半导体产业链"涨价潮"持续。继英飞凌、士兰微、宏微科技等厂商后，近日，又有功率半导体厂商发布 价格调整通知。年初至今，国内外多家芯片龙头企业陆续提价，涨幅位于10%-80%，覆盖MCU、功率 器件等品类。据悉，多数企业在调价原因中均提及，上游原材料及关键贵金属价格大幅上涨是本轮提价 的主因，而供需格局失衡进一步助推动芯片产品涨价。 ...

Market Volatility and Currency Intervention - The People's Bank of China (PBOC) has reduced shorting costs to halt the Yuan's recent rally, which reached its strongest levels since mid-2023, potentially threatening the export-led recovery [2] - Analysts believe Beijing is trying to create a "slow bull" market to avoid historical boom-bust cycles in Chinese equities [2] Regional Performance and Record Highs - The Hang Seng Index (HSI) is facing its steepest monthly decline since October, down 3.2% in February, driven by concerns over AI disruption affecting traditional sectors [3][9] - Malaysia's FBM KLCI index fell over 1% to 1,722.34 points, its lowest since late January, but remains nearly 10% higher than a year ago due to growth in financial services and utilities [4] - Hyundai Motor (HYMTF) shares surged over 6% to a record high, driven by a 60 trillion won proposal for hydrogen fuel cell infrastructure in Canada and optimism surrounding its robotics division [5][9] Geopolitics and Domestic Policy - Russia has warned Britain against deploying troops to Ukraine, stating it could lead to a large-scale military confrontation [6] - A meeting between President Trump and New York City Mayor Zohran Mamdani focused on a proposal for 12,000 housing units at Sunnyside Yards, indicating potential federal-local cooperation on urban development [7] Commodities and Corporate Trends - Copper prices are rising, supported by positive global demand and industrial growth prospects, as Chinese firms seek foreign expertise for global expansion [10]

Investment Summary - Nvidia invested approximately $7.93 billion in Intel Corporation (INTC) during the October-to-December quarter of 2025, resulting in a stock price increase from $24.21 to $46.51, a gain of roughly 92% [1] - Nvidia also invested around $1.08 billion in Nokia (NOK), with the stock rising from $4.23 to $7.57, marking a gain of approximately 79% [2] - However, Nvidia's investment in Synopsys Inc. worth about $2.26 billion saw a decline of around 24% since September 2nd [2] Existing Investments Performance - The value of Nvidia's stake in CoreWeave decreased from $3.32 billion to $1.58 billion, a drop of nearly 50% [3] - Nvidia's investment in Nebius fell from $133.65 million to approximately $99.65 million [3] Divestments - Nvidia divested its entire stakes in four firms: Arm Holdings, Applied Digital Corporation, Recursion Pharmaceuticals, and WeRide Inc., with a combined worth of $387.64 million [4]

寒武纪-U成交额达100亿元，现涨2.5%。 ...

（原标题：芯片制造重大突破？阿斯麦(ASML.US)下一代EUV已能用于大规模量产，造价翻倍！） 据报道，全球光刻机龙头阿斯麦(ASML)的一位高管最新表示，其下一代芯片制造设备已准备就绪，可以供应芯片制造商用于大规模量产，这对芯 片行业来说是一个重大进展。 这家荷兰公司生产着全球唯一的商用极紫外光刻(EUV)设备，这是芯片制造商不可或缺的关键部件。阿斯麦的数据显示，这款新设备将帮助台积 电和英特尔等芯片制造商生产出更强大、更高效的芯片，因为它省去了芯片制造过程中的一些昂贵且复杂的步骤。 阿斯麦首席技术官Marco Pieters周三表示，该公司计划于周四在圣何塞举行的技术会议上公布这一数据，这标志着一个重要的里程碑。 据悉，阿斯麦花了数年时间才研发出这些昂贵的下一代设备，而芯片制造商们则一直在试图确定在何种情况下使用这些设备进行大规模生产才具 有经济意义。 但鉴于当前这一代EUV设备在制造复杂AI芯片方面已接近技术极限，下一代设备(被称为高数值孔径EUV工具)对于AI行业至关重要，可以改进 OpenAI的ChatGPT等聊天机器人，并帮助芯片制造商按时完成其AI芯片研发线路图，以满足不断增长的需求。 不过需 ...

来源：搜狐财经 2月21日，英伟达公布了2026财年第四季度财报，再次以碾压式的数字震撼市场。总营收高达681.3亿美 元，同比增长73%；GAAP净利润达到429.6亿美元，同比猛增94%——相当于平均每天净赚超过4.7亿美 元。营收、利润、现金流三项核心指标均创历史单季新高。 这份财报的核心逻辑只有一个词：AI。所有增长几乎全部由数据中心业务驱动，其单季收入达到623.1 亿美元，占总营收比例历史性地突破了91.5%。这个数字意味着，英伟达已经从一家"游戏显卡公司"彻 底变为全球AI基础设施的核心供应商。 数据中心业务内部，有一个更值得关注的结构性变化：网络业务（Networking）的收入同比暴增 263%，达到110亿美元。这背后是NVLink互联技术的爆发，英伟达通过"计算芯片+高速网络"的系统级 方案，正在把自己的护城河越挖越深。CEO黄仁勋将其描述为"AI工厂"的基础设施：不只是卖一块芯 片，而是卖一套完整的"算力工厂"解决方案。 在一片叫好声中，市场的目光更聚焦于AI的下一战场——推理（Inference）。这里的推理是指AI模型训 练完成后，真正被用户调用、生成回答的过程，所消耗的算力就是推 ...

存储芯片巨头大战下，HBM还在进化，而新型内存HBF（High Bandwidth Flash，高带宽闪存）也加快 了商业化进程。 近日，SK海力士与闪迪联合举办"HBF规格标准化联盟启动"活动，宣布在开放计算项目（OCP）框架 下成立专属工作组，推进HBF的全球标准化。 消息传出后，闪迪盘前股价拉升5%，足以看出市场对该技术的高度期待。 什么是HBF？这并非全新概念，其源于AI产业从训练向规模化推理的转型。 作为"AI推理时代存储缺口解决方案"的核心技术，HBF的标准化启动，标志着下一代存储架构竞争进入 实质阶段。 随着多模态大模型普及，AI服务并发量激增，现有存储架构陷入明显瓶颈：HBM带宽优异但容量有 限、成本高昂，单堆栈仅数十GB；而TB级的SSD虽容量充足，却读写偏慢，无法适配AI推理的实时性 需求。 在全新的AI存储架构下，HBM将负责延迟敏感型任务，而HBF会接替HBM专注于大容量顺序读取的工 作，例如模型参数读取。 两者分工协作、优势互补，精准满足AI推理的性能与功耗双重需求，SK海力士推出的"H3混合架构"就 是用来验证这一逻辑。 该架构将HBM与HBF并列部署于GPU周边，仿真测试显示，其 ...

《广发海外电子通信》 ☄GTC 2026前瞻：LPX、CPO及PCB关键亮点 ☀LPX机架强化英伟达在推理领域的产品组合：如先前所述，LPX（又称LPU）预计将采用基于SRAM的片上内 存，提供快速Token生成及超低延迟，提升英伟达在推理领域的地位。 在英伟达于2025年12月与Groq达成非独家许可协议前，该机架设计将搭载64个Groq LPU，采用RealScale芯片间互 连。 《广发海外电子通信》 ☄GTC 2026前瞻：LPX、CPO及PCB关键亮点 ☀LPX机架强化英伟达在推理领域的产品组合：如先前所述，LPX（又称LPU）预计将采用基于SRAM的片上内 存，提供快速Token生成及超低延迟，提升英伟达在推理领域的地位。 在英伟达于2025年12月与Groq达成非独家许可协议前，该机架设计将搭载64个Groq LPU，采用RealScale芯片间互 连。 对于GTC 2026，我们预计增强型LPX机架将搭载256个LPU，采用多层52LM9 Q-glass PTHPCB，单个LPU对应PCB 价值量约200美元。 因此，我们预计PCB、液冷（冷板）公司将受益。 我们预计英伟达将于GTC 2026 ...

Core Insights - Nvidia reported a significant earnings increase, with revenue rising 73% to $68.1 billion, surpassing estimates of $66.2 billion, and adjusted net income increasing 79% to $39.5 billion, or $1.62 per share, exceeding expectations of $1.54 [1][2] Group 1: Nvidia's Performance - Nvidia's profit margin reached 58%, showcasing its dominance in the AI sector and reflecting strong demand for its chips and AI computing power [2] - Despite the impressive earnings report, investor reaction was muted, with the stock initially rising over 3% but ultimately finishing the after-hours session flat [3][4] - Nvidia's market capitalization is nearing $5 trillion, making it challenging for the stock to experience significant upward movement despite strong growth numbers [4] Group 2: Investment Opportunities - The VanEck Semiconductor ETF (SMH) is highlighted as a strategic investment to capitalize on the AI boom, providing diversified exposure to leading AI stocks [5][7] - The SMH ETF includes top holdings such as Nvidia, Taiwan Semiconductor, Broadcom, Micron, and ASML, which collectively represent about 50% of the fund [7][8] - Other semiconductor stocks within the ETF, like Micron and ASML, have shown substantial growth, with Micron tripling in value over the last six months due to memory chip shortages [9][10] Group 3: Market Outlook - Nvidia's success is expected to positively impact its top holdings, with the potential for other chip stocks to have more upside due to their lower market capitalizations compared to Nvidia [10] - The ongoing AI boom is supported by Nvidia's results, indicating that concerns about an AI bubble are premature [11][12] - The SMH ETF, while not inexpensive with a price-to-earnings ratio of 45, offers a broad range of AI stock exposure and the potential for significant returns as AI spending continues to rise [12]

消息面上，此前，英伟达正式宣布其新一代AI超级计算平台Vera Rubin已进入全面投产阶段。根据英伟 达公布的数据，Rubin GPU搭载第三代Transformer引擎，NVFP4推理/训练算力达到50/35PFLOPS，达到 前代Blackwell的5/3.5倍；HBM4带宽22TB/s，为前代的2.8倍。中邮证券表示，Rubin GPU升级为 HBM4，成为与GPU紧紧绑定的"计算核心"。 有业内人士分析，2025年国产GPU已实现从"单点突破"到"系统级替代"的跨越。随着资本化进程加速、 生态体系完善，国产算力正从追赶国际巨头转向定义自主架构，在AI驱动的算力新时代构建坚实底 座。华鑫证券认为，天数智芯天垓系列训练GPU和智铠系列推理GPU已实现规模化应用，新发布的"彤 央"系列边端产品实测性能超越英伟达同类产品。该行认为，随着2026年AI产业"云端深耕+边缘爆发"格 局成型，公司GPU业务将成为业绩增长引擎。 智通财经APP获悉，天数智芯(09903)一度涨超7%，盘中高见310.6港元创上市新高。截至发稿，涨 2.8%，报294港元，成交额1.62亿港元。 ...