AI是一个五层蛋糕 2026 年 3 月 10 日 作者 黄仁勋 3月11日消息，英伟达CEO黄仁勋发表了一篇以《AI是一个五层蛋糕》为题的，关于人工智能的长篇博 客文章，文章系统阐释了AI产业的底层逻辑，黄仁勋在文中定义了AI的"五层架构"。他表示，当前AI 产业仍处于极早期发展阶段，尽管行业已投入数千亿美元，但AI的真正潜力尚未被完全发掘，未来仍 需数万亿美元的持续投资来完善底层基础设施。 黄仁勋指出，AI已成为当今塑造世界的最强大力量之一，它并非单一的聪明应用程序或模型，而是如 同电力和互联网一样至关重要的基础设施，运行在真实的硬件、能源和经济基础之上，能够吸收原材料 并转化为规模化的智能，未来每家公司都将使用AI，每个国家都将建设AI基础设施。 以下为全文： AI 是塑造当今世界的强大力量之一。它并非仅仅是一款巧妙的应用程序，也不是单一的模型，而是如 同电力和互联网一样必不可少的基础设施。 AI 依托真实的硬件、能源和经济体系运行。它可以将原材料大规模地转化为智能。每家公司都将应用 AI， 每个国家/地区都将发展 AI。 要理解 AI 为何以这种方式发展，我们需要从基本原理进行推理，并了解计算领域发生 ...

核心观点 需求爆发&供给刚性，价格持续上涨，存储迎来超级周期。1）AI大模型技术超预期迭代升级，全球Token消耗量爆发式增长，由此带来海量的数 据存储、处理和检索需求，存储行业迎来超级景气周期。2）海外三大原厂将有限产能向高利润HBM和DDR5产品倾斜，对消费级和低端存储芯片 产能造成严重挤压，供需缺口扩大。3）存储原厂在上轮周期产能和资本开支过度后，本轮周期扩产动作均较为谨慎；且高端HBM存储芯片存在洁 净室建设周期长，良率爬坡困难等问题，短期供给持续偏紧。在需求爆发&供给刚性的背景下，据CFM闪存市场预计，2026年存储价格整体仍将 延续上涨。 需求端：本轮超级周期的根本原因在于，AI大模型技术超预期迭代升级，带来海量的数据存储、处理和检索需求。1）得益于生成式AI的快速迭代 ，AI agent、多模态应用与原生视频技术等新应用爆发式增长，由此带来海量数据的存储、处理和检索需求，直接推高了对HBM、DRAM、NAND 等存储介质的需求，开启一轮存储超级需求周期。2）根据Open Router，2026年2月同期全球主要大模型消耗的token是25年同期的10倍及以上， 彰显出极其旺盛的算力及存储需求。3 ...

Core Viewpoint - The artificial intelligence (AI) boom is significantly driving demand for graphics processing units (GPUs), leading to remarkable sales and stock performance for Nvidia [1] Group 1: Financial Performance - Nvidia's revenue for the fourth quarter of fiscal 2026 reached $68.1 billion, a 73% year-over-year increase, with data center revenue rising 75% to $62.3 billion [4] - The company anticipates continued growth, projecting first-quarter fiscal 2027 revenue of approximately $78.0 billion, indicating a 77% year-over-year growth [5] - Nvidia generated $34.9 billion in free cash flow in the fourth quarter, converting 51.2% of its revenue into free cash flow [6] Group 2: Share Repurchase Strategy - During fiscal 2026, Nvidia returned $41.1 billion to shareholders, primarily through share repurchases, with $58.5 billion remaining under its share repurchase authorization [8] - The company's share repurchases in fiscal Q4 were $3.8 billion, down from $7.8 billion in the previous year, reflecting a more cautious approach due to the stock's significant appreciation [9] - Nvidia's management is focused on opportunistic share repurchases, aiming to buy back shares when prices decline, which could help reduce the share count over time [10] Group 3: Management Approach - Nvidia's management emphasizes a data-driven approach to capital deployment, balancing share repurchases with investments in other growth opportunities [12][13] - The company's thoughtful share repurchase strategy, combined with its cash-generative business model, is expected to enhance total shareholder returns over the long term [14]

截至3月10日，科创芯片设计ETF天弘(589070)近10个交易日内，合计"吸金"1555.00万元。 【产品亮点】 通过聚焦科创板芯片设计企业，实现了三大细分赛道的全覆盖，既能够捕捉单一赛道的爆发式增长，又能通过赛道分散降低个股风险。当前半导体行业 处于复苏周期，叠加政策与需求双重利好，科创芯片设计ETF天弘(589070)的配置价值进一步凸显。 【热点事件】 英伟达GTC 2026开发者大会将至，全新推理芯片引关注 截至2026年3月10日收盘，科创芯片设计ETF天弘(589070)换手10.37%，成交6443.45万元，市场交投活跃。跟踪的上证科创板芯片设计主题指数(950162) 下跌1.13%。成分股方面涨跌互现，龙芯中科领涨6.90%，纳芯微上涨4.24%，佰维存储上涨3.50%。 英伟达GTC 2026开发者大会临近，Feynman新架构、Rubin Ultra强化平台及LPU推理芯片有望集中亮相，其中LPU采用大模型推理专用架构，片上 SRAM存储带宽达80TB/s，显著优于GPU的HBM方案，低时延与高能效特性将加速AI实时推理应用落地，并带动PCB、散热、电源及SRAM相关材料升 级需 ...

Core Viewpoint - The focus is on Oracle's upcoming earnings call, with particular attention to its cloud revenue growth, debt management, and the implications of its relationship with OpenAI [1][2][3]. Financial Performance - Oracle has historically utilized debt, with a higher debt-to-equity ratio a year ago, and investors are concerned about its ability to maintain a triple-B credit rating, which is barely investment grade [2]. - The company has a capital expenditure expectation of approximately $60 billion, with cloud infrastructure sales projected to increase by 79% [3]. Market Dynamics - There is a complex web of interrelated companies in the tech sector, leading to concerns about potential conflicts of interest and the sustainability of investments [4]. - Free cash flow among major hyperscalers has recently improved, contrasting with the rapid decline seen during the AI arms race [5]. Investment Strategy - The company is viewed as vulnerable due to its dependence on OpenAI, and there is a preference for companies with better diversification [6]. - Despite the challenges, the company has been a long-term holding, but there is caution about adding to positions until more clarity is achieved [7]. Market Volatility - Recent market fluctuations, including a significant drop in the Dow followed by a quick recovery, highlight the challenges of navigating investments in a volatile environment [9][12]. - The current market conditions are reminiscent of past periods of volatility, where long-term investors may find opportunities despite short-term challenges [11][12].

英伟达GTC 2026有望带来 *免责声明：文章内容仅供参考，不构成投资建议 *风险提示：股市有风险，入市需谨慎 | | 物理学家费曼命名 | SPF | | --- | --- | --- | | | | 及争 | | CPO | CPO将光引擎与交换芯片 | | | | 共同封装在同一IC载板或 | 本次 | | | 硅中介层上，相较于传统 | 向ブ | | | 网络，功率转换效率提升5 预计 | | | | 倍、网络韧性提升10倍， | 交并 | | | 而部署时间则加快5倍 | | | HBM4 | | 随着 | | | HBM4频宽较HBM3E提升2 提出 | | | | 倍以上，能效提升40%， | GT( | | | 可显著提升AI服务效能 | 元华 | | | | 下良 | | | | LPL | | | | 流才 | | | 英伟达收购Groq技术后推 | 则， | | LPU | 出的专用推理加速单元 | 优萝 | | | (芯片) | 量六 | | | | 次有 | | 产品/技术 | 定位 | | | --- | --- | --- | | Rubin Ultra | 下一代AI超级计 ...

在AI产业下半场转向端侧落地的关键节点，一家成立不到两年的中国芯片黑马横空出世，凭借颠覆性技术与顶尖团队实力，在资本寒冬中逆势突围，累 计斩获近10亿元融资，迅速跻身端侧AI芯片领域第一梯队。 光羽芯辰的快速崛起，离不开掌舵人周强的前瞻布局与硬核实力。这位1983年出生的"芯片老兵"，深耕芯片领域十余年，履历堪称中国芯片产业的进阶缩 影，也是业内罕见的全栈型技术创业者。 周强是复旦大学微电子学院博士、工信部集成电路领军人才，专业功底扎实、产业经验丰富，他的职业生涯从本土上市公司芯原微电子起步，随后加入被 誉为芯片行业"黄埔军校"的AMD，深耕芯片SoC系统研发与管理，后续又历任摩尔线程、燧原科技高管，担任燧原科技首席芯片战略官，主导AI芯片架构 前瞻性研发、战略生态建设与产业资源整合，还带领团队成功量产CPU、AI SoC、汽车AP等多款芯片，精通芯片设计、验证、量产全流程开发。 十余年行业深耕，让周强沉淀出两项核心能力，一是务实的工程化落地能力，能精准把控技术边界与落地难点，不做空中楼阁式研发；二是敏锐的市场洞 察力，习惯以客户需求倒推产品定义，拒绝技术自嗨。2022年ChatGPT引爆云端大模型热潮，赛道 ...

融中咨询认为，涨势至少将延续至2026年下半年，预计2026年一季度存储芯片价格将上升40%~50%， 二季度还将再涨约20%。DRAM供应短缺预计持续至2027年一季度，NAND短缺则延续至2026年三季 度。根据TrendForce数据，2026年DRAM位增长率或达26%，NAND位增长率或达21%，较2025年均有 加速。 公司方面，据上市公司互动平台、公告及券商研报表示， 隔夜美股存储公司再度大涨，闪迪、美光、西部数据等涨超5%此前英伟达CEO黄仁勋公开表示，"存储 器厂产能扩多少，英伟达就会用多少"。 另据韩媒Sedaily 3月8日报道，三星电子计划在第二季度再将NAND Flash价格大幅调高一倍。三星今年 一季度对NAND Flash供应合约价格已上调超过100%，这意味着仅三星一家，上半年NAND价格累计涨 幅已超200%。SK海力士、铠侠等其他NAND制造商也计划跟进调涨。 中信建投证券指出，AI浪潮引发存储芯片需求井喷，深刻改变了存储行业景气周期。单台AI服务器平 均配置1.7TB存储，远高于普通服务器的0.5TB，直接带动HBM、DDR5及企业级SSD供不应求。三季度 DRAM合约价 ...

Jensen Huang Thinks SoNvidia Earnings Live: Chipmaker's Results Blow Past Wall Street Estimates as Customers 'Are Racing to Invest in AI'Block Lays Off 40% of Staff Citing AI. CEO Dorsey Says Other Firms Will Make Similar Moves. -- Apple's Newest Phone and Laptop Are More Budget-Friendly. Premium Products Could Be Next Stocks Little Changed as Oil SlidesAre Investors Ready for 'Violent and Frequent Shocks'?Is War Pushing the Economy Toward 'Stagflation'?Anthropic Identifies the Jobs Most Exposed to AI- Top ...

Core Insights - GSI Technologies has over 30 years of experience in the semiconductor industry, primarily known for high-performance SRAM products used in various demanding applications [2] - The company's current product, Gemini-II, is designed for power and low latency constrained edge environments, such as drones and satellites, aiming to enhance performance per watt compared to traditional architectures [3] - GSI is focusing on the development of its next-generation technology, the Associative Processing Unit (APU), which features a proprietary compute and memory architecture [2][4] Product and Market Focus - The Gemini-II is currently being evaluated for defense and other edge applications, with the company working towards initial design wins [3] - GSI aims to capitalize on edge AI market opportunities by delivering architectures that provide the highest compute per watt, which is crucial for the performance demands of edge applications [4]