AI Agent 推动 CPU 重定价，关注 Infra 与应用层变化 20260126 摘要 互联网大厂广告代理商链条（如省广、蓝标等）有望受益于大模型产品 广告份额增长，成为关注焦点。同时，AI 应用泛化领域如博纳影业、昆 仑万维等也值得关注。 恒生科技板块中，消费互联网（阿里巴巴、京东健康、贝壳）、算力 （中芯国际、华虹半导体）、软件应用（快手、金蝶、哔哩哔哩）、智 能驾驶（小马智行、文远知行、小鹏汽车）及游戏（三七互娱、腾讯、 心动公司）等子板块均存在投资机会。 存储行业正经历技术升级周期，NAND 和 DRAM 合约价格均超预期上涨， 一季度涨幅分别达到 100%以上和 70%。存储需求显著增加，而供给侧 相对刚性，预计未来 NAND 价格仍有超预期上涨可能。 存储行业投资机会集中在下游存储涨价、上游成本相对稳定的设计公司 （如军政、兆易创新），含存量高的设备公司（如中微公司、拓荆科 技），以及受益于存储涨价周期的模组公司。 Q&A 目前对 A 股传媒板块的看法是什么？ 目前我们认为 A 股传媒板块的 AI 应用行情并未结束，产业趋势强劲支撑这一结 果。尽管近期有所调整，但整体行情的确定性依然较强。 ...

最近CPU的话题依然很火，机构们也一直在路演，讲CPU的逻辑，这篇文章我们先讲一些产业信 息，再对CPU的产业做个简单的科普。 在之前的文章中，我们也大概梳理过这边CPU的涨价逻辑，再更新一下： 1、对CPU的利好，肯定需要增量的CPU，无论是CPU性能的增量还是CPU个数的增量；像DeepSeek 的Engram， 将大量静态的嵌入表从HBM 转移到DRAM，对应的CPU的使用就会更多，这部分 可能 会导致对柜内CPU的性能要求越来越高，需要更高性能的CPU，但这个CPU是柜内/服务器内部的 CPU，不会对CPU的个数有增量需求； 2、目前对CPU增量需求最大的是Agentic AI，至于Agentic AI是如何需要CPU的，很多资料都讲过， 我们就不细讲了，结论就是Agentic AI需要的是柜外的CPU，这部分CPU就是增量的，不受GPU配比 关系的影响； CPU分类 由于很多读者并非半导体行业，对CPU的一些产业格局并不是很了解，所以这个章节我们讲一些偏 科普的东西，部分内容参考了长江和东海证券的报告。 首先按照指令集分类，CPU可以大致分为下面几类，其中X86和ARM是最为我们熟知的。其中，以 微 ...

Core Insights - The phenomenon known as "Claude-pilled" is rapidly spreading in Silicon Valley, indicating a significant shift in how software development is approached, with non-programmers now able to create applications using Claude AI [1][3][12] - The traditional programming paradigm is being challenged as AI tools like Claude Code allow users to generate complex software solutions without prior coding knowledge [12][14][46] Group 1: Impact on Software Development - Claude Code enables users to complete projects in a fraction of the time previously required, exemplified by a developer who finished a six-month app project in just one weekend [9][10] - The rise of "Vibe coding" allows individuals without coding backgrounds, such as stay-at-home parents and welders, to create functional applications simply by describing their needs in natural language [12][14] - The shift towards AI-driven development is leading to a decline in the traditional role of programmers, as the barriers to entry in software creation are being dismantled [12][46] Group 2: Changes in Workforce Dynamics - The use of AI tools is causing a divide among software engineers, with junior developers facing job market challenges as AI automates many of their tasks [40][41] - Conversely, senior engineers who can leverage AI effectively are becoming more valuable, transitioning from code writers to AI strategists and system architects [43][44] - The ability to understand and manage AI outputs is becoming crucial for job security in the evolving tech landscape [47][48] Group 3: Future of SaaS and AI Integration - The traditional SaaS model is being disrupted by AI Agents, which automate tasks through natural language processing, reducing the need for manual intervention [32][34] - Predictions indicate that by 2026, AI Agents will be integrated into nearly 80% of enterprise applications, fundamentally changing how software is developed and utilized [36] - The future of software engineering will likely see a shift towards AI-driven solutions, with traditional coding practices becoming less relevant [29][46]

" Manus最大的竞争对手，Genspark的登顶之路。 " 作者丨 齐铖湧 编辑丨 马晓宁 更明显的区别是，景鲲创立的MainFunc是一家扎根于硅谷的国际化团队，肖弘创立的蝴蝶效应，是一家 武汉原生的中国团队。 MainFunc 的融资节奏快得惊人。 MainFunc 的种子轮，于2024年中完成，估值约6000万美元，彼时MainFunc的核心业务，还只是在做AI 搜索。一年后，2025年2月MainFunc完成1亿美元A轮融资，估值跃升至5.3亿美元。 提起来Agent，最先被关注的肯定是被 Meta 收购的 Manus，而在硅谷投资圈内，另一家 Agent 公司 Genspark 同样备受关注，不仅得到了硅谷主流基金的青睐，有传闻称，Genspark 也同样进入到了硅谷 大厂的视野范围之内。 Genspark 与 Manus 同为华人创办，都是 AI Agent赛道的顶流公司，有着类似的产品定位和营销打法， 他们也是中国最新一代全球化AI创业公司的集中缩影。大众对于Manus的故事耳熟能详，我们也即将发布 《我所认识的肖弘》一文，关于Genspark的背景经历，相比较而言知之者要少得多。 01 硅 ...

Group 1: ETF Market Impact - The recent outflow from broad-based ETFs has reached approximately 600 billion yuan over the past two weeks, driven by an abnormal surge in trading volume [1] - Western Securities suggests that the core trigger for this significant ETF outflow is the abnormal trading volume, which peaked near 4 trillion yuan in a single day, likely attracting speculative funds [1] - The current ETF outflow does not indicate the end of the market; rather, a return to stable trading volume could lead to a sustainable slow bull market similar to 2020-2021, aligning with the preferences of bank wealth management and insurance funds for stable returns [1] Group 2: Dynamic Animation Market - The short drama market has approached a scale of nearly 100 billion yuan, demonstrating the commercial potential of fragmented entertainment, with dynamic animation (漫剧) leveraging AI technology to become a new growth driver [2] - By mid-2025, daily GMV for dynamic animation on Douyin is expected to exceed 10 million yuan, marking a year-on-year growth of 1532%, with daily advertising spending surpassing 4 million yuan, indicating a fully realized commercialization loop [2] - The dynamic animation industry is at a critical juncture, transitioning from "concept validation" to "scale performance release," with multi-modal technology enhancing visual expression and enabling "creativity as a product" [2]

[Table_Page] 行业专题研究|电子 2026 年 1 月 26 日 证券研究报告 [Table_Title] AI 的 Memory 时刻 5 | | | | AI NAND | 供需紧张，涨价仍有弹性 | | | --- | --- | --- | --- | --- | --- | | 分析师： [Tabl | 王亮 | 分析师： | 耿正 | 分析师： | 焦鼎 | | e_Author] | SAC 执证号：S0260519060001 | | SAC 执证号：S0260520090002 | | SAC 执证号：S0260522120003 | | | SFC CE.no: BFS478 | | | | | | | 021-38003658 | | 021-38003660 | | 021-38003658 | | | gfwangliang@gf.com.cn | | gengzheng@gf.com.cn | | jiaoding@gf.com.cn | | 分析师： | 张大伟 | | | | | | | SAC 执证号：S0260523050001 | | | | | | | 02 ...

Core Insights - SenseTime's AI Agent application "Kapi Accounting" has announced that its core accounting features will be permanently free, including 30 previously paid functionalities such as automatic bookkeeping, shared ledgers, budget planning, and multi-account management [1] Company Developments - The application leverages SenseTime's "Riri Xin" large model for multimodal understanding, enabling users to automatically recognize consumption information and categorize bookkeeping through methods like photographing receipts, voice input, or tapping on the phone [1] - Future plans include the launch of an "AI Pro" mode, which will provide users with personalized life planning and recommendations in addition to basic financial management [1] - Following this adjustment, users can access the application's bookkeeping features without any barriers [1]

3. 从"规模经济"到"效率经济" 当 Token 消耗增长 10 倍、100 倍时，推理服务成本不再是次要考量，如何能够必须实现"超卖"与"混 部"。考虑到实际上 Agent 需要使用 LLM 和多模态的不同模型，应对 Agent 的不同模型需求流量模 式呈现更强的潮汐效应，推理服务基础设施需要像"数字电网"一样动态调度算力。 AI Agent 对推理基础设施的 作者 | 章明星，清华大学副教授，Mooncake 社区联合发起人、 车漾，阿里云容器服务高级技术专家，Fluid 社区联合发起人 Token 消耗量的结构性转移正在重塑大模型推理服务基础设施的底层逻辑。一个不容忽视的事实是： AI Agent 正从人类手中接过 Token 消耗的指挥棒，背后是大模型从 Chatbot 转化为新质生产力 。 这不是量的变化，而是质的跃迁——推理基础设施的使用者正从"偶尔提问的人类用户"变为"7×24 小 时不间断工作的 Agent"，其单次任务需要几十次工具调用、输入输出比达到 10:1 甚至 100:1、面向 图像和全模态的输入导致上下文窗口常态性突破 100K，其请求模式、负载特征与成本考量正在发生 根本性的变 ...

Core Insights - The report from GF Securities highlights the increasing importance of AI memory as a foundational capability for context continuity, personalization, and historical information reuse, which is expected to accelerate the deployment of applications like AI Agents [1] Group 1: AI Agent and CPU Demand - The demand for CPUs is expected to increase due to AI Agents driven by three main factors: (1) Amplified application call volume as users can simultaneously engage multiple Agents, leading to higher system requests and increased CPU load [1] (2) Orchestration bottlenecks as the decision-making process involves multiple external tools, which increases CPU utilization [1] (3) Sandbox isolation raises overhead costs, necessitating additional CPU, memory, and storage configurations [1] Group 2: CPU Ratio Trends - According to a report by Semianalysis, the CPU ratio per GPU megawatt (MW) is currently below 10% but is projected to rise to 15% by Q2 2026, indicating a significant increase in CPU demand alongside GPU deployment [2] Group 3: Memory and Interface Chip Opportunities - The demand for memory modules is expected to grow as AI servers shift towards full insertion (2DPC) configurations to meet higher capacity and bandwidth requirements, leading to a doubling of DIMM demand relative to CPU growth [3] - The transition from traditional RDIMM to MRDIMM configurations will require more complex interface chips, which will drive up the average selling price (ASP) of supporting chips, thus expanding the market for memory modules and interface chips [3]

Summary of Key Points from Conference Call Records Industry Overview - The records focus on the **fiber optic** and **CPU** industries, highlighting trends in pricing and demand driven by technological advancements and market dynamics. Fiber Optic Industry Core Insights - **Increased Demand**: The demand for G657A single-mode fiber optic cables is expected to surge due to the rise of drones, projected to account for nearly **10%** of global fiber demand by Q2 2025 [1][2]. - **AI Influence**: The development of AI is significantly boosting the demand for fiber optics in data centers, with expectations that AI-related needs will represent **30%** of global fiber demand by **2027** [1][2]. - **Price Increases**: Prices for G657A and G652D single-mode fibers have risen sharply, with G657A increasing from over **30** to more than **55** yuan per kilometer, and G652D from over **20** to above **35** yuan per kilometer, reflecting a tighter market supply [1][2]. - **Supply-Demand Gap**: By **2027**, global fiber demand is projected to reach **800 million** core kilometers, while supply is expected to be around **810 million**, resulting in a **5%** supply-demand gap [1][5]. Additional Considerations - **Short-term Price Influences**: Short-term price fluctuations are influenced by the sustainability of drone demand and procurement strategies from telecom operators [7]. - **Telecom Procurement Impact**: The procurement strategies of major Chinese telecom operators like China Unicom and China Mobile significantly affect supplier profits and market dynamics [8][9]. CPU Industry Core Insights - **Market Changes**: The CPU market has seen notable price increases and secondary market fluctuations due to limited production capacity and competition from high-end GPUs and ASICs [10]. - **Extended Lead Times**: The lead time for CPU shipments has increased from the normal **8-10 weeks** to over **24 weeks** due to supply chain bottlenecks and material shortages [10]. - **Price Trends**: Major server CPUs from Intel and AMD have already seen price increases of several percentage points, with predictions of further increases of **10-20%** in **2026** [11][12]. AI Impact - **Rising Demand**: The advent of AI Agents is expected to significantly increase CPU demand as they require rapid and secure environments for executing AI code [13][19]. - **Market Growth Potential**: The market for AI Agents and their associated sandbox environments is projected to grow substantially, potentially increasing CPU demand by **hundreds of times** by **2028** [19]. Technological Advancements - **Micro-Virtual Machine Advantages**: Micro-virtual machines offer rapid startup times and high elasticity, enhancing CPU resource utilization and reducing costs [14][15]. - **Security Enhancements**: These technologies improve security by managing permissions dynamically and ensuring code integrity, making them suitable for high-value tasks [16]. Conclusion - The fiber optic and CPU industries are experiencing significant transformations driven by technological advancements, particularly in AI and drone applications. The ongoing supply-demand imbalances and price fluctuations present both challenges and opportunities for investors and industry players.