Core Viewpoint - The article discusses the acceleration of global storage technology innovation driven by the increasing memory demands of AI computing, highlighting SK Hynix's new storage architecture called H³, which combines HBM and HBF technologies to significantly enhance performance in AI applications [3][6]. Group 1: H³ Architecture - SK Hynix introduced the H³ architecture at the IEEE global semiconductor conference, which integrates HBM and HBF technologies to improve performance [3]. - In simulations, the H³ architecture demonstrated a performance increase of up to 2.69 times per watt compared to using HBM alone when paired with NVIDIA's Blackwell GPU [3][5]. Group 2: Advantages in AI Inference - The H³ architecture is particularly advantageous for AI inference, where the key-value cache (KV cache) is essential for managing data interactions between AI services and users [5]. - By utilizing HBF for KV cache storage, the burden on GPU and HBM is reduced, allowing them to focus on high-speed computations and data creation [6]. Group 3: Industry Development - SK Hynix, Samsung, and SanDisk are advancing HBF technology, with SK Hynix planning to release HBF1 samples by the end of this year, featuring 16-layer NAND flash stacking [6]. - Samsung and SanDisk aim to implement HBF technology in actual products from NVIDIA, AMD, and Google by late 2027 or early 2028 [6]. Group 4: Market Potential - Current AI models require memory capacities that exceed HBM capabilities, with HBF expected to provide 8 to 16 times the storage capacity of existing HBM, potentially expanding GPU storage to 4TB [7]. - Niche storage markets are anticipated to grow due to AI demand, with SLC NAND likely to be used in AI SSD products and HBF, leading to sustained price increases in the niche storage sector [7].

截至2月12日11:06，上证科创板半导体材料设备主题指数下跌0.34%。成分股方面涨跌互现，欧莱新材 领涨18.32%，晶升股份上涨2.72%，上海合晶上涨1.6%；耐科装备领跌3.38%，兴福电子下跌2.82%，中 微公司下跌1.99%。科创半导体ETF（588170）下跌0.45%，最新报价1.75元。 截至2月12日11:08，中证半导体材料设备主题指数下跌0.12%。成分股方面涨跌互现，江丰电子领涨 8.56%，广立微上涨5.79%，晶升股份上涨3.03%；拓荆科技领跌1.96%，京仪装备下跌1.87%，华峰测控 下跌1.84%。半导体设备ETF华夏（562590）下跌0.16%，最新报价1.9元。 流动性方面，科创半导体ETF盘中换手4.25%，成交3.45亿元；半导体设备ETF华夏盘中换手2.25%，成 交6131.29万元。 规模方面，科创半导体ETF近1周规模增长2.05亿元，实现显著增长，新增规模领先同类；半导体设备 ETF华夏最新规模达27.06亿元。 资金流入方面，科创半导体ETF最新资金净流入1058.84万元。拉长时间看，近5个交易日内有3日资金 净流入，合计"吸金"1.21亿元， ...

存储涨价潮，正从DRAM蔓延向NAND。 继1月三星电子将NAND闪存合约价格上调100%以上后，2月以来，多家市场研究机构纷纷上修了对 NAND闪存价格的预测。 2月2日，TrendForce集邦咨询上修了第一季DRAM、NAND Flash各产品价格季成长幅度，NAND Flash 合约价从季增33-38%上调至55-60%，并且不排除仍有进一步上修空间。 Counterpoint也在2月3日表示，根据该机构的2026年1月存储价格追踪报告，NAND闪存价格预计将在本 季度上涨超过40%。 "闪存涨价主要还是AI需求量太大了。"IO资本创始合伙人赵占祥对21世纪经济报道记者表示，AI推理 以及最新一代AI大模型倾向于在推理过程中做一些训练，双双带动了闪存的需求。 推理催化闪存需求 随着生成式AI从训练阶段全面迈入大规模推理应用，NAND闪存需求一路走高。 AI对NAND的需求，主要来自RAG（检索增强生成）：为了解决大语言模型普遍存在的"幻觉"痛点，能 够通过检索外部知识库来增强生成准确性的企业级RAG应用正加速渗透落地。 然而，这要求企业将TB乃至PB级的私有知识库进行向量化并持久化存储。在推理阶段，系 ...

行业研究丨深度报告丨电子设备、仪器和元件 [Table_Title] 美光科技：乘 AI 之东风，存储龙头高速增长 %% %% %% %% research.95579.com 1 丨证券研究报告丨 报告要点 [Table_Summary] 美光科技是全球存储的龙头企业，以 2025 年统计口径看，公司 DRAM、NAND Flash 全球市 占率分别为 23%、13%。在 45 年的发展史中始终保持产品的推出处于业内领先水平；公司以 IDM 的模式围绕核心存储持续展开产品布局，现拥有 DDR、LPDDR、GDDR、HBM、NAND Flash 等主要存储产品矩阵。近年来，AI 对于存储需求的持续拉动，助力公司 Cloud Memory 营收呈高速增长之势。就新产能规划而言，公司现已启动系列晶圆厂的建设工作，我们认为伴 随着公司新产能的逐步建成落地，美光有望长期在存储行业处于领先地位。 分析师及联系人 [Table_Author] 杨洋 张梦杰 SAC：S0490517070012 SAC：S0490523120002 SFC：BUW100 请阅读最后评级说明和重要声明 2 / 24 %% %% %% %% ...

Investment Rating - The report suggests a focus on companies such as SanDisk (SNDK.O), SK Hynix (000660.KS), Samsung Electronics (005930.KS), and Micron Technology (MU.O) due to their potential in the AI-driven storage market [4]. Core Insights - The industry is transitioning from traditional cycles to AI-driven technological premiums and high growth, with storage becoming a critical bottleneck for AI computing efficiency [5]. - AI is driving exponential growth in storage demand, with significant increases in requirements for high-bandwidth memory (HBM) and solid-state drives (SSD) due to the scaling laws of model parameters [5]. - The valuation framework is being reshaped, with revenue growth and net profit margin expansion occurring simultaneously, indicating a period of high operational leverage in the storage sector [5]. Summary by Sections AI Value Chain - The AI value chain includes components such as computing chips (GPUs, CPUs, ASICs), storage (DRAM, SSD, HDD), networking, and energy solutions, highlighting the interconnectedness of these sectors [10][12]. - The storage segment is characterized by oligopolistic features, with high barriers to entry supporting high profitability [12]. Storage Demand and Hardware Architecture - Storage demand is categorized into cold, warm, and hot data, with different storage solutions (HDD, SSD, DRAM) tailored to each category's speed and capacity requirements [20]. - The architecture is structured in layers, with HBM for high-speed access, SSD for intermediate storage, and HDD for large capacity [29]. Large Language Model Storage Requirements - The report details the storage needs of large language models, emphasizing the importance of storage capacity and bandwidth in both training and inference phases [42][47]. - The growth in model parameters necessitates increased storage capacity, with significant implications for the demand for NAND and DRAM technologies [59]. Company-Specific Insights - Micron Technology (MU.O) is focusing on data centers and has made significant advancements in HBM3E technology, with a strong market demand for its products [64][70]. - SanDisk (SNDK.O) has undergone a restructuring, achieving notable revenue growth and improving profit margins through a focus on high-value enterprise SSDs [72][78]. - SK Hynix is benefiting from its early entry into the AI storage market, while Samsung is striving to catch up [79][81].

海力士在存储行业表现突出，DRAM 产品线完善，并推出新型存储产品 HBF，预计 2026 年下半年出样品，2027 年实现商业化，将显著提升 推理速度，满足 AI 推理需求。海力士在 HBM 技术方面具有领先优势。 英特尔目标价上调至 71.5 美元，基于其制程工艺的显著进步和 IDM 2.0 战略的成功，吸引了政府和私营部门的投资，具备先进封装能力， 外部客户需求将成为推动其进一步发展的关键因素。 全球半导体龙头业绩启示：苹果，ASML，Hynix，三星， Advantest，DISCO20260201 摘要 2026 年半导体行业规模预计接近 1 万亿美元，存储领域增长最为显著， 硬件板块预计持续跑赢软件板块，原材料、存储、半导体设备表现优异， 消费电子品牌受损最大，苹果相对影响较小。 预计 2026 年全球智能手机出货量将下降 6.7%，苹果和三星基本持平， 中国品牌预计下降 14%，主要受存储缺货影响，华为、荣耀、小米、 OPPO、vivo 等品牌均会受到不同程度的影响。 微软和 Meta 报告显示，2026 年 CSP 资本开支预计增长 43%，Meta 从 700 亿美金增加到 1,200 亿美 ...

Summary of Conference Call Notes Industry: Storage Chips - HBF is expected to partially replace HBM in AI servers, balancing performance and cost, with mass production anticipated in Q4 2026 to Q1 2027 at a price of approximately $10–11 per GB [1][2] - HBF is beneficial for SanDisk and Kioxia as they do not engage in HBM business, allowing them to expand their market through HBF [1][2] - Current supply and demand for HBM are generally balanced [1][2] - Production capacity is planned to expand to 476,000 wafers per month by 2026, suggesting a stable to declining price for HBM in 2026 [2] Industry: RF Chips - The RF chip industry is expected to see moderate recovery in 2026, with intense price competition in the 4G sector, while the 5G sector's L-PAMiD modules maintain a profit margin exceeding 20% with relatively eased competition [2] - Satellite direct connection in mobile phones is emerging as a new growth area, with the Mate80 series supporting low-altitude direct connection, primarily in collaboration with Zhaoshengwei; Xiaomi, Vivo, OPPO, and Samsung are following suit [2] Industry: AI Programming - Current AI programming tools are categorized into three main types: plugin-based, AI-native IDEs, and Agent types, represented by GitHub Copilot, Cursor, and Claude Code respectively [2] - GitHub Copilot shows the fastest commercialization progress with a monthly active user payment rate exceeding 20%; Cursor's latest ARR has reached $1 billion; Claude Code's API call volume is approximately 60% of Anthropic's, indicating significant revenue potential [3] - Leading domestic programming models include DeepSeek, Zhipu, Alibaba Qianwen, and Kimi, with a focus on the B-end market, while C-end free IDE products are currently underperforming [3] Industry: Tires - The global demand for giant tires is expected to grow by 35% from 2025 to 2029, driven primarily by increased demand from overseas mining projects [3] - Foreign brands like Michelin, Bridgestone, and Goodyear plan to raise giant tire prices by over 10% in 2026, while domestic brands like Hai'an will not increase prices to capture market share [3] - Hai'an's overseas growth this year is primarily focused on markets in Russia, Northwest Africa, and South Africa, with other domestic brands like Sailun and Zhongce also accelerating their international expansion [3] Industry: Pharmacies - Recent policy documents appear macro in nature and lack specific measures, but they provide a framework and space for subsequent detailed regulations from various ministries [3] - The industry is still undergoing a natural clearance process, with an expected annual exit of 10,000 to 20,000 stores, predicting a dynamic balance when the total number of stores stabilizes around 600,000 [3] - The O2O average transaction value has increased from below 50 yuan to approximately 55 yuan, with future O2O growth expected to maintain over 20% [3] Industry: Innovative Drugs - Competition in the CXO sector from South Korea is intensifying, with Samsung entering the ADC and cell therapy production markets [4] - To address patent cliff issues, BMS has launched seven new core products, while Merck has engaged in extensive mergers and acquisitions to enter new disease areas [4] - Major pharmaceutical companies are actively investing in AI, but few have the capability for significant computational investment like Eli Lilly [4]

Summary of Key Points from Conference Call Records Storage Chips 1. HBF is expected to partially replace HBM in AI servers, balancing performance and cost, with mass production anticipated between Q4 2026 and Q1 2027 at a price of approximately $10–11 per GB [1] 2. HBF is beneficial for SanDisk and Kioxia as they do not engage in HBM business, allowing them to expand their market through HBF [1] 3. The supply and demand for HBM is currently balanced, with plans to increase production capacity to 476,000 wafers per month by 2026, suggesting a stable decline in HBM prices in 2026 [1] RF Chips 1. The RF chip industry is expected to experience moderate recovery in 2026, with intense price competition in the 4G sector, while the 5G sector's L-PAMiD modules maintain a profit margin exceeding 20%, indicating relatively eased competition [2] 2. Satellite direct connection for mobile phones is emerging as a new growth driver, with the Mate80 series supporting low-altitude direct connection in collaboration with Zhaoshengwei; Xiaomi, Vivo, OPPO, and Samsung are following suit [2] AI Programming 1. Current AI programming tools are categorized into three main types: plugin-based, AI-native IDEs, and agent-based, represented by GitHub Copilot, Cursor, and Claude Code respectively [2] 2. GitHub Copilot is leading in commercialization with a monthly active user payment rate exceeding 20%; Cursor's latest ARR has reached $1 billion; Claude Code's API call volume is approximately 60% of Anthropic's, indicating significant revenue potential [2] 3. Leading domestic programming models include DeepSeek, Zhiyu, Alibaba Qianwen, and Kimi, with a focus on the B2B market, while C2C offerings remain free and IDE products are currently underperforming [2] Tires 1. The global demand for giant tires is projected to grow by 35% from 2025 to 2029, primarily driven by increased demand from overseas mining projects [3] 2. Foreign brands like Michelin, Bridgestone, and Goodyear plan to raise giant tire prices by over 10% in 2026, while domestic brands like Haian will not increase prices to capture market share [3] 3. Haian's growth in overseas markets this year is mainly focused on Russia, Northwest Africa, and South Africa, with other domestic brands like Sailun and Zhongce also accelerating their international expansion [3] Pharmacies 1. Recent policy documents appear macro in nature and lack specific measures, but they provide a framework and space for subsequent detailed regulations from various ministries [4] 2. The industry is still undergoing a natural clearance process, with an expected annual exit of 10,000 to 20,000 stores, suggesting a dynamic balance may be reached when the total number of stores stabilizes around 600,000 [4] 3. The average order value for O2O has increased from below 50 yuan to approximately 55 yuan, with future O2O growth expected to maintain over 20% [4]

AI需求创造突破口 NAND正在从周期品重新变成AI基础设施的关键部件，需求端的结构性变化抬升了价格与估值，也把曾 被视为二线阵营的铠侠与闪迪推到聚光灯下。 随着AI工作负载从训练转向推理，SSD在AI数据中心的部署扩大，数据中心对NAND闪存的采购规模激 增，为非传统领先厂商创造了市场机遇窗口。 铠侠与闪迪维持了超过25年的深度合作，通过合资与联合开发，双方共同分担昂贵设备与研发支出，以 规模效应摊薄成本。 在技术路线上，闪迪聚焦将HBF集成进英伟达、AMD与Google的产品。铠侠则更侧重SSD端的性能跃 迁。 市场重心分化，押注不同技术路线 市场重心上，据Knowing.asia报道，铠侠和闪迪两者的市场重心明显分化，铠侠主要向日本及全球大型 电子制造商供货，闪迪则主导消费存储市场，并在北美及海外企业级SSD领域占据强势地位。 技术路线上，HBM受限于容量与成本，NAND通过HBF（高容量近存储）与AI SSD（智能预处理）两 种技术路线优化AI架构。 据EE Times报道，随着AI工作负载从训练转向推理，SSD在AI数据中心的部署扩大，NAND因此迎来新 的需求推力。 对投资者而言，这意味着NAND ...

今年以来，在AI应用和电力这两个板块，大家可以看到，不断有破圈的AI应用出来，opencode和 CoWork还么研究明白，最近clawdbot又突然爆火；在电力板块也是一样的，冷不丁就会有个新闻， 上周达沃斯论坛上，马斯克提到电力是AI规模化的核心瓶颈，今天紧接着出来的两个跟电力相关新 闻。这说明了AI应用和电力这个话题，会是今年不断持续的大题材。 ClawdBot 的核心定位是一个全天候运行的开源 AI 代理员工，它强调通过直接部署在用户的本地硬 件（如 Mac mini 或 RTX 4090）上来实现完全的隐私保护与系统掌控权，彻底摆脱对云端平台的依 赖 。它被视为一个真正能够执行实操任务的自动化引擎，不仅能处理日常的邮件与日程管理，还能 跨越到自主市场研究甚至是视频剪辑等高级领域 。通过与 WhatsApp 或 Telegram 等现有聊天工具集 成，它让用户能像指挥真实员工一样，通过对话即可驱动 AI 在本地系统执行复杂的自动化工作 。 AI应用 在之前 的文章中，我们就提到 过AI应用的产业拐点已至，即便前段时间因为监管的因素，AI应用回 撤都非常大，但这个观点我们依然坚持。具体的逻辑之前讲过很多 ...