Training Data and Fine-tuning - The training data for GLM-4.5 encompasses 15 trillion tokens of general corpus, supplemented by 8 trillion tokens specifically fine-tuned for coding, reasoning, and agent tasks, enhanced through reinforcement learning [7]. Group 4: Agent Capabilities and Demonstrations - Zhipu AI has released multiple real-world scenario demos to showcase the agent capabilities of GLM-4.5, including a simulated search engine, a video platform simulator, a playable Flappy Bird game, and an automated PPT tool [14].

Core Insights - The 2025 WAIC indicates a shift towards practical commercialization in the robotics industry, with wheeled robots becoming mainstream due to their ease of deployment and cost-effectiveness [2][3][4] - Despite advancements in mobility, fine manipulation remains a significant challenge, hindering robots from fully replacing human labor [8][9] - The cost of robots is decreasing, but a clear turning point has not yet been reached, necessitating more definitive industry signals for sustainable stock performance [3][11] Group 1: Commercialization Trends - The WAIC showcased over 60 robotic products, a significant increase from 25 static prototypes last year, indicating a move towards real-world applications in various sectors [2][4] - The design trend is shifting towards wheeled bases (AGV-style) rather than complex bipedal designs, prioritizing immediate commercial viability over technical sophistication [4][5] - The event's scale and participation have grown, with a 35% increase in venue size and a 60% rise in exhibitors, reflecting heightened investment and government support in the robotics sector [4] Group 2: Application Scenarios - Robots are increasingly tailored for specific applications, moving away from a one-size-fits-all approach [6] - In industrial settings, robots are being developed for tasks such as power line inspections and quality control in harsh environments [6] - In consumer and service industries, robots are being designed for practical tasks like making ice cream, organizing rooms, and providing retail assistance [6][7] Group 3: Technical Challenges - Despite improvements in autonomous navigation and dynamic movement, the precision of robotic manipulation remains a critical limitation [9] - Demonstrations at WAIC revealed frequent operational failures, with task completion times significantly lagging behind human capabilities [9] - The integration of visual-language-action (VLA) models and reinforcement learning (RL) is seen as essential for enhancing robotic performance and achieving commercial success [9] Group 4: Cost and Data Considerations - The introduction of entry-level robots priced at 40,000 RMB marks a notable development, yet mainstream robots still range from 400,000 to 500,000 RMB [11] - High-quality real-world interaction data is crucial for training models, but the cost of data collection is high, leading companies to adopt mixed strategies for data sourcing [11]

三星电子与特斯拉签署价值165亿美元的芯片制造协议，合同期至2033年底。马斯克确认合作细节，称三星德州工厂将专 门制造特斯拉AI6芯片，并指出"这还只是最低金额"。分析指出，这份大单是特斯拉从汽车公司向AI和机器人公司转型的 重要一步，另一方面对陷入产能利用率不足的三星代工业务意义重大，有望使其代工销售额每年增长10%。 硬·AI 作者 | 董 静 编辑 | 硬 AI 另据媒体报道指出，这份持续至2033年底的大单，不仅标志着三星在激烈的代工竞争中取得重要突破，也 为三星提供了在AI芯片制造领域重新证明自己的重要机会。 消息传出后，三星电子在首尔交易的股价大幅上涨6.8%，报70400韩元/股， 为去年9月以来最高水平 。 三星电子与特斯拉达成165亿美元芯片制造协议，马斯克确认与三星合作细节，不仅强调"将亲自参与提升 生产效率"，还指出"这还只是最低金额"。 7月28日，据媒体报道，知情人士透露，三星周一宣布获得价值22.8万亿韩元的芯片制造合同，客户正是已 与其代工部门有业务往来的特斯拉。 特斯拉CEO马斯克在社交平台X上的多条帖文中详细披露了这一合作的战略布局，并强调"将亲自参与提升 生产效率"。 分 ...