本文架构： 核心观点就是标题，李想对AI进步速度预期与 Anthropic CEO Dario Amodei很类似。 李想认为现在的人形机器人很像2025年2月的Manus，机器人速度可能比软件发展慢一点，但会超所有人预期。 Dario认为最快 1-2 年内就会有超强AI到 来，尽管也可能显著更久。 据TOP2观察，言语上倾向说理想首先应该把车造好的人，普遍对AI进步速度没有这个预期。李想现在重心完全放在具身智能上了，对造好具身智能的汽 车非常上心。 花长篇幅介绍为什么 Dario Amodei的观点值得深度参考，核心锚点主要是 Anthropic近期取得的成果水准，以及 Dario比Ilya早三年发现应该解除 Sam Altman的CEO职务。 TOP2并不是想表达，因为李想与 Dario 是这么想，他们的观点就一定是对的。TOP2认为不应该盲从李想/ Dario或任何人的观点，应该从第一性上充分批 判性思考，但他们的观点本身 值得深入参考。他们的观点是什么，为什么这么想都值得深入求证。 本文核心是指出李想的想法是什么，李想在AI进步速度的预期与AI核心领袖之一大方向很接近。这种接近不是轻飘飘的一句话，往小 ...

作为有70%投票权的创始人, 李想真打算全力做硅基司机/家政了, 不是demo，奔着量产去的。做硅基 生命要求的组织形式与车企完全不同了。 李想能不能做成是另外一回事，基于现阶段综合条件，没有主体能阻止李想将调集理想全部的资源做 这事。 李想已经把话说的非常清楚了。 站在2026年1月，不论是原有巨头还是全新AI公司，组织方式基本变成一样，各个团队整合在一 起进行联合设计，共同设计硅基生命的感知/大脑/心脏/神经/软件本体/硬件本体。和过去做软 件做硬件的方式发生了根本性变化，完全不同于互联网移动互联网时代。 理想要对技术和研发团 队进行大调整，把工作权限按照最先进的公司，按协作造硅基生命的方式进行工作。 硅基司机不是从A到B点，而是生活助理， 承担充电/接孩子/取快递/处理违章等事， 家政解决做 饭、洗衣、收纳等家务。 李想认为做家政的人形机器人的紧迫感强到 必须立刻启动，再不启动就会错过第一波最后的上车 机会。 空间机器人和人形机器人一些团队从水面下浮到水面上。去招聘最好的人。把原来由于没 有机器人，看到了很多机器人创业公司的一些人再招回来。 基于李想判断今天看机器人遇到的很多难题，非常像2025年2月7 ...

详见：《 新智驾独家｜理想调整研发架构，智驾部门将被重组 》 2026年1月李想对AI与Agent的理解是 据新智驾消息 理想汽车将研发体系重组为三大团队：基座模型团队、软件本体团队、硬件本体团队。 郎咸朋将成为硬件本体负责人，主要负责机器人研发，将向理想汽车总裁马东辉汇报，不再负责 自动驾驶。 自动驾驶团队将被并入到勾晓菲负责的软件本体团队。勾晓菲是理想智能空间副总裁。此次调整 后，他将统筹理想智能座舱和智能驾驶研发。勾晓菲也将向马东辉汇报。 此外，詹锟将领导基座模型团队，负责统筹 VLA 和自研芯片融合，詹锟向理想汽车 CTO 谢炎汇 报。 站在2026年1月，不论是原有巨头还是全新AI公司，组织方式基本变成一样，各个团队整合在一 起进行联合设计，共同设计硅基生命的感知/大脑/心脏/神经/软件本体/硬件本体。和过去做软 件做硬件的方式发生了根本性变化。完全不同于互联网移动互联网时代。 都会有Infra团队来确保足够的算力和数据稳定性，相当于硅基生命的云端心脏器；都有基座模型 团队，相当于硅基生命的大脑，都会去做视频音频多模态，做机器人的大脑。 都需要招大量的研究员，都以校招生为主，且发现本科生也行。都需要 ...

Core Viewpoint - AWS and Google Cloud have initiated price increases for their cloud services, indicating strong downstream demand for computing power, which may lead domestic cloud providers to follow suit [1][2]. Group 1: Price Increases - AWS has 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 [1]. - Google Cloud has announced a price adjustment for data transfer methods effective May 1, 2026, with North America seeing a price increase from $0.04 to $0.08 per GB, while Europe and Asia also experience increases [2]. Group 2: Demand for Computing Power - The price hikes are attributed to recent increases in CPU and storage costs, alongside a surge in computing power demand driven by complex task execution by agents [2]. - Clawdbot, an open-source project, has gained significant attention and is expected to further increase demand for computing power due to its ability to execute complex tasks continuously [3]. Group 3: Development of General-purpose Agents - Domestic model manufacturers are actively developing general-purpose agents, with Kimi releasing the K2.5 model that enhances capabilities in everyday office software [4]. - Alibaba's Qianwen app has integrated various services, evolving into an entry point for AI-driven tasks within the Alibaba ecosystem, reflecting a trend towards multi-agent collaboration [4]. Group 4: Investment Recommendations - Recommended stocks include Deepin Technology (300454.SZ) and other beneficiaries such as Parallel Technology (920493.BJ), Qcloud Technology-U (688316.SH), and others in the cloud computing sector [5].

Investment Rating - The industry investment rating is "Overweight" (maintained) [1] Core Insights - AWS and Google have initiated a price increase in cloud services, which is expected to be followed by domestic cloud vendors. This price adjustment reflects strong downstream demand for computing power due to rising costs of CPUs and storage, as well as increased computational needs from complex task execution by AI agents [4][7] - The open-source project Clawdbot has gained significant attention in the tech community, indicating a growing demand for general AI agents that can operate continuously and handle complex tasks, thereby increasing the need for computing power [5][7] - Domestic model vendors are actively developing general-purpose AI agents, which is anticipated to lead to breakthroughs in productivity tools and daily life assistance [6][7] Summary by Sections Cloud Pricing Trends - AWS has raised its EC2 machine learning capacity block prices by approximately 15%, with specific instance costs increasing from $34.61 to $39.80 per hour [4] - Google Cloud will adjust its data transmission prices starting May 1, 2026, with North America seeing a price increase from $0.04 to $0.08 per GB [4] AI Agent Development - Clawdbot, created by PSPDFKit founder Peter Steinberger, has received over 65,800 stars on GitHub since its launch in late 2025, showcasing its compatibility with multiple operating systems and productivity tools [5] - Kimi has released the K2.5 model, enhancing its agent capabilities for everyday office software, while Alibaba's Qianwen app has integrated various services to function as an AI shopping assistant [6] Investment Recommendations - The report suggests focusing on the AI infrastructure supply chain, with key recommendations including companies like Deepin Technology, Parallel Technology, Qingyun Technology, and others that are expected to benefit from the rising demand for computing power [7]

Core Viewpoint - The company is transitioning from a traditional automotive manufacturer to a focus on embodied intelligence, emphasizing the development of humanoid robots and general-purpose agents as part of its strategic upgrade [1][3] Group 1: Strategic Shift - The CEO outlined a strategic shift from "creating a mobile home" to focusing on embodied intelligence, with a core emphasis on silicon-based life forms [1] - The company plans to restructure its R&D teams into three main segments: foundational models, software entities, and hardware entities, integrating automotive and robotics under the hardware team [1] Group 2: R&D Focus - Since 2022, the company has been developing reasoning chips and operating systems, and it initiated foundational model research in 2023, aiming to enhance its self-research capabilities [1] - The company aims to deepen its self-research in core components like batteries and motors while building a general-purpose agent ecosystem for seamless integration between mobile applications and vehicle systems [1] Group 3: Industry Insights - The CEO identified four paths for general-purpose agent development, focusing on improving life scenarios through autonomous driving, humanoid robots, and home services [2] - The company plans to accelerate its humanoid robot business and will initiate specialized recruitment to attract core talent in the robotics field [2] Group 4: Market Positioning - The strategic upgrade signifies a formal transition from an automotive manufacturer to an embodied intelligence enterprise, leveraging its advantages in family user scenarios for autonomous driving and home robotics [3] - The effectiveness of organizational changes and technological collaboration will require time to validate, and the progress of this transformation will provide important insights for the competitive landscape of the electric vehicle industry [3]

站在2026年1月，不论是原有巨头还是全新AI公司，组织方式基本变成一样，各个团队整合在一起进 行联合设计，共同设计硅基生命的感知/大脑/心脏/神经/软件本体/硬件本体。和过去做软件做硬件的 方式发生了根本性变化。完全不同于互联网移动互联网时代。 都会有Infra团队来确保足够的算力和数据稳定性，相当于硅基生命的云端心脏器；都有基座模型团 队，相当于硅基生命的大脑，都会去做视频音频多模态，做机器人的大脑。 都需要招大量的研究员，都以校招生为主，且发现本科生也行。都需要做预训练与后训练。 在2025年下半年，逐渐都形成一个软件本体团队，组织中出现专门负责造手的团队，构建完整的工具 链和SaaS调用能力，让模型具备在数字世界的执行力。 免费提供数字世界所有使用的工具，免费提供一个完整的团队，如Lovart提供完整的设计团队，从想 法一直到最后发布的效果；Claude Code从需求分析到架构设计/前端/后端/部署/验证都做。 都要做所有的微调，context工程与记忆，从而把一个很好的脑子变成一个人，配合工具能够去大规模 生产。 都要评估模型的能力。 很多做应用做Agent的团队有两种错误的做后训练方式 1.类似 ...

Investment Rating - The report suggests that the general Agent will continue to be a competitive focus for major model manufacturers in 2026, indicating a positive investment outlook for Agent-related applications and ecosystem opportunities [5]. Core Insights - The competition in the general Agent field is intensifying among leading model manufacturers, with new iterations of Agent applications being released by companies like Anthropic, ByteDance, and Alibaba [3][4]. - The transition from ChatBot to Agent as the primary entry point for general model applications is underway, driven by advancements in core capabilities such as coding and Agentic functions [23]. - The report highlights that the next generation of Agent products will likely follow the aggressive path set by Anthropic's Claude Cowork, which integrates user file permissions for more direct interaction [12][13]. Summary by Sections Industry Dynamics - In early 2026, major players like Anthropic, ByteDance, and Alibaba have launched new iterations of general Agent applications, focusing on desktop environments as the new product definition [3]. - Anthropic's Claude Cowork, released on January 12, 2026, is designed as a desktop work Agent for non-programmers, allowing for direct file operations on users' computers [4][8]. Product Developments - Claude Cowork enhances user interaction by allowing direct access to local files, enabling tasks such as reading, writing, and organizing files without manual uploads [9]. - Domestic companies are also iterating their Agent products, with Alibaba's Qianwen and ByteDance's Coze 2.0 updates enhancing their respective ecosystems and productivity platforms [13]. Valuation and Recommendations - The report emphasizes the importance of monitoring the evolution of Agent products and the related application ecosystem, as well as the increasing computational demands associated with these advancements [5][25]. - The anticipated growth in the Agent sector is expected to create significant investment opportunities, particularly in vertical industries with accumulated data and customer bases [25].

Claude Cowork 正式发布，CPU 需求加速增长 [Table_Industry] 电子元器件 | | | [Table_Invest] 评级： 增持 | [姓名table_Authors] | 电话 | 邮箱 | 登记编号 | | --- | --- | --- | --- | | 舒迪(分析师) | 021-38676666 | shudi@gtht.com | S0880521070002 | | 段笑南(研究助理) | 021-38031382 | duanxiaonan@gtht.com | S0880124070028 | 本报告导读： Claude Cowork 正式发布，通用 Agent 加速落地。Agent 任务中 CPU 需承担更多工 作负荷，CPU 需求或加速增长。 投资要点： [Table_Report] 相关报告 行业跟踪报告 表1：推荐标的估值表（收盘价参考 2026 年 1 月 21 日） | 证券代码 | 公司名称 | 收盘价 | EPS（元/股） | | | PE | | | 评级 | | --- | --- | --- | --- | --- | --- | -- ...

Group 1 - The core viewpoint of the news highlights the significant growth and investment in AI technologies, particularly through the performance of AI ETFs and advancements in AI models [1][2] - As of January 22, 2026, the AI ETF has surpassed 10 billion yuan in size, marking a record high since its inception, with a net inflow of 1.1 billion yuan over four out of the last five trading days [1] - Baidu's "Wenxin Moment" conference showcased the official release of the Wenxin model 5.0, which features 2.4 trillion parameters and supports various forms of information processing, ranking first domestically and eighth globally in text processing [1] Group 2 - Major tech companies are increasing their investments in AI infrastructure, with projected capital expenditures reaching 596.4 billion dollars in 2026, reflecting a 47% year-on-year growth [1] - The focus of competition among AI model manufacturers is shifting towards general-purpose agents capable of autonomous planning and execution, with several companies launching new products in this space [2] - The Huafu AI ETF (515980) offers investors a strategic tool to capture long-term growth opportunities in the AI sector, with a unique index that adjusts quarterly to reflect the latest industry trends [2]