Core Insights - The report from Caitong Securities highlights the potential growth in CPU demand driven by the ongoing development of AI Agents and the implementation of sandbox technology to mitigate associated risks [1][2]. Group 1: AI Agent Sandbox Development - Both domestic and international markets are gradually deploying AI Agent sandboxing technology. Notably, Meta's acquisition of Manus for over $2 billion by December 2025 is expected to accelerate the promotion and application of sandbox technology [1]. - Major domestic cloud platforms, such as Alibaba Cloud, are launching and iterating AI Agent Infra products centered around AI Agent Sandbox, indicating a gradual implementation of sandboxing in AI Agent deployment [1]. Group 2: CPU Demand and Risk Control - The sandbox isolation technology is crucial for controlling potential risks associated with AI Agents, which in turn creates additional demand for CPUs. The action module of AI Agents translates abstract instructions into specific operations, heavily relying on the richness and reliability of the tools available [2]. - The core technology enabling this tool usage is function calling, which allows large language models (LLMs) to output structured JSON objects alongside text generation. This separation of understanding and execution is vital for AI Agent functionality [2]. Group 3: Investment Recommendations - Investment focus should be on overseas CPU manufacturers such as AMD and Intel, as well as domestic CPU alternatives like Haiguang Information and Loongson Technology. Additionally, supply chains related to AMD (Tongfu Microelectronics, Aoshikang, Shiyun Circuit) and Intel (Lanke Technology, Shiyun Circuit, Xingsen Technology) are recommended for attention [3].

Core Insights - The AI large model sector is expected to undergo deeper integration and innovation in the coming year, transitioning from a phase of "technical frenzy" to "value realization" [1][24]. Group 1: Market Dynamics and Competition - Alibaba's AI model, Qianwen, has integrated with various platforms like Taobao and Alipay, creating a closed-loop consumer ecosystem [2]. - The competitive landscape has shifted from nearly 300 companies to a concentrated market dominated by major players like ByteDance, Alibaba, and Baidu, driven by resource and capital advantages [4][10]. - ByteDance's strategy of "stronger models + lower prices + easier implementation" has allowed it to rapidly capture market share, with its Doubao model achieving over 150 million monthly active users [5]. - Tencent's Yuanbao has heavily invested in advertising, with expenditures reaching 1 billion yuan in June and July 2025, indicating significant efforts to promote AI products [6]. - Baidu and Alibaba are enhancing their core competencies through technological iterations and open-source strategies, with Baidu's Wenxin model outperforming competitors in multi-modal capabilities [7][8]. Group 2: Industry Trends and Challenges - The large model market in China is projected to grow significantly, with a market size of 294.26 billion yuan in 2024 and expected to exceed 700 billion yuan by 2026, reflecting a shift towards commercialized applications [13]. - B-end applications face challenges in reliability and standardization, while C-end applications are diversifying, with major companies integrating AI into various daily life scenarios [14][15]. - The competition in the C-end market is intensifying, requiring companies to invest in continuous iteration and resource allocation to meet diverse user needs [19][23]. - The rise of AI agents is transforming large models from mere conversational tools to actionable partners, with a focus on creating measurable commercial value in vertical industries [21][22]. Group 3: Future Outlook - The AI large model sector is anticipated to see further integration and innovation, with a focus on the collaboration between general and specialized models, enhancing commercialization efficiency [24]. - The next phase for companies in the large model space will be a sustained battle for technological depth and the realization of commercial value [25].

Investment Rating - The report provides a "Buy" rating for the industry, indicating an expectation of stock performance exceeding the market by more than 10% over the next 12 months [31]. Core Insights - The demand for CPU is increasing due to AI agents, driven by three main factors: increased application call volume, orchestration becoming a bottleneck, and elevated overhead from sandbox isolation [2][11]. - The CPU ratio is expected to rise, with projections indicating that by Q2 2026, the CPU ratio per MW of GPU will increase from below 10% to 15% [2][17]. - There is significant growth potential in memory modules and interface chips, with a shift from traditional RDIMM to MRDIMM expected to enhance performance and capacity [2][19]. - The value of AI memory is transitioning from a cost item to an asset item, highlighting the increasing importance of related upstream infrastructure [2][25]. Summary by Sections AI Agent's Impact on CPU Demand - AI agents are driving a substantial increase in CPU demand due to higher application call volumes, orchestration challenges, and increased overhead from sandboxing [2][11]. Increasing CPU Ratio - Current estimates suggest that the CPU ratio per MW of GPU will rise to 15% by Q2 2026, with significant implications for the overall demand for CPUs in AI clusters [2][17]. Expanding Memory and Interface Chip Market - The transition to MRDIMM solutions is expected to significantly enhance bandwidth and capacity, creating opportunities in the memory and interface chip markets [2][19]. Investment Recommendations - The report emphasizes the importance of AI memory as a foundational capability for AI applications, recommending a focus on key beneficiaries within the industry chain [2][25].

Group 1 - Neynar will acquire and take over Farcaster, transferring the protocol, code repository, official applications, and Clanker to Neynar for operation and maintenance [1] - Mask founder Suji Yan commented that increased competition is beneficial and congratulated Neynar, revealing his small shareholder status in Neynar [1] - Aave founder Stani indicated that the management rights of the social protocol Lens will be transferred to the Mask team, with the original team serving as technical advisors [1] Group 2 - The first World Liberty Forum organized by World Liberty Financial (WLFI) will be held on February 18, 2026, at Mar-a-Lago, focusing on the future of finance, technology, and policy [2] - Notable attendees will include Goldman Sachs CEO David Solomon, Franklin Templeton CEO Jenny Johnson, and CFTC Chairman Michael Selig, among other political and business leaders [2] - The forum will discuss digital assets, AI, global financial market changes, and geopolitical risks, with new important developments to be announced [2] Group 3 - The NFT project Moonbirds announced that the BIRB token will have its Token Generation Event (TGE) on January 28, leading to a 7.9% increase in the floor price of Moonbirds NFTs to 2.46 ETH within 24 hours [3] Group 4 - The Ethereum-based privacy protocol Railgun has seen a record high in usage, with daily token protection transactions reaching 328 [4] - The ERC-8004 AI Agent ecosystem is expanding, with various projects emerging across infrastructure, orchestration, frameworks, developer tools, identity, marketplaces, and storage [4] Group 5 - Payment blockchain Keeta has signed an agreement to acquire a bank to enhance its compliance and regulatory capabilities, aiming for deeper integration with traditional financial infrastructure [4] - Keeta Inc. has approved the allocation of 35 million KTA tokens for acquisition-related arrangements [4] Group 6 - The Cardano Foundation has completed the first milestone of its governance roadmap, delegating approximately 220 million ADA to 11 community representatives (DReps), increasing the total delegated amount to 360 million ADA [5] Group 7 - Pendle announced the launch of sPENDLE, which will replace vePENDLE as the new governance and yield token, with staking opening on January 20 [5] - The new token will feature a 14-day exit period and will not require long-term locking, with protocol revenue used for PENDLE buybacks [5] Group 8 - PumpFun has launched the Pump Fund, initiating a $3 million Build in Public Hackathon to support 12 projects, each receiving $250,000 in funding [5] - Participating projects must issue tokens, build actual products, and hold at least 10% of the token supply [5] Group 9 - Ondo Global Market's stock token market cap on the BNB Chain has recently surpassed $100 million, growing nearly tenfold since early December [6] - With expansion to Solana, Ondo Global Market's total stock token market cap across Ethereum, BNB Chain, and Solana is approaching $500 million, representing over 50% of the total market cap for tokenized stocks [6]

Investment Rating - The report provides a "Buy" rating for multiple companies within the retail, technology, and internet sectors, indicating a positive outlook for their future performance [11]. Core Insights - Alibaba's Qianwen app has surpassed 100 million monthly active users within two months of its launch, showcasing significant user growth and a strategic shift towards becoming a "super entry" for task management [4]. - WeChat's small store has entered a new growth phase, with a 1.7 times year-on-year increase in the average number of active merchants, and aims to enhance its e-commerce infrastructure through various initiatives [5]. - ByteDance's Douzi 2.0 has been launched, focusing on three core functionalities that aim to transform AI into a collaborative partner for workplace efficiency [9][10]. Summary by Sections Alibaba - The Qianwen app has achieved over 100 million monthly active users, marking a significant milestone in its growth trajectory [4]. - The app's recent update integrates various Alibaba ecosystem services and introduces a task assistant feature, indicating a strategic pivot towards practical AI applications [4]. Tencent - WeChat's small store has seen a 1.7 times increase in active merchants year-on-year, with plans to enhance brand presence through reduced deposit requirements and new social features [5]. - The projected gross merchandise volume (GMV) for WeChat's small store is expected to double in 2025, reaching between 0.6 to 1 trillion [5]. Pinduoduo - Pinduoduo is currently under investigation by multiple government departments, which may impact its business operations and marketing activities [6][7]. - The company has faced a fine of 100,000 yuan due to tax compliance issues, which could further affect its market position during peak shopping seasons [6][8]. ByteDance - The launch of Douzi 2.0 emphasizes a shift from a single-task AI tool to a comprehensive workplace assistant, enhancing user capabilities through modular skills and programming features [9][10]. - The new functionalities aim to create a closed-loop system for task management, promoting proactive engagement from the AI [10].

Core Insights - Anthropic's AI programming tool, Claude Code, is leading a pivotal shift in the global software development industry, transitioning from manual coding to AI-driven coding with its new "Agentic" capabilities [1][2] - The annual recurring revenue (ARR) for Claude Code is projected to exceed $1 billion by the end of 2025, contributing 12% to Anthropic's total ARR of approximately $9 billion [1][2] Group 1: Product Development and Market Position - The breakthrough is attributed to the launch of Anthropic's latest AI model, Claude Opus 4.5, which significantly enhances programming capabilities beyond mere code completion [2][3] - Claude Code's ability to understand natural language instructions and autonomously complete development tasks is driving competition among startups and tech giants like OpenAI and Google [2][3] - The evolution of AI programming tools has accelerated, with a shift from basic auto-completion features to early "agent-based" programming products by startups like Cursor and Windsurf [3][4] Group 2: User Adoption and Internal Practices - The internal adoption of Claude Code is notable, with nearly 100% of Anthropic's technical staff frequently using the tool, resulting in 95% of their code being generated by Claude Code [4][5] - Corporate clients exhibit similar usage patterns, allowing Anthropic to optimize its products through internal use [5][6] Group 3: Expansion into Non-Programming Areas - Anthropic has launched Cowork, an AI agent product aimed at non-programmers, which can manage files and interact with software without requiring coding [6][7] - Cowork is designed to handle various tasks such as filling out forms and sending emails, indicating a broader application of AI agents beyond programming [6][7] Group 4: Competitive Landscape - The success of Claude Code is attracting more competitors into the AI programming market, with Cursor reporting an ARR of $1 billion and strong month-over-month revenue growth [7] - Major players like OpenAI, Google, and xAI are accelerating the development of their own agent-based programming products, intensifying competition in the market [7] - The core of this competition lies in the capabilities of underlying AI models, with Claude Opus 4.5 giving Anthropic a temporary advantage [7]

①AI服务器高端CPU严重缺货，供需失衡驱动服务器CPU价格上涨，这个环节是导致CPU涨价的核心因 素，国内产业链相关公司有望承接海外溢出订单；②CPU厂商扩产意愿不强，产能瓶颈短期内或难以解 决，国产CPU有望填补英特尔、AMD缺货留下的市场空白；③AI Agent带来芯片需求重构，若后续CPU 与存储产品持续提价，这些上游环节存在涨价或者增量的可能性。 前言 《金牌纪要库》是财联社VIP倾力打造的一款高端会议纪要类产品，结合财联社的媒体资源和行业圈层 优势，为投资者提供全面、深入的市场及行业洞察，以及专业分析和解读。 栏目专注于捕捉投资市场 的最新题材机会，通过一线记者的即时报道、资深编辑的专业整理，以及行业资深专家的深度访谈，为 投资者提供前瞻性、独家性、热门性及专业性的市场分析。 ...

AI已经深度扎根电商、零售的具体业务流程，以"数字员工"的形式，切实解决企业问题，带动实体经济提质增效。 消费者的需求正快速变化与升级，他们既要求更专业的解答，也渴望获得情绪共鸣与服务温度。提升服务质量已成为留住客 户、拉高转化的关键。 传统的配置型客服机器人已难以为继，什么才是"更智能的服务"？答案是：精准与温度并存。 AI智能体正是为此而生。它通过强大的知识库与学习能力，确保解答的专业性与一致性，稳定承接绝大部分日常服务，为用户 提供即时满足。更重要的是，它释放了人的精力，让客服人员能专注于更复杂、更需创造力和情感连接的工作，并通过实践不 断优化AI。 AI驱动客服从"成本中心"升级为"四大增长引擎" AI如何将客服从传统"成本中心"转变为"增长引擎"成为本次峰会的核心议题。 推动"AI生产力时代"落地 探迹科技子公司探域科技CTO黄加胜向21世纪经济报道记者表示，探域智能体作为探迹科技的B2C智能体产品，专注为电商与 零售企业提供全流程AI Agent服务，全面覆盖客服、营销、运营、私域等核心业务场景。 黄加胜表示，探域智能体坚持LLM Native技术路线，将传统客服辅助工具升级为具备"思考与行动"能 ...

Investment Rating - The industry investment rating is "Buy" with a previous rating of "Buy" as well [2]. Core Insights - The large-scale application of AI agents is expected to drive the demand for CPUs and foundational software tools, as AI agents require substantial computational resources for their operations [6][19]. - The report highlights that the demand for CPUs is anticipated to grow due to the CPU-intensive nature of tasks performed by AI agents, which cannot be easily replaced by GPUs [19][22]. - The report suggests that the current tight capacity in wafer fabs is leading to an increase in CPU prices, with Intel's LGA1700 products experiencing a price rise of 23% from October 2025 to January 2026 [22]. - The report emphasizes the potential for a dual trend of quantity expansion and performance upgrades in CPUs driven by the penetration of AI agents across various industries [19][22]. - Key companies to watch include domestic CPU leaders and IT firms such as StarRing Technology and Deepin Technology, which are expected to benefit from this trend [26][29]. Summary by Sections AI Agent and Infrastructure Demand - AI agents rely on runtime environments that provide the necessary tools and resources for their operations, which include task planning, data processing, and tool invocation [9]. - Virtual machines play a crucial role in determining the hardware resources available to AI agents, as they manage the allocation and scheduling of CPU and memory resources [11][16]. CPU Market Dynamics - The report notes that the configuration of server CPUs is evolving from traditional 8-16 cores to 32-64 cores and higher, indicating a shift towards more powerful processing capabilities [19]. - The report also mentions that the growth in storage demand is impacting CPU production capacity, further tightening the supply of CPUs [22]. Investment Recommendations - The report recommends focusing on key players in the software and hardware infrastructure related to AI, including: 1. AI foundational software tools: StarRing Technology, Paradigm Intelligence 2. Virtual machines: Deepin Technology, UCloud 3. CPUs: China Great Wall [26][29].

Core Insights - Yutaka Sasaki, CEO of NTT DATA Group, predicts that 2026 will be the year of the "AI Agent" as advancements in artificial intelligence continue to evolve [1] - The integration of AI into organizations is expected to significantly impact job structures across various industries [1] Group 1 - NTT DATA Group is actively engaging in discussions about the future of AI and its implications for the workforce [1] - The company is positioning itself to lead in AI developments, anticipating a transformative shift in how AI will be utilized in business operations [1]