Agent 趋势乘风起，关注 CPU 产业机遇 电子 证券研究报告/行业点评报告 2026 年 01 月 28 日 | 分析师：王芳 | | --- | | 执业证书编号：S0740521120002 | | Email：wangfang02@zts.com.cn | 2026-01-05 沿主线，买缺口》2025-12-10 3、《【中泰电子】4F2+CBA 是国产 颈》2025-12-01 | 增持（维持） 评级： | | 重点公司基本状况 | | | | | | | | | | | | | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | | | 简称 | 股价 | | | EPS | | | | | PE | | | 评级 | | 分析师：王芳 | | （元） | 2023A | 2024A | 2025E | 2026E | 2027E | 2023A | 2024A | 2025E | 2026E | 2027E | | | 执业证书编号：S0740521120002 | 海光信息 ...

Core Insights - The competition for AI entry points among major internet giants like Alibaba, ByteDance, and Tencent is intensifying, with each company adopting distinct strategies [2][10] - Alibaba has launched its flagship reasoning model Qwen3-Max-Thinking, which significantly enhances the native agent capabilities of its Qwen app, aiming to become a leading AI assistant [1][3] - ByteDance has introduced the Doubao Model 1.8, which focuses on multi-modal understanding and improved agent capabilities, enhancing its applications in various scenarios [5][6] - Tencent is taking a decentralized approach, emphasizing user-defined AI interactions and prioritizing enterprise productivity over consumer-focused AI entry points [7][8] Alibaba's Strategy - Alibaba's Qwen3-Max-Thinking model boasts over one trillion parameters and has achieved state-of-the-art performance in 19 benchmark tests, showcasing its advanced reasoning and agent capabilities [3][4] - The Qwen app has integrated over 400 AI functionalities, making it the first AI assistant capable of handling complex real-life tasks [4] - The model's ability to think and use tools like a professional enhances user experience, allowing for more intelligent and fluid interactions [3][4] ByteDance's Approach - ByteDance's Doubao Model 1.8 is optimized for multi-modal agent scenarios, improving its ability to follow complex instructions and execute tasks effectively [5][6] - The collaboration with ZTE on the Nubia M153 AI phone aims to validate the feasibility of "end-side large models + system-level agents" [6] Tencent's Position - Tencent's strategy focuses on a decentralized AI ecosystem, allowing users to define when and how to engage with AI, which contrasts with the centralized approaches of Alibaba and ByteDance [7][8] - Tencent's AI applications, such as the "Yuanbao" social AI feature, are designed to enhance user engagement while maintaining a focus on enterprise solutions [8][9] Industry Trends - The evolution of agent capabilities is seen as a critical factor in the AI landscape, with potential to create new application scenarios and software forms [9][10] - The competition is shifting from a focus on technology alone to a differentiation in operational models and compliance with safety standards [10]

Core Insights - The transition from dialogue models to action-oriented agents is reshaping the division of labor between CPUs and GPUs, with a structural differentiation in computing power demand as AI moves towards agent-based applications [1][2] - The execution control flow is becoming CPU-centric, as agent actions involve numerous conditional statements and system calls, making CPUs more suitable for branching tasks compared to GPUs [1][2] - The memory architecture is shifting away from GPUs, as long-context reasoning in agent scenarios generates large KV caches that quickly exhaust GPU memory, while CPUs with large memory capacities are becoming the preferred architecture [1][3] Industry Developments - Major cloud service providers (CSPs) like AWS and Google Cloud are accelerating the development of sandbox environments for agents, enhancing software-level isolation and orchestration capabilities [3] - CPU manufacturers are evolving towards ultra-multicore architectures driven by agent demands, with AMD's Turin reaching up to 192 cores and Intel's Sierra Forest potentially offering up to 288 cores [3] - NVIDIA is increasing the importance of CPUs in its new architectures, indicating a recognition of the need for large-memory CPUs to support extensive KV caches in high-concurrency agent scenarios [3] Investment Recommendations - The company is optimistic about HaiGuang Information and Lanqi Technology, as the former benefits from the elevation of CPU's role in high-concurrency execution driven by agentic AI, while the latter benefits from the synergy of memory and computing power in long-context scenarios [4] - Attention is also recommended for Longxin Technology and Wantong Development (Shudu Technology) due to their potential in this evolving landscape [4]

Market Overview - On January 28, the cloud computing sector experienced a significant surge, with Meili Cloud hitting the daily limit, and companies like Wangsu Science and Technology and Oulutong rising over 10% [1] - Meili Cloud's stock price reached 14.28, reflecting a 10.02% increase, while Wangsu Science and Technology saw a price of 16.04, up by 18.46% [2] Price Increases by Major Cloud Providers - Google Cloud announced on January 27 that it will increase global data transfer service prices starting May 2026, with North American rates expected to double from current levels [3] - Amazon Web Services raised its EC2 machine learning capacity block prices by approximately 15%, with the p5e.48xlarge instance's hourly cost increasing from $34.61 to $39.80 [3] Institutional Insights - The recent price hikes confirm the high demand for AI computing power globally, indicating a growing scarcity of resources in the AI cloud industry [4] - The price increases mark a departure from the long-standing trend of declining cloud service prices, suggesting that once a price increase is successfully implemented without significant customer loss, further increases may follow more easily [4] - The global cloud computing market is projected to reach $692.9 billion in 2024, with a year-on-year growth of 20.3%, driven by the demand for IaaS due to AI model training [4] - The emergence of Clawdbot signifies a shift in AI product forms from "scene-level assistants" to "system-level agent platforms," enhancing overall efficiency in various business operations [4]

Investment Rating - The report rates the computer industry as "Overweight," indicating that the industry is expected to outperform the overall market [9]. Core Insights - Clawdbot signifies a transition in AI product forms from "scene-level assistants" to "system-level Agent platforms," emphasizing comprehensive execution capabilities across tools and systems [3]. - The evolution of Agents is marked by a clear path from Skills to Claude Excel and Clawdbot, shifting market focus from model parameters to the ability to execute complex tasks and integrate external tools [3]. - Clawdbot's revolutionary aspect lies in its ability to understand high-level user intentions and autonomously break them down into ordered steps, enhancing automation in complex workflows [3]. - The deployment of system-level Agents faces challenges such as high complexity in enterprise deployment, data security concerns, and the need for improved stability and traceability across systems [3]. - The acceleration of Agent deployment does not imply the replacement of traditional software but rather positions Agents as intelligent interaction and automation hubs, enhancing overall efficiency [3]. Summary by Sections AI Application Key Companies Valuation - The report includes a valuation table for key AI application companies, detailing their market capitalization and projected net profits for 2024 to 2027, with notable companies such as Kingsoft Office, Yonyou Network, and Zhongkong Technology [4].

让AI像人类一样记忆， 这家公司如何拿下AI竞赛的下半场门票。 前不久， LMArena.ai 对全球大模型的市场地位变化做了统计后，得到了一个有意思的发现： 自 2023 年年中起， SOTA 模型 的迭代周期被 快速 压缩至 35 天， 曾经的 SOTA 模型，只要 短短 5 个月就可能跌出 Top5 ， 7 个月后连 Top10 的 门槛都摸不到。 但 SOTA 不断更新的背后，模型的确在进步，但曾经 ChatGPT 、 Deepseek 这样让人眼前一亮的新产品却越来越少，技术进步已经进入了不断小修小补 却始终难以突破的瓶颈期。 与逐渐偃旗息鼓的模型进化形成鲜明对比的，是过去两年多围绕 AI 记忆形成的你方唱罢我登场的热闹。 其中，最先一步出发的，是 2023 年先后涌现出的诸如 Milvus 、 Pinecone 、 faiss 为代表的向量数据库产品。 此后一年，建立在成熟的语义、知识图库以及关键词检索基础上， 2024 — 2025 年期间， Letta （ MemGPT ）、 Mem0 、 MemU 、 MemOS 为代 表的各种 AI 记忆框架，如雨后春笋般冒出， GitHub 上各种 Me ...

匿名潜伏在 OiiOii 的用户群里，闹闹看到了不少和她当年一样的人，热爱ACG，等一个合适的机会展示自己的热情与天赋：一个性格内敛的北漂女孩 用动画拍出了不敢出镜的 Vlog；一位小学老师因为做出了让学生惊叹的课件而找回了成就感；锦秋基金创始合伙人杨洁自己，也用 OiiOii 为转学的女 儿重构了一个"太空学院"的奇遇。 在闹闹看来，这不仅是工具的迭代，更是产业逻辑的重组。过去，动画是集体的重资产消耗；而未来，当内容生产的边际成本趋近于零，行业将进 入"意图主导"的 时代。 在北京那些被数据和业务填满的深夜里，闹闹偶尔会被认为是一个"潜伏者"。 过 去的十年里，她的名字出现在腾讯微信、字节跳动、B站、阶跃星辰这些互联网巨头的名人名单里。 作为一名资深产品经理，她不仅擅长用理性的逻 辑去拆解用户增长，也在长期的产品实战中，建立起一种对用户需求极其敏锐的职业直觉。 但在这些职业标签之 下，其实还一直藏着一个关于动画的人生伏笔。 十年前，闹闹曾试图入行。她辞职去学动画，却发现那是一座由复杂的软件操作、昂贵的渲染成本和冗长的工业管线筑起的围墙。那时，一个缺乏资源 支撑的个体创作者，很难获得表达的机会。此后，即便身处 ...

Core Viewpoint - Tencent's AI assistant, Yuanbao, has launched a new social AI feature called "Yuanbao Club," aiming to integrate AI technology into multi-user social scenarios, enhancing group interaction and entertainment [6][8][10]. Group 1: Features of Yuanbao Club - Users can create or join a "club," where they can interact with Yuanbao AI to summarize chats and organize activities like fitness and reading [1]. - The platform allows for creative content generation, enabling users to transform ordinary photos into memes or stickers, fostering collaborative fun [1]. - Yuanbao Club will support features like "watch together" and "listen together," leveraging Tencent Meeting's audio-visual capabilities for synchronized media experiences [6][9]. Group 2: Strategic Implications - This initiative marks a shift in Tencent's AI application strategy from enhancing individual productivity to fostering group interactions, providing a new approach to AI integration in daily life [8][9]. - The company is positioning itself to explore AI's role beyond traditional chatbot functionalities, aiming to create a more natural and engaging user experience [9][10]. - Tencent has committed to investing significantly in this initiative, as evidenced by the distribution of 1 billion cash red envelopes within the Yuanbao app, highlighting its determination to accelerate in the AI sector [9].

Summary of Conference Call Notes Industry and Company Involved - The discussion primarily revolves around the **CPU** industry, particularly focusing on **x86** and **ARM** architectures, with mentions of major players like **Intel**, **AMD**, and cloud service providers such as **Alibaba Cloud** and **Tencent Cloud**. [1][2][4][24] Core Points and Arguments 1. **Demand Growth for CPUs**: The demand for CPUs is expected to grow exponentially, driven by three linear factors rather than a simple linear increase. High-performance x86 and ARM architectures will benefit from this trend. [1][21] 2. **Impact of AI and Agents**: The rise of AI agents is significantly driving CPU demand. The increase in agent numbers and their complexity is leading to higher CPU utilization, as evidenced by various workloads where CPU tasks dominate processing time. [4][6][13] 3. **Price Increases**: CPU prices are anticipated to rise steadily over the next year, although not as dramatically as storage prices. This increase is viewed as a healthy adjustment for the industry. [14][21] 4. **Supply Chain Constraints**: The supply of CPUs is constrained by production capacity issues, particularly for advanced nodes (3nm and 5nm) at foundries like TSMC. This is leading to a competitive environment for resources between CPUs and GPUs. [20][21] 5. **Cloud Service Pricing**: The rising costs of CPUs and storage are expected to be passed on to cloud service providers, leading to potential price increases for services offered by companies like Alibaba and Tencent. [22][24] 6. **Server Manufacturers**: Companies involved in server manufacturing, such as Inspur and Huqian, are expected to experience increased demand and flexibility due to the rising CPU prices. [2][25] 7. **Investment Opportunities**: Key investment opportunities are identified in domestic companies like **Haiguang Information** and **Zhongke Shuguang**, as well as international firms like **AMD** and **Intel**. [24][25] Other Important but Possibly Overlooked Content 1. **Agent Workloads**: Detailed analysis of agent workloads shows that CPU tasks can account for up to 90% of processing delays, indicating a critical need for CPU resources in AI applications. [7][11] 2. **Batch Size and Performance**: Increasing batch sizes in processing tasks can lead to diminishing returns in throughput and increased CPU context-switching bottlenecks, highlighting the need for efficient CPU management in AI workloads. [10][11] 3. **Long-term Trends**: The transition to a new paradigm in AI processing, where CPU tasks are more prominent due to the nature of agent interactions, suggests a long-term shift in resource allocation from GPUs to CPUs. [15][19] 4. **Market Dynamics**: The competitive landscape is shifting, with domestic firms potentially expanding into markets previously dominated by larger players, indicating a changing dynamic in the CPU market. [23][24] This summary encapsulates the key insights and projections discussed during the conference call, emphasizing the evolving landscape of the CPU industry and its implications for investment and market strategies.

Summary of Conference Call Notes Industry and Company Involved - The discussion revolves around the demand for CPUs driven by the increasing number of Agents in AI systems, focusing on the implications for CPU usage and performance in AI applications. Core Points and Arguments - **Increased Demand for CPUs**: The rise in the number of Agents significantly increases the demand for CPUs, as each Agent requires substantial computational resources for data processing and logical scheduling [1][4] - **Virtual Machine Technology Changes**: Current AI clusters emphasize hardware resource binding, requiring virtual machines to start within 1 second and maintain a resident state, which escalates the need for high-performance CPUs [1][5] - **CPU Load Factors**: The core factors affecting CPU load include the duration and frequency of tasks. Long-duration tasks (2-4 hours) have a more significant impact on CPU load compared to short, frequent tasks [1][6] - **Memory Management Needs**: The development of large models necessitates more CPUs for memory scheduling, particularly with DRAM and SSD storage, which involves complex data communication [2][15] - **Agent Task Complexity**: AG tasks impose a heavy load on CPUs, with token consumption during processing being significantly higher than user input, leading to increased computational demands [1][11] - **Future CPU Usage Growth**: CPU usage growth is expected to be between linear and exponential, potentially doubling or quadrupling in the next few years, depending on the complexity of long-term tasks [2][12] - **Deepseek and Anagram Technologies**: These technologies enhance computational efficiency by offloading some calculations to CPUs, reducing GPU burden and improving query efficiency [1][10] - **CPU vs. GPU**: While CPUs can support smaller language models, GPUs remain essential for complex tasks in AI servers, indicating that CPUs are not a complete substitute for GPUs in high-demand scenarios [2][12][18] - **Agent Support by CPU Cores**: A single CPU core can support 2-5 Agents, but this number decreases for complex tasks, highlighting the need for more cores to handle increased workloads [2][13] - **Market Supply and Alternatives**: Despite the tight supply of CPUs, established vendors like Intel and AMD maintain a competitive edge due to their stable ecosystems, while newer architectures are still in development [2][22] Other Important but Potentially Overlooked Content - **Impact of High Concurrency**: In high-concurrency situations, even optimized simple tasks can place significant demands on CPUs, especially during peak usage times [2][19] - **Challenges in Performance Optimization**: As user scale increases, the effectiveness of CPU performance optimizations may diminish, with potential performance gains dropping during peak usage [2][20] - **General Computing vs. AI Servers**: General computing servers focus on storage integration, while AI servers prioritize GPU capabilities, indicating a divergence in design and application [2][21] - **Future Trends in General Computing Servers**: The maturity of general computing servers suggests a continued reliance on established platforms like Intel and AMD, particularly in cloud technology [2][23]