Group 1 - Nvidia's CEO Jensen Huang stated that the recent $30 billion investment in OpenAI may be the last, indicating a significant shift in their previously announced $100 billion infrastructure partnership due to OpenAI's preparations for an IPO [1][2] - The ambitious $100 billion investment plan is now deemed "highly unlikely" to be realized, as OpenAI's upcoming IPO will fundamentally change its capital structure and financing strategy, making it difficult for Nvidia to continue large-scale investments as a private equity investor [2] - The confirmed $30 billion investment is part of OpenAI's total $110 billion financing plan, which also includes $50 billion from Amazon and $30 billion from SoftBank, providing OpenAI with dedicated inference and training capabilities to support its growing AI data center needs [2] Group 2 - Nvidia's investment in another AI giant, Anthropic, amounting to $10 billion, is also likely to be the last, as Anthropic plans to go public in 2026, although the IPO decision has not been officially confirmed [3] - Huang expressed that this may be Nvidia's last opportunity to invest in such significant companies, as leading AI startups move towards the public market, redefining Nvidia's dual role as a chip supplier and strategic investor [3] - The AI industry is experiencing a profound shift in demand from model training to inference, which requires new chip efficiency and latency standards, prompting Nvidia to develop new chips optimized for inference tasks [4]

本文是由投资/产品专员而非分析师撰写的文章汇编。 它不构成研究报告，也不应被解释为研究报告，也不旨在提供 专业、投资或任何其他类型的建议或推荐。 CSIWM 个股点评 2026 年 3 月 4 日 英伟达 NVDA US 本文内容由 Ryan Lee (李昊谦) 提供 摘要 中信证券财富管理与中信里昂研究观点一致。根据中信里昂研究在 2026 年 3 月 4 日发布的题为《Why is the stock dull?》的报告，分析指出，即便在近期地缘政治事件发生之前，英伟达（Nvidia）股价已与盈利表现出现脱钩。研究 表示市场仍对超大规模资本支出保持警惕，担忧风投机构可用资金减少，同时关于 AI 建设周期性的争论持续存在。 尽管承认这些担忧，分析仍锚定指数级增长的代币处理需求及强劲盈利表现。 超大规模资本支出可信吗？分析认为确实可信 亚马逊、谷歌、微软和 Meta 四大资本支出方的经营现金流预计将在 2028 年达到 1 万亿美元。这些企业间的竞争持 续白热化，亚马逊近期资本支出激增，表明其希望挽回在 AI 云市场份额的损失。分析承认私募/风投领域可用资金较 2023 年峰值有所下降，但也注意到未量化的顺风 ...

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 ...

AI播客：换个方式听新闻 下载mp3 音频由扣子空间生成 对科技股投资者而言，今年开局颇为艰难。八家市值超过1万亿美元的美国科技公司中，有七家的股价 年初至今录得下跌。唯一的例外是英伟达(NVDA.O)。截至周二收盘，这家芯片制造商在2026年股价上 涨了3.4%。 周三美股盘后，英伟达将公布最新的季度财报。根据FactSet的共识预期，英伟达预计将在第四财季实现 经调整后每股收益1.54美元、营收661亿美元。 其中，1月季度的数据中心业务收入预计为607亿美元。就整个财年来看，FactSet跟踪的分析师预计公司 营收将达到2138亿美元。展望未来，分析师目前预计英伟达第一财季营收为729亿美元。 华尔街普遍看好未来的AI支出 目前看来，华尔街对这家公司的处境已有相当清晰的判断。这是因为其最大的一批客户几周前已陆续公 布业绩，并向投资者明确表示，在AI基础设施上的巨额投入只会进一步增加。 韦德布什证券的分析师在周一发布的英伟达财报前瞻报告中写道： "我们对2026年自然年超大规模云厂商资本开支的预测已超过此前预期……由于服务器和AI 基础设施构成未来支出的主体，我们预计AI投资的增长幅度将略高于整体资本开支 ...

Summary of Key Points from Conference Call Industry Overview - The conference call discusses the technology sector, particularly focusing on companies like Meta, AMD, Intel, TSMC, and Apple, highlighting a positive market reaction to AI developments [1][2]. Core Insights and Arguments 1. **Market Catalysts**: - On February 24, a significant agreement between Meta and AMD boosted confidence in AI, leading to an overall increase in tech stocks. Notable stock price changes included AMD up by 8.77%, Intel by 5.71%, TSMC by 4.28%, AAOI by 4.28%, Qualcomm by 3.11%, and Apple by 2.24% [1]. 2. **Production Shifts**: - Apple plans to relocate part of its Mac Mini desktop production from Asia to Houston, USA. The production will continue in Asia to meet local demand, while the US assembly line will cater to domestic needs [1]. 3. **Computing Power Expansion**: - Meta and AMD have signed a multi-year agreement, with expectations to deploy up to 6GW of graphics processing units (GPUs). The first batch of 1GW equipped with AMD's MI450 GPUs is scheduled for delivery in the second half of 2026 [1]. 4. **Chip Procurement**: - Apple announced plans to procure over 100 million advanced chips produced at TSMC's Arizona facility by 2026, indicating a significant investment in domestic semiconductor manufacturing [2]. Other Important Content - The Shanghai Stock Exchange approved the IPO application of Shenghe Jingwei on February 24, which may indicate growing interest and investment in the semiconductor sector [2].

Core Insights - Google is exploring ways to expand its AI chip market to better compete with market leader Nvidia, leveraging its financial strength to build a broader AI ecosystem [2][8] - The company's chips are gaining wider adoption for AI workloads, including clients like the startup Anthropic, but Google faces challenges such as manufacturing partner capacity constraints and limited interest from cloud computing competitors [2][3] - To expand its potential market, Google is increasing financial support for its data center partner network to provide computing power to a broader customer base [2][3] Investment and Partnerships - Google is reportedly negotiating to invest approximately $100 million in cloud computing startup Fluidstack, which has a valuation of about $7.5 billion [2][3] - Google has also provided financial guarantees for projects related to Hut 8, Cipher Mining, and TeraWulf, which are transitioning from cryptocurrency mining to data center development [3][9] - Discussions are ongoing about potentially restructuring the TPU team into an independent department to explore investment opportunities, although this poses challenges due to Google's reliance on Nvidia chips [3][10] TPU Development and Market Position - Google has been selling TPU computing power through its cloud services since 2018 and is also selling TPU chips directly to external customers [10] - The TPU team has gained importance, evidenced by the promotion of Amin Vahdat to Chief Technology Officer of AI Infrastructure, reporting directly to CEO Sundar Pichai [5][10] - The seventh generation TPU, named Ironwood, was launched in April last year, specifically designed for AI inference tasks [5][10] Supply Chain Challenges - Google may face obstacles in increasing TPU shipments due to tight advanced capacity at TSMC, which may prioritize Nvidia as its largest customer [11] - The company is also affected by a global shortage of storage chips, which are critical components of AI chips [11] - Interest in Google's TPU has grown among AI developers seeking cost-effective computing power to reduce dependence on Nvidia [11]

Core Viewpoint - Arm, a company deeply embedded in the semiconductor industry, faces challenges despite its widespread design usage in smartphones and connected devices. The company is optimistic about future growth driven by artificial intelligence, even as its stock price has declined due to weak demand in consumer electronics [2][3]. Group 1: Company Overview - Arm does not sell chips but licenses its designs, allowing clients to modify and produce chips themselves, generating revenue through upfront licensing fees and royalties [2]. - The company has shipped over 300 billion chips, with more than 30 billion shipped in the last year alone [2]. Group 2: Market Dynamics - Despite the growth in the Philadelphia Semiconductor Index by 65% since early 2025, Arm's stock has decreased by 2% due to sluggish demand in smartphones and consumer electronics [2]. - The rise of artificial intelligence is expected to stimulate demand for Arm's chips, particularly in data centers, which are increasingly reliant on both GPUs and CPUs [4]. Group 3: Strategic Choices - Arm may need to evolve from merely selling design blueprints to developing its own chips to capture more value from the AI boom, which could alienate existing clients [3][6]. - The company is exploring new revenue streams through pre-assembled processor modules, which could triple income per chip compared to traditional designs [5]. Group 4: Financial Performance - Analysts project Arm's revenue for the current fiscal year to be around $5 billion, with half coming from patent royalties and the other half from licensing fees, marking a 20% growth from 2025 [4][5]. - The patent royalty per mobile chip was approximately $0.86, representing 2.5% to 5% of the chip's price [5]. Group 5: Competitive Landscape - Arm's revenue is significantly lower compared to major chip manufacturers like Nvidia and Intel, highlighting the need for strategic shifts to enhance profitability [5]. - The company faces competition from China's push for RISC-V architecture, which is seen as a domestic alternative to Arm and Intel designs [7]. Group 6: Future Outlook - Arm's CEO expresses concern over the company's ability to keep pace with the rapid evolution of AI technologies, as chip design and manufacturing can take years, while AI models can evolve in months [7]. - The company's ownership structure, with SoftBank holding over 85% of shares, may influence its strategic direction, particularly as SoftBank seeks to build a competitive chip portfolio [6].

智通财经APP获悉，美东时间周三，网络设备巨头思科(CSCO.US)公布了截至1月24日的第二财季财报。尽 管季度业绩与全年营收指引均超出市场预期，并展现出人工智能(AI)业务带来的增长动力，但由于当前季 度的盈利预测平平，且利润率指引显著低于分析师预估，公司股价在盘后交易中大幅下挫。 财报显示，思科第二财季营收同比增长10%至153亿美元，市场预期为151亿美元，净利润从上年同期的 24.3亿美元增至31.8亿美元，调整后每股收益为1.04美元，优于市场预期的1.02美元。 在AI基础设施需求的强力拉动下，思科核心网络业务增长已有所提速。第二财季，来自超大规模数据中心 提供商的AI基础设施订单达到21亿美元，较前一季度的13亿美元显著增长。公司核心网络业务营收同比增 长21%至83亿美元，超越分析师预测的79亿美元。 虽然行业竞争在不断加剧——博通(AVGO.US)及收购了瞻博网络的慧与科技(HPE.US)等传统网络设备商， 也试图从开发及运行AI模型所需的硬件投资热潮中分得一杯羹。 但首席执行官Chuck Robbins表示，公司正处于"为AI时代提供可信基础设施"的独特位置，并预计整个2026 财年来自 ...

近日，瑞银发布博通(AVGO.US)专项研报，维持其"买入"评级及475美元目标价不变。报告围绕TPU(张量处理单元)需求激增展开，指出LLM开发者加 速推进定制ASIC路线，TPU作为GPU的中间替代方案需求显著增长，成为博通业绩增长的核心驱动力。 除谷歌外，Anthropic、META等TPU核心客户与传统云服务客户存在显著差异：这类企业可完全掌控自身的软件栈，因此对英伟达统一计算架 构(CUDA)的依赖度远低于传统企业级云服务客户。 经过一系列供应链调研分析，该行优化了自下而上的定制化专用集成电路模型，目前预测博通公司2027年将出货超500万颗张量处理单元 (2026年出货量约为370万颗);在2028年v8ax(太阳鱼)型号成为出货主力前，2027年出货的产品中略超半数为v7(铁木)型号。上述两款产品均 基于台积电3纳米工艺打造，该行认为凭借台积电充足的晶圆供应配额，博通能够充分把握这一需求增长机遇。 该行当前预测，博通2026财年人工智能业务营收约为600亿美元(同比增长约200%)，2027财年将增至约1060亿美元(同比增长约80%)， 2028财年进一步升至约1500亿美元。定制化计算业务营 ...

智通财经APP获悉，近日，瑞银发布博通(AVGO.US)专项研报，维持其"买入"评级及475美元目标价不 变。报告围绕TPU(张量处理单元)需求激增展开，指出LLM开发者加速推进定制ASIC路线，TPU作为 GPU的中间替代方案需求显著增长，成为博通业绩增长的核心驱动力。 博通作为成立于1961年的半导体巨头，总部位于美国圣何塞，产品覆盖数据中心网络、通信、智能手机 等多个领域。此次TPU业务的爆发式增长，不仅巩固了其在半导体行业的地位，也为公司开启了AI时代 的增长新周期，成为全球科技行业硬件创新的重要风向标。 瑞银在报告中写道，随着大语言模型开发商加快推进定制化专用集成电路(ASIC)的研发规划，众多厂商 开始将张量处理单元作为图形处理器(GPU)的过渡替代方案，该行认为该产品的需求正出现显著增长。 估值方面，研报采用SOTP分部门估值法，给予2027财年基础设施软件业务25xEV/FCF、半导体业务 30xEV/FCF的估值倍数。上行情景下目标价可达560美元(潜在涨幅63%)，下行情景为290美元(潜在跌幅 16%)，当前343.94美元的股价存在38%的上涨空间，预测总潜在回报39%，超额回报30. ...