1）英伟达发布 4Q25 业绩，单季度每股盈利 1.62 美元，较分析师预期的 1.53 美元高出 5.81%，同比+82.02%；单季度 营收 681.27 亿美元，较分析师预期的 660.03 亿美元高出 3.22%，同比+73.21%。同时公司给出 1Q26 780 亿美元业绩 指引，超分析师预期的 727.8 亿美元，整体业绩&指引双 beat。但从股价看市场反应温和，我们认为主要系担心竞争 格局、预期交易充分的原因。展望 3 月，NVIDIA GTC 大会即将开幕。将首次公开下一代核心产品 Feynman 芯片，且将 搭载全球首款 1.6nm 制程工艺，同时我们预计有望展示 CPO 交换机、rubin ultra 等方案，或将给再催化光通信板块。 2）旭创发布业绩快报，2025 年全年收入 382.40 亿元，我们测算对应 4Q25 单季度收入 132.35 亿元，同比+102%/环比 +30%，增长稳健。2025 年全年营业利润 135.97 亿元，我们测算对应 4Q25 单季度营业利润 47.59 亿元，同比+190%/环 比+20%。整体处于此前业绩预告中枢。利润增速小于收入增速，我们认为主要 ...

光模块：本周光模块指数+4.84%，本月以来光模块指数+4.14%。NV 业绩&指引双 beat。但从股价看市场反应温和， 我们认为主要系担心竞争格局、预期交易充分的原因。展望 3 月，NVIDIA GTC 大会即将开幕,我们预计有望展示 CPO 交换机、rubin ultra 等方案，或将给再催化光通信板块。 IDC：本周 IDC 指数+2.41%，本月以来，IDC 指数+2.44%。OpenRouter 数据显示，2026 年 2 月 9 日至 15 日当周，中 国模型的调用量为 4.12 万亿 Token，首次超过美国模型（2.94 万亿 Token）。中国领先的芯片制造商正努力在两年 内将采用尖端工艺技术制造的芯片产量提高五倍，以满足国内人工智能行业的需求。 核心数据更新： 电信业务量收增速逐步提升。2025 年电信业务收入累计完成 1.75 万亿元，同比增长 0.7%。按照上年不变价计算的电 信业务总量同比增长 9.1%。12 月我国光模块出口金额当月同比增加 0.9%；1-12 月累计同比降低 16%。4Q25 微软/谷 歌/Meta/亚马逊资本支出分别为 299 亿/279 亿/214 亿/39 ...

行 业 研 究 2026 年 03 月 01 日 投资评级：看好（维持） 行业走势图 数据来源：聚源 -38% -19% 0% 19% 38% 58% 77% 2025-03 2025-07 2025-11 电子 沪深300 相关研究报告 《海内外大模型密集更新，推动 AI 算 力需求持续增长—行业点评报告》 -2026.2.23 《JX 金属指引明确行业景气度，靶材 成为半导体金铲子—行业点评报告》 -2026.2.13 《关注半导体自主可控和涨价连锁反 应—行业点评报告》-2026.2.9 SK 海力士打响存储扩产第一枪，3 月看好设备+耗材 扩产链 ——行业点评报告 陈蓉芳（分析师） 向俊儒（联系人） chenrongfang@kysec.cn 证书编号：S0790524120002 xiangjuru@kysec.cn 证书编号：S0790125070018 事件：SK 海力士 150 亿美金追投，打响全球存储扩产第一枪 2 月 25 日，SK 海力士宣布 2030 年前投资 21.6 万亿韩元（约 150.7 亿美元）， 用于建设首座厂房及洁净室设施，预计洁净室的开放时间将从原定的 2027 年 5 ...

GTC 2026焦点或从Vera Rubin转向Feynman Chosun Biz的报道指向一个关键信号，英伟达准备在GTC 2026把叙事重心从Vera Rubin转向Feynman。 当前市场高度关注GTC大会。英伟达可能在GTC大会中抛出下一代芯片代号Feynman，并首次公开展 示采用台积电A16，1.6nm工艺的产品方向，这将把市场对其算力路线图的关注点，从Vera Rubin进一步 推向更远的周期。 据Wccftech援引韩国媒体Chosun Biz报道，英伟达的GTC 2026演讲规划已"超越Vera Rubin"，今年的大 会可能成为Feynman的首次公开亮相。GTC 2026将于3月15日开幕，活动回到美国加州圣何塞举行。 黄仁勋此前也表示，其主题演讲将展示"从未公开过"的技术。对投资者而言，这类表态往往意味着新一 轮产品节奏与关键供应链选择即将被确认，尤其是先进制程与封装形态的取舍。 如果Feynman确实采用台积电A16，Wccftech认为英伟达将成为该节点初期大规模量产阶段的首家，甚 至可能是唯一客户，这将把先进产能与良率爬坡的市场预期进一步绑定到英伟达。 同时，市场还在评估Fe ...

Core Viewpoint - NVIDIA aims to dominate the inference stack with its next-generation Feynman chip by integrating LPU units into its architecture, leveraging a licensing agreement with Groq for LPU technology [1][18]. Group 1: NVIDIA's Strategy and Technology Integration - NVIDIA plans to integrate Groq's LPU units into its Feynman GPU architecture, potentially using TSMC's hybrid bonding technology for stacking [1][3]. - The LPU modules are expected to enhance inference performance significantly, with Groq's LPU set to debut in 2028 [5]. - The Feynman core will utilize a combination of logic and compute chips, achieving high density and bandwidth while maintaining cost efficiency [6]. Group 2: Inference Market Dynamics - The AI industry's computational demands have shifted towards inference, with major companies like OpenAI and Google focusing on building robust inference stacks [9]. - Google’s Ironwood TPU is positioned as a competitor to NVIDIA, emphasizing the need for low-latency execution engines in large-scale data centers [9][10]. - Groq's LPU architecture is designed specifically for inference workloads, offering deterministic execution and on-chip SRAM for reduced latency [10][14]. Group 3: Licensing Agreement and Market Position - NVIDIA's agreement with Groq is framed as a non-exclusive licensing deal, allowing NVIDIA to integrate Groq's low-latency processors into its AI Factory architecture [18][21]. - This strategy is seen as a way to circumvent antitrust scrutiny while acquiring valuable talent and intellectual property from Groq [19][21]. - The transaction is viewed as a significant achievement for NVIDIA, positioning LPU as a core component of its AI workload strategy [16][21].

Investment Rating - The report assigns a "Buy" rating for NVIDIA (NVDA US) with a target price of $168.00, indicating a potential upside of 45.5% from the current price of $115.43 [1][12]. Core Insights - The report highlights NVIDIA's transition towards generative AI, with CEO Jensen Huang predicting that capital expenditures in the cloud service provider (CSP) sector could exceed $300 billion by 2025 and reach $1 trillion by 2028 [2]. - NVIDIA is actively participating in AI transformations across vertical industries, including mobile communications and smart driving, with a vision of establishing "AI factories" for production planning and intelligent assistance [2]. - The company continues to release new products, with significant upgrades in data center offerings, emphasizing the importance of scale-up architecture over scale-out in server design [6]. Financial Overview - Revenue projections for NVIDIA show substantial growth, with expected revenues of $60,922 million in 2024, increasing to $311,585 million by 2028, reflecting a compound annual growth rate (CAGR) of 15.2% [5][14]. - Net profit is forecasted to rise from $32,312 million in 2024 to $180,404 million in 2028, with a notable increase in earnings per share (EPS) from $1.30 to $7.30 over the same period [5][14]. - The report indicates a significant improvement in profit margins, with gross margins expected to remain around 75% and net margins projected to increase to 57.9% by 2028 [16]. Product Development and Roadmap - NVIDIA's product roadmap includes the launch of the Blackwell Ultra chip in the second half of 2025, followed by the Rubin and Rubin Ultra chips in 2026 and 2027, respectively, showcasing substantial performance upgrades [6][7]. - The introduction of new switch products, Spectrum-X and Quantum-X, is planned for 2025 and 2026, respectively, with advancements in technology such as optical-electrical hybrid packaging [8]. Market Position and Valuation - NVIDIA's current price-to-earnings (P/E) ratio is noted at 24 times the FY26 earnings, which is considered attractive compared to its historical average of 38 times [6]. - The report emphasizes NVIDIA's strong product and technology advantages, particularly in the early stages of AI deployment both domestically and internationally [6].