公众号记得加星标⭐️，第一时间看推送不会错过。 过去一年里，凡是在AWS办公室电信部门附近走动的人，或许都曾听到过碎纸机低沉的嗡嗡声，那 是它正在处理AWS云RAN项目文件的情景。在去年的MWC世界移动通信大会上，AWS展示了一款搭 载Graviton3的Outposts服务器，Graviton3是AWS最新的基于Arm架构的中央处理器（CPU）。这款 服务器原本是为电信网络中的无线接入网（RAN）功能而设计的，而这些功能通常由爱立信和诺基 亚等公司开发。但如今，AWS却突然改变了策略，放弃了整个项目。 这可能会令那些一直寻求该领域更多选择的电信运营商感到不满，而且听起来像是对云无线接入网 （Cloud RAN）概念的又一次打击，云无线接入网的发展多年来一直断断续续。如今，几乎所有电 信运营商都依赖于专为无线接入网设计的硬件，这些硬件通常托管在他们自己的设施中。云无线接入 网或"虚拟"无线接入网推动了向通用计算机行业提供的通用CPU的转变。英特尔凭借其x86架构在该 领域占据主导地位，但包括AWS在内的其他公司则基于Arm（x86的主要替代方案）的蓝图设计芯 片。 以AWS为例，其设想是运营商可以使用搭载Gr ...

导语：在AI的新时代，计算能力就等同于收入，我们已经进入了转折点——自主人工智能正在颠覆全 球的商业模式，而销售人工智能工具也开始带来实实在在的利润。 黄仁勋称 "计算模式已然改变"，智能体人工智能（agentic AI）成为公司利润激增 94% 的核心 驱动力。 英伟达（ NVDA ）股价上涨 1.41% ，该公司公布第四季度利润同比增长 94% 、营收创下历史 纪录，这一业绩缓解了近几个月来席卷市场的人工智能泡沫担忧。 这家芯片制造商披露，当季净利润达 430 亿美元，较去年同期的 221 亿美元大幅增长；营收为 681 亿美元，较去年同期的 393 亿美元增长 73% ，轻松超出市场普遍预期。 纽曼称： "英伟达的业务与部分资产负债状况脆弱的公司深度绑定，因此面临着一定的风险。" 英伟达的发展前景或将受一个关键因素影响，即人工智能产业正从模型训练阶段向推理阶段转 型。推理是人工智能工具对用户的查询做出回应的过程，训练和推理需要不同类型的算力支撑， 相应的硬件设备也截然不同。 据 FactSet 调查的分析师此前预测，该季度英伟达净利润为 375 亿美元，营收为 661 亿美 元。 英伟达面向人工智能和云 ...

事件： 来源：招商电子 英伟达发布FY26Q4季报，本季营收681亿美元，同比+73%/环比+20%，创历史新高；non-GAAP毛利率 为75.2%，同比+1.7pcts/环比+1.6pcts。综合财报及交流会议信息，总结要点如下： 评论： 1、FY26Q4营收高增再超预期，公司开启战略备货以应对未来市场需求。 FY26Q4营收681亿美元，同比+73%/环比+20%，超出此前预期（650亿美元），本季营收、营业利润和 自由现金流均创历史新高；non-GAAP毛利率为75.2%，符合此前指引（74.5%-75.5%），同比+1.7pcts/ 环比+1.6pcts，环比增长主要得益于Blackwell架构产能持续爬坡；本季库存环比增长8%，采购承诺也大 幅增加，公司已战略性储备库存并锁定产能，以满足未来数个季度的市场需求。 2、数据中心营收再创新高网络业务表现亮眼，游戏受供应链影响环比下降。 1）数据中心：营收623.14亿美元，同比+75%/环比+22%，主要得益于Blackwell架构的持续强劲需求以 及Blackwell Ultra的产能爬坡。已部署并投入使用的Blackwell架构基础设施算力达近 9 ...

点击招商研究小程序查看PDF报告原文 事件： 英伟达发布FY26Q4季报，本季营收681亿美元，同比+73%/环比+20%，创历史新高；non-GAAP毛利率 为75.2%，同比+1.7pcts/环比+1.6pcts。综合财报及交流会议信息，总结要点如下： 评论： 1、FY26Q4营收高增再超预期，公司开启战略备货以应对未来市场需求。 FY26Q4营收681亿美元，同比+73%/环比+20%，超出此前预期（650亿美元），本季营收、营业利润和 自由现金流均创历史新高；non-GAAP毛利率为75.2%，符合此前指引（74.5%-75.5%），同比+1.7pcts/ 环比+1.6pcts，环比增长主要得益于Blackwell架构产能持续爬坡；本季库存环比增长8%，采购承诺也大 幅增加，公司已战略性储备库存并锁定产能，以满足未来数个季度的市场需求。 2、数据中心营收再创新高网络业务表现亮眼，游戏受供应链影响环比下降。 1）数据中心： 营收623.14亿美元，同比+75%/环比+22%，主要得益于Blackwell架构的持续强劲需求以 及Blackwell Ultra的产能爬坡。已部署并投入使用的Blackwell ...

证券研究报告|行业简评报告 2026 年 02 月 26 日 本季营收与指引均高增，战略备货以满足未来市场需求 英伟达（NVDA.O）FY26Q4 跟踪报告 TMT 及中小盘/电子 事件： 英伟达发布 FY26Q4 季报，本季营收 681 亿美元，同比+73%/环比+20%，创 历史新高；non-GAAP 毛利率为 75.2%，同比+1.7pcts/环比+1.6pcts。综合财 报及交流会议信息，总结要点如下： 评论： 1、FY26Q4 营收高增再超预期，公司开启战略备货以应对未来市场需求。 FY26Q4 营收 681 亿美元，同比+73%/环比+20%，超出此前预期（650 亿美 元），本季营收、营业利润和自由现金流均创历史新高；non-GAAP 毛利率为 75.2%，符合此前指引（74.5%-75.5%），同比+1.7pcts/环比+1.6pcts，环比 增长主要得益于 Blackwell 架构产能持续爬坡；本季库存环比增长 8%，采购承 诺也大幅增加，公司已战略性储备库存并锁定产能，以满足未来数个季度的市 场需求。 2、数据中心营收再创新高网络业务表现亮眼，游戏受供应链影响环比下降。 1）数据中心：营收 ...

花旗 银行分析师在研究报告中表示，智能体 人工智能不太可能对旅游预订平台产生实质性影响。阿里 巴巴已将其通义千问应用程序连接到包括飞猪在内的核心生态系统，使用户能够直接在通义千问应用中 预订机票和酒店。然而分析师指出，将单一旅游预订平台与人工智能助手连接并不具备足够的吸引力， 因为用户更倾向于跨平台进行价格比较。他们表示："影响旅游预订决策的因素多种多样，其复杂程度 超出了几个简单提示词所能涵盖的范畴。"分析师指出，短期内能够连接多个平台的AI助手将更具吸引 力。他们还提到，如果AI助手能够直接连接并服务于旅游产品供应商，旅游预订平台将面临更大的冲 击。 ...

Core Viewpoint - Analysts maintain a "sell" rating on Nvidia (NVDA) due to concerns over slowing growth and high valuation, despite a recent recovery in growth as indicated by the latest quarterly earnings report [1][2]. Group 1: Growth and Valuation Concerns - Nvidia's stock price has remained stagnant since August, indicating a potential peak [1]. - The analyst's bearish outlook extends beyond just "slowing growth + high valuation" leading to valuation compression [2]. Group 2: AI Industry Bubble Concerns - The AI sector's capital expenditures are unlikely to generate profits, suggesting it is in a bubble that could burst if spending growth slows [5]. - Michael Burry's short position on Nvidia has drawn attention to the potential risks in the AI industry [5]. - Burry argues that large cloud service providers are artificially inflating profit levels by extending the depreciation period of AI chips [5]. Group 3: Product Lifecycle and Market Dynamics - Nvidia's new chips follow an annual iteration cycle, similar to Apple's iPhone strategy, which is crucial for meeting revenue growth expectations [5]. - The depreciation period set by cloud service providers for AI chips (5-6 years) exceeds their actual lifespan, leading to potential write-downs when new chips are released [5][6]. - The recent launch of Nvidia's third-generation AI platform, Rubin, has rendered the previous Blackwell platform nearly obsolete [7][8]. Group 4: Financial Implications and Credit Risks - The introduction of the Rubin platform, which significantly reduces inference token costs and improves efficiency, raises concerns about the financial health of cloud service providers [9][11]. - Many cloud service providers are now relying on debt issuance to fund AI capital expenditures, indicating a shift towards a credit crisis in the AI sector [11][12]. - Companies with weak balance sheets, such as CoreWeave and Oracle, are experiencing rising credit default swap spreads, indicating increased default risk [12]. Group 5: Future Outlook and Market Trends - The upcoming CES in 2027 raises questions about Nvidia's next product and whether it will be a disruptive innovation [12]. - If Nvidia cannot maintain its annual iteration pace, the implied growth expectations in its valuation may not be met, leading to a potential stock price drop [13]. - The AI industry's capital expenditures are substantial, and Nvidia holds a near-monopoly in the AI accelerator market, complicating the competitive landscape [13][14]. - The potential for significant growth in the application of agentic AI by 2026 is a key trend to monitor, as is the ability of AI application companies to achieve profitability [14].

Core Insights - AMD showcased its upcoming Venice series server CPUs and MI400 series data center accelerators at the 2026 CES, marking the first public display of these product lines [1] - The Venice processor features significant changes in its packaging method, utilizing a more advanced approach compared to previous EPYC CPUs, and includes two I/O chips instead of one [1] - The MI400 accelerator is designed with a large package size, incorporating 12 HBM4 memory chips and multiple computing and I/O chips, indicating a substantial upgrade in performance capabilities [4] Venice Processor Details - Each Venice chip contains 8 CCDs, with each CCD housing 32 cores, allowing for a maximum of 256 cores per package [2] - The area of each CCD is approximately 165 square millimeters, with a potential L3 cache of 128MB per CCD, leading to a total of 3GB if V-Cache is implemented [6][2] - The I/O chips have a total area exceeding 700 square millimeters, significantly larger than previous EPYC I/O chips, indicating enhanced capabilities [2] MI400 Accelerator Insights - The MI400 features a large package size with two base chips and additional I/O chips, enhancing its functionality for high-performance computing [4] - The estimated area for the base chips is around 747 square millimeters, with I/O chips adding another 220 square millimeters [4] - The MI400 series will include a new product, MI440X, designed for 8-way UBB chassis, alongside MI430X and MI455X [5] AI and Future Developments - AMD is positioning itself in the AI computing space, aiming to compete with Nvidia by enhancing its Instinct GPU offerings and forming partnerships with companies like OpenAI [7][8] - The upcoming Helios server platform is designed for Yotta-scale computing, featuring the latest AI GPU Instinct MI455X and Venice CPU, set to launch later this year [10][14] - AMD's roadmap indicates a significant increase in GPU performance, with the MI455X expected to deliver ten times the inference throughput compared to its predecessor [16] Helios Server Specifications - The Helios platform will support over 18,000 CDNA5 GPU compute units and 4,600 Zen 6 CPU cores, achieving up to 2.9 exaflops of performance [18] - Each Helios rack will include 31 TB of HBM4 memory and high bandwidth capabilities, indicating a robust infrastructure for AI workloads [18] - AMD's advancements in memory and GPU bandwidth are expected to enhance data transfer speeds significantly, showcasing the company's commitment to high-performance computing [16][18]

Core Insights - The article discusses the current state of artificial intelligence (AI) development and trends for the upcoming year, highlighting a paradox of high adoption rates but low conversion rates in enterprises [1][2]. Group 1: Adoption and Impact - 88% of organizations report regular use of AI in at least one business function, up from 78% the previous year [4] - Only 39% of organizations attribute revenue growth to AI [4] - Over 90% of enterprises are attempting to adopt AI solutions to maintain competitiveness, yet 95% of organizations see no measurable return from generative AI investments [5][6]. Group 2: Challenges in Implementation - Major barriers to scaling AI include organizational and cultural inertia rather than technical issues, with many organizations failing to redesign workflows [9] - Approximately 80% of organizations have conducted pilot projects, but only about 5% of custom enterprise AI tools make it to production [9][10]. - Psychological safety significantly impacts the success of AI projects, with 83% of leaders acknowledging its importance [10]. Group 3: Successful Companies - Only about 5% of companies successfully implement AI, often treating it as a strategic priority managed by top leadership [16]. - Successful organizations undergo operational transformation and redesign workflows to maximize AI's inherent capabilities [17]. - They focus on establishing a culture that reduces resistance to new technologies and emphasizes productivity enhancement rather than immediate layoffs [19]. Group 4: Industry Adoption Rates - The highest adoption rates for AI are found in technology, sales and marketing, media and entertainment, and e-commerce sectors [24]. - Other notable sectors include professional services, finance, healthcare, and telecommunications [25]. Group 5: Future Trends - Global AI spending is expected to approach $1.5 trillion by 2025, driven by applications in consumer products and data center infrastructure [52]. - By 2030, AI is projected to contribute up to $15.7 trillion to the global economy, with significant contributions from regions like India [53]. - The rise of autonomous AI agents capable of independent reasoning and execution of multi-step tasks is anticipated to transform operational risks and governance requirements [42].

英国金融监管机构表示，各大银行竞相采用具备决策与自主执行能力的智能体人工智能（agentic AI），这给零售客户带来了新的风险。该监管机构承诺，将确保零售客户的利益不会被忽视。 人工智能 "智能体" 有望彻底改变人们的预算规划、储蓄和投资方式。例如，它可自动将闲置资金转入 高收益账户，或根据市场波动调整投资组合。 英国国民西敏寺银行（NatWest）、劳埃德银行（Lloyds）以及星展银行（Starling）向路透社透露，它 们正与英国金融行为监管局（FCA）协作，筹备面向零售客户的试点项目。这与银行业此前仅将人工智 能用于后台办公的模式相比，是一次重大转变。 2026 年正式推向市场 英国金融行为监管局首席数据官杰西卡・鲁苏预计，面向消费者的智能体人工智能应用最早将于明年初 正式大规模投放市场。 鲁苏在接受路透社采访时表示："所有人都意识到，智能体人工智能带来了新的风险，这主要源于其高 速执行任务的能力。" 人工智能智能体的自主性及其与其他智能体交互的速度，放大了金融稳定和治理层面的相关风险。 鲁苏指出，英国金融行为监管局将实施高级管理人员制度及消费者权益保护准则，要求企业负责人对违 规行为承担责任，并确 ...