Core Insights - The evaluation system for cloud-based large models is shifting from purely capability metrics to the actual completion of tasks, with a focus on code capabilities and multi-agent systems by leading overseas companies since 2026 [1] - The dual capability stack of "fast interaction + long reasoning" is expected to become a significant evolution direction for general-purpose agents in the near future [2] - The collaboration between edge models and cloud models is emphasized, with edge models handling high-frequency, lightweight tasks locally, while heavier reasoning tasks are processed in the cloud [3] Cloud Models - The expansion of capability boundaries and cost restructuring are occurring simultaneously in cloud models, with a focus on task completion [1] - Leading companies are intensively laying out code capabilities and multi-agent systems to enhance performance [2] Code Models - The reasoning demands in the era of intelligent agents are evolving along two optimization directions: long-chain complex reasoning and real-time interaction [2] - Low-latency agents like OpenAI's Codex-Spark prioritize interactive AI experiences, while agents like Claude4.6 focus on improving success rates in complex tasks through increased context length [2] Edge Models - The evolution of edge models is characterized by efficiency optimization and capability compression under a collaborative framework with cloud models [3] - Multi-modal capabilities are becoming a key competitive point for edge models, with a focus on achieving zero-latency interactions [3] Hardware Reconstruction - The industry is expected to focus on high-frequency demand scenarios in 2024, with a shift towards multi-modal creative capabilities by 2025 [4] - Key components for edge models are undergoing upgrades in memory and power consumption to enhance user experience [4] Future Outlook - Next-generation flagship SoC platforms like Qualcomm's Snapdragon 8 Elite Gen 6 are anticipated to provide enhanced hardware support for the complexity and multi-modality of edge AI functions [5]

证券研究报告·行业深度报告·电子 [Table_Tag] [Table_Summary] 投资要点 ◼ 端侧模型牵引硬件重构：算力、存力与散热协同升级。从整机 AI 功能 看，2024 年行业整体仍以高频刚需场景为切入点，重点围绕图像消除、 文本摘要等低门槛功能；进入 2025 年，厂商明显加速向多模态创作能 力延展，覆盖语音、生成式图像等更复杂交互形态，并进一步向操作系 统底层渗透。整机 AI 竞争正从功能数量比拼，转向多模态体验与系统 级整合深度的综合较量。在整机级 AI 能力向多模态等方向升级的背景 下，端侧核心部件也正围绕内存与功耗等制约端侧体验的关键变量上进 行新一轮升级。在存储侧，三星 LPDDR6 产品在支持更高数据传输速率 和内存带宽的情况下，还从电路架构到电源管理进行了系统性重构，使 LPDDR6 在保持高速性能的同时，实现较上一代约 21%的能效提升。在 散热侧，三星于 2025 年 12 月 19 日发布 Exynos 2600 芯片，首次在移 动 SoC 中引入 High-k EMC 材料优化热传输路径，使热阻较 Exynos 2500 降低约 16%。在重载场景（如游戏与端侧 AI ...

而在台下，路线之争也在酝酿着分歧，纯粹"卖Token"的商业模式正面临潜在危机。OpenClaw等开源框架暴露出不小的安全风险与不稳定的调用体验，且其 本身的开发壁垒不高。 各大厂慢慢意识到，真正的护城河不在于开放API，而在于通过自主研发的本地化开发工具与端侧智能体，锁死用户的数据上下文与业务场景。 2026年之春，"养龙虾"在AI圈彻底火了。这个让人人都能拥有自己的贾维斯的OpenClaw，自己出圈的同时，顺手把卖大模型token的玩家给养肥了。 由于Agent运行需要大量token，API调用量指数级飙升。作为"卖铲人"，智谱、MiniMax、Kimi、阿里云等国内大模型厂商纷纷推出Coding Plan的API套餐， 来迎接这泼天的富贵，智谱的顶配套餐甚至被抢空。 "卖铲人"在此期间赚得盆满钵满，营收随之暴增，在资本市场也迎来了史无前例的估值狂欢。 智谱与MiniMax在港股的市值飙升，双双突破三千亿港元大关，成了大厂身价的"计量单位"，未上市的Kimi则以破百亿美元估值，刷新了国内大模型独角兽 的纪录。 眼下，当AI变成能够自我循环、疯狂消耗Token来纠错和执行的自动化机器时，算力资源的消耗速度超 ...

Investment Rating - The report maintains a rating of "Buy" for the electronic industry, indicating a positive outlook for investment opportunities in this sector [1]. Core Insights - The IoT market is identified as the largest blue ocean market, presenting significant opportunities for domestic substitution, particularly in customized solutions and software ecosystems [2]. - The report emphasizes the importance of hardware supply chain enterprises in the AI transformation, as major internet and cloud computing companies accelerate their hardware ecosystem development [2]. - The evolution of edge AI is seen as a critical trend, with the need for high-performance edge hardware driving innovation in traditional mobile and PC markets [5][6]. - The automotive sector is highlighted as a prime application area for edge AI, with significant opportunities arising from the upgrade of in-vehicle chips and the construction of domestic ecosystems [5]. Summary by Sections 1. Edge AI and Domestic Supply Chain Opportunities - The transition of edge AI from concept to a well-defined industry path marks a strategic shift towards physical world applications, driven by privacy, security, and latency considerations [15]. - The deep restructuring of edge hardware provides a systemic elevation opportunity for domestic supply chains, particularly in new terminal markets like AI glasses and embodied intelligent robots [16]. 2. AI Empowering Mobile and PC Market Innovations - The demand for high-end smartphones is increasing due to the rapid adoption of AI technology, with projections indicating that by 2028, 54% of smartphones will feature edge AI capabilities [18]. - The average selling price (ASP) of smartphones is expected to rise, with a notable increase in the proportion of high-end models, driven by the demand for AI functionalities [21][19]. - The report notes that the semiconductor industry is experiencing a shift towards higher-end chip manufacturing processes, with TSMC's 2nm technology expected to enhance performance and efficiency significantly [23][24]. 3. Automotive Electronics and Edge AI Growth - The automotive sector is positioned as a second growth engine for edge AI, with in-vehicle chips evolving to meet the demands of intelligent driving and user interaction [5]. - The report discusses the competitive landscape of automotive chips, highlighting the rapid advancements in domestic chip manufacturers and their collaboration with new energy vehicle companies [5]. 4. Internet Giants Building Edge-Cloud Collaborative Ecosystems - Major internet companies are establishing comprehensive strategies that integrate cloud, AI, and chip development to strengthen their hardware foundations for AI transformation [10]. - The report outlines the strategic moves of companies like Alibaba, ByteDance, and Tencent in creating a cohesive hardware ecosystem that supports AI applications across various sectors [10].

Core Insights - The article discusses the strategic shift of Mianbi Intelligent towards edge-side large models, which gained credibility after Apple's entry into the market. This shift has led to the release of the first large model capable of "instant free dialogue" and the AI hardware Pinea Pi for full-stack development [2][3]. Group 1: Model Development - Mianbi officially released and open-sourced the new generation multimodal flagship model MiniCPM-o 4.5, which features an end-to-end "watch, listen, and speak" capability, allowing for real-time dialogue interactions [3][5]. - The model introduces a full-duplex mechanism where multimodal inputs and outputs do not block each other, enabling continuous perception of external audio and video streams while generating responses [5][6]. - The development faced challenges in unified training of various modalities, but the team successfully maintained text capabilities while improving efficiency and response speed [6][11]. Group 2: Hardware Development - Mianbi emphasizes the importance of collaboration with chip manufacturers to optimize model training and performance on specific hardware [13][14]. - The launch of Pinea Pi, an AI-native edge intelligent development board, aims to facilitate the development and application of models in various scenarios, focusing on market education rather than immediate commercialization [16][14]. - The hardware integrates multimodal components and is designed to reduce the adaptation effort for developers, with plans for future iterations based on user feedback [16][14]. Group 3: Market Strategy - Mianbi's core philosophy is based on the "Knowledge Density Law," suggesting that the knowledge density of large models doubles approximately every 100 days, necessitating continuous model innovation [17][18]. - The company aims to create a system capable of consistently training high-density knowledge models, which is crucial for maintaining a competitive edge in the rapidly evolving AI landscape [18][19]. - Mianbi focuses on the edge market, which is fragmented and offers numerous opportunities for startups to target specific applications without competing directly with larger companies [19][20]. Group 4: Future Directions - Mianbi envisions a future where edge and cloud collaboration will be the mainstream model, addressing issues like latency and privacy while enhancing user interaction with intelligent terminals [23][24]. - The company believes that advancements in multimodal capabilities will be foundational for future multi-agent systems, enabling efficient collaboration among different intelligent agents [25][26]. - Mianbi anticipates that within the next one to two years, models will gain stronger autonomous learning capabilities, leading to significant breakthroughs in multi-agent collaboration and the emergence of intelligent assistants that understand user needs [26].

Core Insights - AI technology is reshaping various industries, with storage technology becoming increasingly critical as a backbone for AI applications. Jiangbolong has emerged as a key player in the AI storage transformation with innovative solutions and precise market positioning [1]. Group 1: Full-Stack Solutions - Jiangbolong provides a full-stack solution for AI servers and computing integrated machines, covering all scenarios of AI training and inference. Key products include eSSD, RDIMM, SOCAMM2, and innovative memory solutions, which offer efficient and reliable storage performance tailored to complex AI needs [1]. - The new UNCIA 3856 SATA eSSD features high-quality 3D eTLC NAND and self-developed firmware algorithms, achieving a balance of large capacity, low power consumption, and high endurance, thus providing a solid data storage foundation for AI servers [1]. Group 2: Memory Solutions - Jiangbolong's DDR5 RDIMM and MRDIMM memory modules are core choices for general servers and AI infrastructure due to their high bandwidth, low latency, and excellent compatibility. The DDR5 MRDIMM significantly enhances data transfer rates through a multi-channel architecture, delivering unprecedented performance for AI computing integrated machines [3]. - The SOCAMM2 memory product, based on LPDDR5/5x particles and CAMM modular design, meets the stringent performance and energy efficiency requirements of data centers. It offers ultra-high transmission rates (up to 8533 Mbps) and low power consumption (one-third of standard DDR5 RDIMM), providing dual enhancements in capacity and bandwidth for intelligent computing centers [3]. Group 3: Edge AI Solutions - Jiangbolong has introduced the integrated packaging mSSD for edge AI applications, utilizing wafer-level system-in-package (SiP) technology to integrate the controller, NAND, PMIC, and other components into a single package. This makes mSSD an ideal storage solution for edge AI devices such as AI PCs and robots, offering flexibility and efficiency [5]. Group 4: Strategic Vision - In the AI era, mere improvements in storage performance are insufficient to meet the complex and changing application demands. Jiangbolong aims to achieve efficient utilization of storage resources and flexible release of computing power through an edge-cloud collaborative strategy. The company strengthens partnerships within the industry chain to inject critical storage capabilities into AI intelligent computing center construction and promote the continuous advancement and widespread application of edge AI storage technology [7]. - Jiangbolong's outstanding performance in the AI storage sector is reflected not only in the continuous launch of innovative products but also in its deep insights into industry trends and precise understanding of customer needs. The company will continue to adhere to an innovation-driven development strategy, focusing on core media like mSSD to iterate more forms and scenarios of innovative packaged storage, contributing further to the popularization and application of AI technology [7].

Core Insights - The article discusses the launch of the new open-source agent model Step 3.5 Flash, which features a total of 196 billion parameters and 11 billion active parameters, supporting a context window of 256K [2][36]. Model Performance - The model achieves a peak inference rate of 350 TPS, comparable to closed-source models in agent scenarios and mathematical tasks, capable of handling complex, long-chain tasks [5][41]. - In benchmark tests, Step 3.5 Flash scored 97.3 in the AIME 2025 benchmark, 74.4% in the SWE-bench Verified coding tasks, and 88.2 in the τ²-Bench for agent tasks, indicating strong performance across various applications [7][6]. Technical Architecture - Step 3.5 Flash employs a MoE sparse mixture of experts architecture, activating approximately 11 billion parameters during inference to control computational and deployment costs effectively [36]. - The model incorporates a 3:1 sliding window attention mechanism to address long context issues, enhancing its ability to manage lengthy texts [37]. - It features a self-developed MIS-PO reinforcement learning framework to improve inference and agent execution capabilities, reducing data noise and gradient variance for stable optimization in long-sequence tasks [42]. Ecosystem Integration - The model is designed to work seamlessly with major AI acceleration chip platforms from various manufacturers, including Ascend, Mu Xi, and Alibaba's T-head, ensuring compatibility with current mainstream domestic AI hardware [4]. - Step 3.5 Flash emphasizes a cloud-edge collaboration approach, where the cloud handles complex planning and reasoning while the edge focuses on secure data retrieval and local execution [30][32]. Future Developments - The development team is already working on Step 4, indicating ongoing advancements in the model's capabilities [43].

Core Insights - The Alibaba Cloud Tongyi Intelligent Hardware Exhibition showcased over 1,500 hardware products integrated with AI, demonstrating how AI is becoming tangible in everyday devices [1][2] - AI is transforming various products, from automatic cat litter boxes to sports equipment, enhancing their functionality and intelligence [2][3] - The exhibition highlighted the role of Alibaba Cloud in accelerating the integration of AI into hardware, making advanced technology more accessible [3][4] Group 1: AI Integration in Hardware - The automatic cat litter box now monitors the health of cats by analyzing sounds and urine pH levels, showcasing AI's ability to enhance pet care [2] - A basketball selfie device can track shooting accuracy and analyze game strategies, illustrating AI's application in sports [2] - The ACEMATE AI tennis robot can analyze player movements and predict ball trajectories, providing a virtual training experience [2] Group 2: Market Trends and Opportunities - The exhibition emphasized the need for diverse cloud service providers that can support various AI model requirements, indicating a growing demand for tailored AI solutions [7] - Alibaba has open-sourced over 300 models, facilitating faster adoption of AI technologies among small and medium enterprises, thus promoting a robust AI hardware ecosystem [7][8] - The "end-cloud collaboration" model is essential for entrepreneurs to quickly identify and develop critical features that resonate with users [8] Group 3: Future of AI Hardware - Multi-modal interactions are reshaping user experiences, moving towards a more seamless and intuitive interface in AI consumer hardware [9] - AI is transitioning from merely adding features to becoming a critical entry point for user interaction with digital services [9][10] - The exhibition served as a collective demonstration of how hardware can be redefined through advanced AI models, indicating a shift in the industry towards innovative product development [10]

Core Insights - The CES 2026 event highlights the significant presence of Chinese companies, which account for one-quarter of all exhibitors, signaling a shift in AI hardware from novelty to independent intelligent agents [1][3] - Alibaba Cloud is playing a crucial role in this transformation, providing foundational support for hardware innovation without pursuing independent hardware development [4][13] - The evolution of AI hardware is marked by a transition from "functional intelligence" to "system-level intelligence," driven by advancements in underlying technologies and business models [5][6] Group 1: Industry Transformation - The AI hardware industry is entering a phase of explosive growth, moving away from "pseudo-intelligence" to devices capable of understanding complex human interactions [5][6] - The collaboration between cloud computing and edge computing is becoming the industry standard, with Alibaba Cloud offering tailored intelligence solutions for various hardware needs [7][8] - The integration of AI capabilities into hardware is redefining user interactions, allowing devices to respond to more abstract user requests and adapt to human habits [8][10] Group 2: Alibaba Cloud's Role - Alibaba Cloud is recognized as an "emerging leader" in key areas such as cloud infrastructure and model capabilities, positioning itself alongside top global AI companies [9] - The company emphasizes the importance of making technology accessible and affordable, ensuring that AI hardware becomes a common tool rather than an expensive luxury [4][24] - By providing extensive model capabilities, Alibaba Cloud enables over 150,000 hardware manufacturers to innovate without the need for substantial resources [27][28] Group 3: Market Dynamics - The market for AI consumer hardware is shifting towards a more discerning consumer base that values stability, experience, and real-world application [10][12] - Users are increasingly willing to pay for perceived value, with a notable rise in subscription models for AI functionalities, particularly in health-related applications [12][10] - The relationship between large tech companies and smaller manufacturers is evolving from competition to collaboration, fostering a more symbiotic ecosystem [24][28] Group 4: Product Innovations - New AI products, such as the Mooni M1 for children, demonstrate how AI can enhance user interaction by focusing on individual needs and preferences [14][16] - In the smartphone sector, companies like OPPO and Honor are integrating advanced AI capabilities to improve user experience and interaction [18][19] - The automotive industry is also seeing significant advancements, with AI systems transforming vehicle interfaces into intelligent hubs that understand and respond to user needs [19][21]

Core Viewpoint - The article emphasizes that AI hardware has evolved from being mere gimmicks to becoming independent intelligent entities, showcasing significant innovation from Chinese manufacturers at CES 2026 and the role of Alibaba Cloud in this transformation [3][6][19]. Group 1: AI Hardware Evolution - AI hardware has crossed the threshold from being toys to independent intelligent agents, indicating a major shift in the industry [3][6]. - The transition from "China manufacturing" to "China creation" is highlighted, with Alibaba Cloud playing a crucial role in providing the necessary technological support for hardware innovation [7][19]. - The year 2026 marks the beginning of a "species explosion" in AI hardware, driven by continuous technological advancements and a shift in business logic [9][10]. Group 2: Technological Advancements - AI hardware is undergoing a transformation from "functional intelligence" to "system-level intelligence," with the ability to understand complex environments [10][11]. - The collaboration between cloud computing and edge computing is becoming the industry standard, allowing for a balance between performance and power consumption [11][12]. - Alibaba Cloud's models, ranging from 0.5 billion to 480 billion parameters, provide tailored intelligence solutions for various hardware needs [11][12]. Group 3: Market Dynamics - The market for AI consumer hardware is shifting from an "enlightenment market" to a more mature one, where users demand stable performance and real-life value [13][14]. - Users are increasingly willing to pay for AI functionalities, particularly in health-related applications, indicating a significant change in consumer behavior [14][16]. - The evolution of payment models from one-time purchases to subscription services reflects a deeper integration of AI into everyday life [16]. Group 4: Role of Alibaba Cloud - Alibaba Cloud is positioned as a silent innovator, supporting over 150,000 hardware manufacturers without competing directly in the hardware market [19][34]. - The company’s approach fosters trust among manufacturers, allowing them to innovate collaboratively rather than defensively [19][34]. - By providing a robust AI infrastructure, Alibaba Cloud enables Chinese manufacturers to transition from low-margin assembly to high-value innovation [39][40]. Group 5: New Product Innovations - New AI products, such as the Mooni M1 for children, demonstrate how AI capabilities are integrated into hardware, enhancing user interaction [22][28]. - The collaboration with various companies has led to significant advancements in wearable technology and creative hardware, showcasing the versatility of AI applications [24][26]. - The smartphone industry is also experiencing a transformation, with enhanced interaction capabilities through AI, making devices more intuitive and responsive [29][30]. Group 6: Global Market Position - Chinese hardware is evolving from being a follower in the global tech landscape to becoming a leader in defining new standards [36][38]. - The collaboration between Alibaba Cloud and hardware manufacturers is reshaping the global value chain, allowing for a more competitive stance in the international market [39][40].