【中信科技产业】海外AI叙事或重回乐观情形，重视海外算力链新一轮上涨机遇！ 核心判断 近期海外推理与训练算力需求同步走强，亚马逊云、谷歌云相继涨价，台积电上修Capex。 尽管当前AI应用大规模商业化能见度仍有限，但在模型与应用密集催化下，未来3–6个月海外算力需求有望进一步 上行，算力"泡沫论"担忧或阶段性缓解，产业链有望迎来新一轮修复。 推理侧：Agent落地 【中信科技产业】海外AI叙事或重回乐观情形，重视海外算力链新一轮上涨机遇！ 核心判断 近期海外推理与训练算力需求同步走强，亚马逊云、谷歌云相继涨价，台积电上修Capex。 尽管当前AI应用大规模商业化能见度仍有限，但在模型与应用密集催化下，未来3–6个月海外算力需求有望进一步 上行，算力"泡沫论"担忧或阶段性缓解，产业链有望迎来新一轮修复。 推理侧：Agent落地抬升推理算力消耗 MoltBot等新一代Agent显著提升对电脑操作与复杂任务的处理能力，带来更高推理算力消耗。 Anthropic持续推出Claude Code、Agent Skills等产品，拓展Agent应用场景。 OpenRouter数据显示，2026年1月初以来Token调用量持 ...

Core Viewpoint - Recent demand for inference and training computing power is strong, leading to price increases from both Amazon Web Services (AWS) and Google Cloud [1][2] Group 1: Demand for Computing Power - The demand for computing resources for inference and training has significantly increased, supporting the need for training computing power [1] - AWS raised prices by approximately 15% for EC2 machine learning capacity blocks on January 23, 2026, followed by Google Cloud announcing price hikes for its cloud network transmission services, with North America seeing a doubling of prices [2] Group 2: Inference Side - The rapid emergence of AI agents, such as MoltBot, is expected to support the demand for inference computing power, as these agents can perform more complex tasks and require more computing resources [3] - Anthropic has raised its revenue expectations for 2026 and 2027 to $18 billion and $55 billion, respectively, indicating a strong upward trend in inference demand [3] Group 3: Training Side - The industrial sector is continuously exploring scaling limits, which supports the demand for training computing power, with new models expected to be released in Q1 2026 [8] - Models like Grok-5 and GPT-6 are anticipated to utilize larger datasets and parameter scales, increasing the demand for training computing power [8] Group 4: Financial Reporting Catalyst - The upcoming earnings reports from major cloud service providers (CSPs) will be critical in confirming the demand for computing power and capital expenditure continuity [9] - Key earnings dates include Microsoft and Meta on January 29, 2026, followed by Google on February 5 and Amazon on February 6, with NVIDIA's report on February 26 expected to further influence market sentiment [9][10]

Core Insights - Claude Skills have gained significant popularity, surpassing MCP, with various CLI clients supporting them and products like扣子 and MiniMax launching similar offerings [2] - The value of Claude Skills is still greatly underestimated, as a well-designed Skill can match or exceed the intelligence of complete AI products [3][4] - Skills allow non-technical users to create their own applications, demonstrating the potential for widespread adoption and innovation in AI [4][5] Group 1: Understanding Skills - Skills are modular capabilities that extend the functionality of AI Agents, packaging instructions, metadata, and optional resources for automatic use [22][24] - The architecture of Skills resembles a comprehensive SOP package for task delegation, enabling Agents to execute tasks reliably [24][28] - Skills can be seen as an extension of general Agents, allowing them to perform specialized tasks by loading different Skills [14][16] Group 2: Value and Future of Skills - Skills represent a future-oriented approach to AI applications, enabling Agents to autonomously understand and execute tasks based on human-provided guidelines [28][30] - The design of Skills allows for zero-code development, making it accessible for non-technical users to create intelligent Agents [31][34] - Skills can be combined flexibly, allowing multiple Skills to work together to complete complex tasks [68][69] Group 3: Implementation and Usage - The article provides a comprehensive guide on how to create and use Skills, emphasizing the ease of development and the potential for rapid testing of ideas [105][130] - Users can install Skills from repositories or create their own using tools like skill-creator, which simplifies the development process [121][130] - The article outlines scenarios where Skills are particularly beneficial, such as when specific knowledge or templates are required for task completion [139][141]

此外，扣子官方视频Skill也正式上线，该功能支持自动生成视频脚本、匹配视觉素材，并完成剪辑、转 场、配乐等后续流程。 扣子2.0推出Agent Plan，即"长期计划"，让AI从"即时问答工具"升级为"可持续运作的智能体"。用户只 需要确定目标，规定好怎么完成、怎么实现，扣子能够持续执行，并向用户主动汇报、最终交付任务。 以自媒体账号运营为例，扣子能够和用户一起讨论账号定位，拆解每个阶段的运营策略，再帮用户创作 内容。如果用户想写一本书，只需把写作主题和目标发给扣子，它会自己搜集资料、撰写初稿，再根据 反馈自行调整，达成3周内写出10万字初稿的计划。用户还可以用长期计划来完成学习目标，比如考雅 思，扣子会设定每日任务并准备学习资料。 （文章来源：科技日报） 1月19日，字节跳动旗下AI Agent（智能体）平台"扣子"宣布2.0品牌升级。扣子2.0集成了Agent Skill、 Agent Plan、Agent Coding、Agent Office能力，让AI真正成为用户的"工作伙伴"。 通用Agent已展现出强大的基础任务处理能力，在日常对话、信息整合与简单推理方面有了长足进步， 但要应对高度专业化、高 ...

Core Insights - ByteDance's AI Agent platform "Kouzi" has announced a brand upgrade to version 2.0, integrating capabilities such as Agent Skill, Agent Plan, Agent Coding, and Agent Office, positioning AI as a true "work partner" for users [1][3] Group 1: Agent Skills - The new Agent Skills feature encapsulates domain knowledge and standardizes operational processes, combining general AI cognitive abilities with specific task requirements to meet diverse and high-standard application scenarios, ensuring output stability [1][3] - Agent Skills are essentially a combination of "best practices for scenarios + required tools," aimed at helping users leverage professional skills to enhance their ability to solve complex professional problems [1][3] Group 2: Agent Plan - The introduction of Agent Plan allows AI to evolve from an "instant Q&A tool" to a "sustainable operating agent," where users can set goals and specify how to achieve them, enabling continuous execution and proactive reporting until task completion [2][4] - This long-term planning feature facilitates closed-loop management of complex goals, breaking down tasks that may take hours or days into manageable steps while tracking progress and managing intermediate states [2][4] Group 3: Agent Office - Agent Office is designed to deeply understand workplace scenarios, allowing users to delegate tasks such as writing strategic reports in Word, creating analysis presentations in PPT, and organizing data in Excel to the AI [2][4] Group 4: Vibe Coding Platform - The Vibe Coding platform offers a one-stop cloud-based solution, enabling users to easily build agents, workflows, websites, and mobile applications through continuous dialogue, with core functionalities available out of the box [5] - The platform provides Vibe Infra infrastructure for one-click deployment, allowing agents to write prompts, integrate knowledge bases, and develop tools, with the capability for self-iteration during multi-turn conversations [5]

Core Insights - The article discusses a new AI feature called "Agent Skills" developed by Anthropic, which allows AI to learn and perform various tasks more efficiently than traditional prompt-based methods [2][4][24] - This feature is seen as a significant advancement in AI capabilities, enabling users to create and share skills that the AI can utilize without the need for extensive prompts [15][23] Group 1: Introduction of Agent Skills - "Agent Skills" is a new project that has gained significant attention in AI communities, with claims that it is more user-friendly than traditional prompt writing [2] - The feature allows AI to learn new skills, similar to how Pokémon learn abilities, enhancing its functionality [7][8] Group 2: Functionality and User Experience - Users can enable various built-in skills for tasks such as document processing and web design, allowing for direct commands to create outputs like PowerPoint presentations [8][18] - The AI can assist with coding tasks by generating documentation based on provided code snippets, streamlining the coding process [13] Group 3: Custom Skill Creation - Users can create custom skills using the "Skill Creator," which guides them through the process of defining their needs, making it more accessible than traditional prompt writing [14][15] - Skills can be packaged and shared easily, allowing users to benefit from community-shared skill sets [15][24] Group 4: Technical Mechanism - The underlying architecture of Skills is modular, enabling the AI to determine which skill to use based on the task at hand, rather than relying on a fixed algorithm [23] - This approach allows the AI to "discover and load on demand," making it more efficient in task execution [23] Group 5: Comparison with MCP - The article clarifies that while "Agent Skills" and the previously introduced MCP (Multi-Channel Protocol) both enhance AI functionality, they serve different purposes; MCP focuses on data access, while Skills focus on task execution [24] - The introduction of Skills is expected to set a new trend in AI development, similar to the impact of MCP [24]

Group 1: Nvidia and TSMC Collaboration - Nvidia and TSMC unveiled the first Blackwell chip wafer produced in the U.S., marking a significant milestone in domestic chip manufacturing [1] - The TSMC Arizona factory has a total investment of $165 billion and will produce advanced chips using 2nm, 3nm, and 4nm processes [1] - The Blackwell chip features 208 billion transistors and achieves a connection speed of 10TB/s between its two sub-chips through NV-HBI [1] Group 2: Anthropic's Agent Skills - Anthropic launched the Agent Skills feature, allowing users to load prompts and code packages as needed, enhancing the capabilities of AI [2] - Skills can be used across Claude apps, Claude Code, and API platforms, with a focus on minimal necessary information loading [2] - The official presets include nine skills for various document formats, and users can upload custom skills [2] Group 3: New 3D World Model by Fei-Fei Li - Fei-Fei Li's World Labs introduced a real-time generative world model, RTFM, which can render persistent 3D worlds using a single H100 GPU [3] - RTFM employs a self-regressive diffusion Transformer architecture to learn from large-scale video data without explicit 3D representations [3] - The model maintains spatial memory for persistent world geometry through pose-aware frames and context scheduling technology [3] Group 4: Manus 1.5 Update - Manus released version 1.5, introducing a built-in browser that allows AI to interact with web pages, test functions, and fix bugs [4] - A new Library file management system enables collaborative editing within the same Agent session, reducing average task completion time significantly [4] - The system allows for no-code music web application construction through natural language, supporting real-time updates [4] Group 5: Windows 11 Major Update - Windows 11's major update features "Hey Copilot" for voice activation and Copilot Vision for screen understanding, enhancing user interaction [5][6] - Copilot Actions can perform operations on local files, while Copilot Connectors integrate with OneDrive, Outlook, and Google services [5][6] - Manus AI operations are integrated into the file explorer, allowing for automatic website generation and video editing functionalities [6] Group 6: Baidu's PaddleOCR-VL Model - Baidu open-sourced the PaddleOCR-VL model, achieving a score of 92.6 on the OmniDocBench V1.5 leaderboard with only 0.9 billion parameters [7] - The model supports 109 languages and excels in text recognition, formula recognition, table understanding, and reading order prediction [7] - It utilizes a two-stage architecture combining dynamic resolution visual encoding and a language model, achieving high inference speed on A100 [7] Group 7: AI in Fusion Energy Development - Google DeepMind collaborates with CFS to accelerate the development of the SPARC fusion device using AI [8] - The partnership focuses on creating precise plasma simulation systems and optimizing fusion energy output [8] - The TORAX simulator is a key tool for CFS, enabling extensive virtual experiments and real-time control strategy exploration [8] Group 8: Harvard Study on AI's Impact on Employment - A Harvard study tracking 62 million workers found a significant decline in entry-level positions in companies using AI, primarily through slowed hiring [9] - The impact of AI is most pronounced among graduates from mid-tier universities, while top-tier and bottom-tier institutions are less affected [9] - The wholesale and retail sectors face the highest risk for entry-level jobs, with a trend towards skill polarization [9] Group 9: Concerns Over AI-Generated Content - Reddit co-founder Ohanian warned that much of the internet is "dead," overwhelmed by AI-generated content [10] - Reports indicate that automated traffic could reach 51% by 2024, with AI-generated articles surpassing human-written ones [10] - Research suggests that training models on AI-generated data may lead to a decline in model performance [10] Group 10: Andrej Karpathy on AGI Development - AI expert Andrej Karpathy expressed skepticism about the current state of AI agents, predicting that AGI is still a decade away [11] - He criticized the noise in reinforcement learning and the limitations of pre-training methods [11] - Karpathy anticipates that AGI will contribute modestly to GDP growth, emphasizing the importance of education in the AI era [11]

Product Update - Anthropic Engineering Blog 发布了关于 Agent Skills 的开发者使用技巧，Agent Skills 是一种通过指令文件夹、脚本和资源扩展 Claude 功能的新方法 [1]