2025 年 8 月，分析师将英伟达公司（NVDA）评级定为 "卖出"，彼时指出英伟达增长增速放缓，而估 值仍处于高位，因此预计其市盈率倍数将收缩，股价面临下行压力。 不过，在最新披露的季度财报中，英伟达的增长态势实则有所回升，为其提供了一定支撑。总体而言， 自 8 月以来，英伟达股价基本维持横盘震荡，呈现出触顶态势。 基于此，分析师重申对英伟达的 "卖出" 评级。 此外，正如本文后续将进一步阐述的，分析师看空的逻辑已不止于 "增长放缓 + 高估值" 引发的估值收 缩这一点。 伯里强调的关键矛盾是：超大规模云服务商将人工智能芯片的折旧年限设定为 5-6 年，这一期限远超芯 片的实际使用寿命 —— 它们通过将高价芯片的成本分摊至 5-6 年，来维持财报利润的正向表现。 然而，一旦新一代人工智能芯片发布，旧款芯片基本会被淘汰。 之所以说 "基本"，是因为旧款芯片仍可用于诸多非核心功能，但无法承载先进的人工智能工作负载。 由此，超大规模云服务商或将被迫对旧芯片相关的人工智能资本开支进行资产减记，计入亏损 —— 这 正是人工智能泡沫破裂、英伟达股价崩盘的导火索。 分析师一贯的观点是，人工智能领域的资本开支难以产生盈利 ...

分析师一贯的观点是， 人工智能领域的资本开支难以产生盈利，因此人工智能行业属于泡沫领 域，一旦资本开支增速放缓，泡沫便会破裂。 而当迈克尔・伯里披露其对英伟达的空头头寸后，市场才开始更多地关注人工智能行业的潜在 风险。 伯里指出，超大规模云服务商通过延长人工智能芯片的折旧年限，人为抬高了利润水平。此 外，他认为英伟达人工智能生态中存在的循环投资与供应商融资模式，与互联网泡沫时期的情 况存在相似性。 2025 年 8 月，分析师将英伟达公司（NVDA）评级定为 "卖出"，彼时指出英伟达增长增速 放缓，而估值仍处于高位，因此预计其市盈率倍数将收缩，股价面临下行压力。 不过，在最新披露的季度财报中，英伟达的增长态势实则有所回升，为其提供了一定支撑。总 体而言，自 8 月以来，英伟达股价基本维持横盘震荡，呈现出触顶态势。 基于此，分析师重申对英伟达的 "卖出" 评级。 此外，正如本文后续将进一步阐述的，分析师看空的逻辑已不止于 "增长放缓 + 高估值" 引发 的估值收缩这一点。 【如需和我们交流可扫码添加进社群】 伯里提出的 核心问题在于，英伟达的新芯片遵循年度迭代节奏，即每年都会推出新一代人工智 能加速器及相关硬件 ...

英伟达高管一再释放乐观信号，看好Blackwell及下一代Rubin芯片的销售前景。 美东时间6日周二，英伟达首席财务官Colette Kress在摩根大通主办的活动上称，由于需求强劲，英伟达如今更为看好数据 中心业务，到2026年末，英伟达数据中心芯片的预期收入"肯定"会超过去年10月给出的预测5000亿美元。 Kress所说的5000亿美元是指，黄仁勋在两个多月前的GTC大会上表示，到2026年底，英伟达现有和未来数据中心芯片将 带来约5000亿美元收入。高盛10月末指出，这一预期比当时的市场共识高出约12%。 Kress这一表态延续了英伟达CEO黄仁勋本周一在消费电子展（CES）上讲话的乐观情绪。英伟达周一宣布正式发布Rubin 芯片。黄仁勋周一在CES的主题演讲中透露，新一代Vera Rubin平台已全面投产，推理成本降至Blackwell平台的十分之 一。 黄仁勋周二还在媒体采访中强调，英伟达新款芯片性能比上一代高出10倍，并表示中国客户需求强劲。 尽管Kress和黄仁勋都释放积极信号，英伟达股价仍未扭转跌势。英伟达周二早盘一度涨超2%后，持续回落，午盘小幅转 跌，收跌近0.5%，连续两日收跌。20 ...

英伟达CFO Kress称，自2025年10月给出5000亿美元的现有和未来数据中心芯片收入预期以来，英伟达客户的兴趣持续增加；公司对AI应用的乐观预 期不仅来自AI需求本身，企业数据处理需求也在推动下一代计算需求增长。 美东时间6日周二，英伟达首席财务官Colette Kress在摩根大通主办的活动上称，由于需求强劲，英伟达如今更为看好数据中心业务，到2026年末，英 伟达数据中心芯片的预期收入"肯定"会超过去年10月给出的预测5000亿美元。 Kress所说的5000亿美元是指，黄仁勋在两个多月前的GTC大会上表示，到2026年底，英伟达现有和未来数据中心芯片将带来约5000亿美元收入。高 盛10月末指出，这一预期比当时的市场共识高出约12%。 Kress这一表态延续了英伟达CEO黄仁勋本周一在消费电子展（CES）上讲话的乐观情绪。英伟达周一宣布正式发布Rubin芯片。黄仁勋周一在CES的 主题演讲中透露，新一代Vera Rubin平台已全面投产，推理成本降至Blackwell平台的十分之一。 黄仁勋周二还在媒体采访中强调，英伟达新款芯片性能比上一代高出10倍，并表示中国客户需求强劲。 黄仁勋周一在拉斯维 ...

英伟达正试图挑战特斯拉和其他少数几家公司在自动驾驶汽车业务领域的主导地位。就目前而言，特斯 拉首席执行官马斯克表示他并不担心。 马斯克周一晚间在其社交媒体平台X上发帖写道："我并不会为此而失眠。而且我真心希望他们成功。" 英伟达周一发布了一系列名为Alpamayo的开源自动驾驶模型，并称这些模型将实现"类人"决策。该公 司表示，其模型将能够利用推理来应对罕见或棘手的场景，而不是采用传统的基于规则的方法，后者可 能更难覆盖车辆面临的各种不可预测的情况。 这个新模型系列可能会提升英伟达在不断增长的自动驾驶汽车行业中的影响力，该公司将该行业视为一 个巨大的机遇。据首席执行官黄仁勋称，英伟达八年前就开始研发自动驾驶汽车技术，该部门有数千名 员工。 "在我看来，毫无疑问，这将成为最大的机器人产业之一，"黄仁勋周一在消费电子展（CES）上 说。"我们的愿景是，有朝一日，每一辆轿车，每一辆卡车，都将实现自动驾驶。" 根据公司高管的声明，优步、捷豹路虎和Lucid Group都有兴趣使用Alpamayo来开发完全无人驾驶的汽 车。这三家公司都已在自动驾驶技术上与英伟达合作，比亚迪也是如此，该公司在2025年的销量已超过 了特 ...

Rubin平台的训练性能是Blackwell的3.5倍，运行AI软件的性能则提升5倍，训练混合专家模型所需GPU数量减少4倍。黄 仁勋称，全部六款Rubin芯片已通过显示其可按计划部署的关键测试。英伟达称该平台已全面投产，亚马逊AWS、谷歌 云、微软和甲骨文云等云服务商将率先部署。 硬·AI 作者 | 李 丹 编辑 | 硬 AI 英伟达在CES展会推出新一代Rubin AI平台，标志着其在人工智能（AI）芯片领域保持年度更新节奏。该 平台通过六款新芯片的集成设计，在推理成本和训练效率上实现大幅跃升，将于2026年下半年交付首批客 户。 美东时间5日周一，英伟达CEO黄仁勋在拉斯维加斯表示，六款Rubin芯片已从合作制造方处回厂，并已 通过部分关键测试，正按计划推进。他指出"AI竞赛已经开始，所有人都在努力达到下一个水平"。英伟达 强调，基于Rubin的系统运行成本将低于Blackwell版本，因为它们用更少组件即可实现相同结果。 微软和其他大型云计算提供商将成为下半年首批部署新硬件的客户。微软的下一代Fairwater AI超级工厂 将配备英伟达Vera Rubin NVL72机架级系统，规模可扩展至数十万颗 ...

AI正在重塑全球经济的底层逻辑——标普500指数75%的涨幅、80%的利润、90%的资本开支都直接源于AI产业。 在这场颠覆性浪潮中，英伟达（NVDA）早已超越"AI芯片龙头"的单一标签，从AI时代的"卖铲人"进化为定义算力规则、垄断产业命脉的"总设计师"。 从1999年IPO时仅6.26亿美元的市值，到如今一度突破5万亿美元，26年间8000倍的涨幅、约40%的年化回报率，英伟达用实打实的成长证明：每一次计算 时代的更迭，都会诞生一位无可撼动的王者。 PC时代属于英特尔，移动互联网时代属于高通和苹果，而AI时代，注定是英伟达的主场。 01 算力革命的核心：为什么非英伟达不可？ 英伟达的行业统治力，并非偶然的产品成功，而是技术、生态、需求三重共振的必然结果，构筑起对手难以企及的竞争壁垒，成为AI基建狂潮中无可替 代的核心。 1.需求端：指数级爆发的算力缺口，五大云厂商"军备竞赛"托底 AI的本质是算力的较量，而当前算力需求正以指数级速度爆发。 根据英伟达FY26Q3电话会披露，预训练、后训练、推理三大缩放定律依然有效，形成了"计算资源普及→更先进模型诞生→应用场景落地→催生更多算力 需求"的良性循环。 这一循 ...

Group 1 - The global wafer foundry industry is expected to see a revenue increase of 8.1% quarter-on-quarter, reaching approximately $45.1 billion in Q3 2025, driven by demand for AI high-performance computing (HPC) and new consumer electronics chips [3] - Despite the positive outlook for Q3 2025, there are concerns about potential geopolitical disruptions affecting demand in 2026, leading to a more conservative supply chain outlook for mainstream terminal demand [3] - TSMC's market share increased to 71% in Q3 2025, supported by strong demand from Apple for iPhone and NVIDIA's Blackwell platform, with both wafer shipments and average selling prices rising [3] Group 2 - UMC's capacity utilization slightly improved in Q3 2025 due to demand for mature process ICs from smartphones and PCs, resulting in a market share of 4.2% [4] - GlobalFoundries experienced a slight decrease in market share to 3.6% despite a small increase in shipments, attributed to a one-time adjustment in average selling prices [4] - The "China for China" trend benefited Hefei Jinghe, which saw increased demand for DDIC, CIS, and PMIC, allowing it to surpass Tower Semiconductor in rankings [4]

Core Viewpoint - Nvidia's FY26Q3 financial report indicates a robust demand for AI computing power, with a significant revenue forecast of $500 billion from the Blackwell and Rubin platforms by the end of 2026, amidst a backdrop of global GPU shortages and a backlog of 5 million orders [1][2][30]. Financial Performance - Revenue reached $57.01 billion, exceeding expectations by nearly $2 billion and showing a year-over-year increase of 62% [2][3]. - Adjusted net profit was $31.7 billion, translating to an average daily profit of $350 million, surpassing the annual revenue of many tech companies [2][4]. - Adjusted gross margin was 73.6%, with a target of 75% for FY27, reflecting improved profitability [5][7]. Business Segments - Data center revenue was $51.2 billion, accounting for 90% of total revenue, with a year-over-year growth of 66% [9]. - The computing business grew by 56%, supported by the GB300 product line [10]. - Networking revenue doubled to $8.2 billion, driven by NVLink and Spectrum-X [13]. - Gaming revenue was $4.3 billion, up 30% year-over-year, while professional visualization and automotive segments also showed strong growth [16]. Demand and Orders - Nvidia reported a staggering demand for GPUs, with a backlog of 5 million units and a projected revenue visibility of $500 billion from AI projects through 2026 [18][30]. - Major partnerships include a 1 GW computing power agreement with Anthropic and a three-year supply deal with Saudi Arabia for 400,000 to 600,000 GPUs [18]. Technological Advancements - The Blackwell platform, particularly the GB300, is expected to dominate the market, offering performance improvements over previous models [19]. - The upcoming Rubin platform, set to launch in late 2026, aims for significant performance enhancements [19][31]. Strategic Partnerships and Supply Chain - Nvidia's growth is supported by strategic partnerships across various sectors, including collaborations with OpenAI and traditional industries like Caterpillar and Toyota [22][25]. - The company is focused on building a resilient global supply chain, with key partnerships with TSMC and other suppliers to ensure production capacity [26][27]. Future Outlook - Nvidia's financial results challenge the notion of an AI bubble, demonstrating real demand and profitability in the AI sector [30]. - Key indicators for future growth include meeting revenue targets for FY26Q4, successful launch of the Rubin platform, and maintaining supply chain stability [30][31].

Core Viewpoint - Nvidia's FY26Q3 financial report indicates a robust and sustained demand for AI computing power, with significant revenue growth and a strong order backlog, dispelling concerns about an AI bubble [2][4][49]. Financial Performance - Revenue reached $57.01 billion, exceeding expectations by nearly $2 billion and showing a year-over-year increase of 62% [3][5]. - Adjusted net profit was $31.7 billion, translating to an average daily profit of $350 million, surpassing the annual revenue of many tech companies [3][4]. - The adjusted gross margin was 73.6%, with a target of 75% for FY27, indicating improved profitability [7][9]. Business Segments - Data center revenue was $51.2 billion, accounting for 90% of total revenue, with a year-over-year increase of 66% [11]. - The computing business grew by 56%, supported by the GB300 product line [12]. - Networking revenue doubled to $8.2 billion, driven by NVLink and Spectrum-X [15][38]. - Automotive revenue reached $592 million, a 32% increase year-over-year, indicating future potential in autonomous driving [17]. Demand Dynamics - Nvidia has a backlog of $500 billion in orders, with 5 million GPUs needed for AI projects this quarter [18][22]. - The demand spans across cloud providers, sovereign nations, and tech companies, with no signs of a seasonal slowdown [23][24]. - Key partnerships include a 1 GW computing power agreement with Anthropic and a three-year supply deal with Saudi Arabia for 400,000 to 600,000 GPUs [21]. Technological Leadership - The Blackwell platform is a key driver, with the GB300 product contributing nearly two-thirds of its revenue and outperforming previous models [31][32]. - The upcoming Rubin platform, set to launch in the second half of 2026, aims for significant performance improvements [34]. - The CUDA ecosystem remains a critical competitive advantage, ensuring customer retention and performance upgrades [36]. Supply Chain and Partnerships - Nvidia is building a resilient global supply chain, with key partners including TSMC and Foxconn [44]. - TSMC's CoWoS capacity is projected to reach 105,000 wafers per month by the end of 2026 [45]. - The company maintains a cooperative stance with competitors like Intel, balancing competition with collaboration [46]. Future Outlook - Key indicators for Nvidia's continued growth include meeting FY26Q4 revenue targets, successful Rubin platform production, and stable supply chain costs [51]. - Nvidia's core advantages include its technological monopoly, the CUDA ecosystem, and a substantial order backlog, positioning it as a critical player in the AI infrastructure market [52].