Core Viewpoint - The software ETF (515230) rose over 0.7% on January 8, driven by the potential of the mHC architecture to reshape industry logic through a new paradigm in AI chip design [1] Group 1: mHC Architecture - The mHC architecture is expected to introduce a new paradigm in AI chip design by combining manifold constraints with engineering optimization, addressing the mismatch between computing power and bandwidth [1] - This architecture promotes the logic of "software actively adapting to hardware bottlenecks," steering the industry towards a direction of "efficiency first" in hardware-software collaboration [1] - mHC significantly reduces bandwidth requirements through optimizations like kernel fusion and selective recomputation, indicating that chip design does not need to solely pursue high-bandwidth memory [1] Group 2: Innovation and Market Impact - The manifold constraint logic of mHC may drive innovation in dedicated computing units for chips, potentially breaking the current monopoly of "general computing units" in AI chips and advancing towards a "general + dedicated" heterogeneous architecture [1] - The mHC architecture, as released by DeepSeek, has shown stable performance, excellent scalability, high cost-effectiveness, and efficiency in large-scale model training [1] - mHC may indicate a new direction for the evolution of next-generation infrastructure, deepening the understanding of how topological structures influence optimization and representation learning [1] Group 3: Software ETF Overview - The software ETF (515230) tracks the software index (H30202), which selects securities from publicly listed companies involved in software development, services, and related technology fields to reflect the overall performance of the software industry [1] - The software index primarily focuses on the information technology sector, characterized by significant growth potential and innovation [1]

Core Insights - DeepSeek has introduced a new architecture called mHC aimed at addressing stability issues in large-scale model training while maintaining performance improvements [1][11]. Group 1: Problem Identification - Large models face a dilemma in training stability, where traditional single-channel connections lead to information congestion as model size increases [3][5]. - Previous solutions, like the hyper-connection approach, improved efficiency but introduced new issues such as uncontrolled information amplification or suppression, leading to gradient explosion and training failures [5][7][9]. Group 2: mHC Architecture - The mHC architecture incorporates an intelligent scheduling system for multi-channel connections, utilizing the Sinkhorn-Knopp algorithm to maintain energy conservation during information transmission [11][13]. - Additional design features include non-negative constraints on input-output mappings to prevent useful signal loss due to coefficient cancellation [15]. Group 3: Infrastructure Optimization - DeepSeek has optimized its infrastructure by merging multiple computation steps into a single operator, reducing memory read/write cycles and employing recomputation strategies to lower memory usage [16][18]. - These optimizations have resulted in significant stability improvements with minimal increases in training time, even at an expansion factor of 4 [18]. Group 4: Performance Validation - Testing on various model sizes, particularly a 27 billion parameter model, demonstrated that mHC effectively resolved training instability issues, achieving lower loss values compared to traditional baseline models [21][22]. - The performance advantages of mHC were consistent across different model sizes, indicating its practical value for both small and large models [24]. Group 5: Industry Implications - The introduction of mHC suggests a shift in the industry towards refined architectural designs rather than merely increasing parameters and computational power, potentially lowering entry barriers for smaller companies in the large-scale model domain [26][29]. - This pragmatic technological innovation is expected to facilitate the deployment of AI technologies, making it easier for more enterprises to engage in large-scale model development [29].

Core Viewpoint - The mHC architecture is expected to introduce a new paradigm in AI chip design by combining manifold constraints with engineering optimization, addressing the mismatch between computing power and bandwidth [1] Group 1: mHC Architecture - The mHC architecture promotes the logic of "software actively adapting to hardware bottlenecks," driving the industry towards a "efficiency-first" approach in hardware-software collaboration [1] - Through optimizations like kernel fusion and selective recomputation, mHC significantly reduces bandwidth requirements, allowing chip design to move away from solely pursuing high-bandwidth HBM memory [1] - The manifold constraint logic of mHC may foster innovation in dedicated computing units for chips, breaking the current monopoly of "general computing units" in AI chips and advancing towards a "general + dedicated" heterogeneous architecture [1] Group 2: Performance and Market Impact - The mHC architecture, as released by DeepSeek, demonstrates stable performance and high scalability in large-scale model training, indicating a potential direction for the next generation of infrastructure [1] - The framework is expected to reignite academic interest in macro-architecture design and deepen the understanding of how topological structures influence optimization and representation learning, potentially breaking current limitations [1] Group 3: ETF and Market Representation - The Guotai AI ETF (159388) tracks the ChiNext AI Index (970070), which has a daily price fluctuation limit of 20% [1] - This index selects listed companies involved in AI technology and related applications from the ChiNext market, covering various segments from hardware manufacturing to software development, reflecting the overall performance of AI-related listed companies in the ChiNext market [1]

Group 1 - The mHC architecture is expected to introduce a new paradigm in AI chip design by combining manifold constraints with engineering optimization, providing a novel approach to address the mismatch between computing power and bandwidth [1] - This architecture promotes the logic of "software actively adapting to hardware bottlenecks," driving the industry towards a direction of "efficiency-first" hardware-software collaboration [1] - mHC significantly reduces bandwidth requirements through optimizations like kernel fusion and selective recomputation, indicating that chip design does not need to solely pursue high-bandwidth memory [1] - The manifold constraint logic of mHC may foster innovation in dedicated computing units for chips, breaking the current monopoly of "general computing units" in AI chips and advancing towards a "general + dedicated" heterogeneous architecture [1] - The mHC framework, released by DeepSeek, demonstrates stable performance and good scalability in large-scale model training, offering high cost-effectiveness and efficiency, potentially guiding the evolution of the next generation of infrastructure [1] - This framework is likely to reignite academic interest in macro-architecture design and deepen the understanding of how topological structures influence optimization and representation learning, breaking current limitations [1] Group 2 - The Computer ETF (512720) tracks the CS Computer Index (930651), which selects listed companies involved in software development, IT services, and hardware manufacturing from the Shanghai and Shenzhen markets to reflect the overall performance of the computer industry [2] - The CS Computer Index focuses on the information technology sector, with constituent stocks exhibiting high growth potential and technological innovation capabilities, effectively representing the development trends of China's computer industry [2]

Investment Rating - The industry investment rating is "Positive" and is maintained [8] Core Insights - The domestic AI industry is experiencing positive developments, with significant events such as Meta's acquisition of Manus and the IPOs of Zhiyu and MiniMax in Hong Kong. These changes indicate that 2026 may be a pivotal year for the AI industry, transitioning from technological breakthroughs to large-scale implementation [2][4][6] - The current phase of the AI large model market in China has shifted from an early "hundred model battle" to a critical stage of "application heat" and "value verification," suggesting that resources may concentrate on leading firms [6] - The report suggests focusing on domestic large model vendors, major cloud service providers, vertical scenario agent vendors, and the domestic computing power supply chain as potential investment opportunities [2][6] Summary by Sections Event Description - The report highlights that the domestic AI industry has seen a surge of positive news around the New Year holiday, with key developments indicating that 2026 could be a transformative year for the industry [4] Event Commentary - The report discusses the IPOs of Zhiyu and MiniMax, marking a significant step for China's large model industry as it enters a phase of value verification. The funds raised will primarily support AI model development and infrastructure optimization [6] - The acquisition of Manus by Meta is noted as a strategic move to enhance Meta's capabilities in agentic AI, potentially leading to scalable and practical AI applications [10]

Core Viewpoint - DeepSeek has introduced a new network architecture called mHC (Manifold-Constrained Hyper-Connections) aimed at addressing instability issues in large-scale model training, potentially guiding the evolution of next-generation infrastructure [3][6]. Group 1: Technical Innovations - The mHC architecture improves upon traditional hyper-connection frameworks by stabilizing information transmission in neural networks, akin to adding "traffic rules" to information channels, thus enhancing model training efficiency and scalability [7]. - The paper suggests that mHC opens up numerous promising research avenues, potentially reigniting academic interest in macro-architecture design and deepening understanding of how topological structures affect optimization and representation learning [8]. Group 2: Industry Implications - mHC may enable companies to reduce hardware investments and shorten training cycles when developing larger foundational models, lowering the barrier for small to medium AI enterprises to create more complex models [8]. - Enhanced training stability and scalability could facilitate the deployment of large models in more complex scenarios, such as multi-modal models requiring extensive parameters and industrial-grade intelligent decision systems [8]. - Industry experts view DeepSeek's research as foundational innovation, predicting significant updates in the upcoming V4 version based on this architecture [8]. Group 3: Recent Developments - Despite not launching major versions like R2 or V4 in 2025, DeepSeek has continued to iterate and open-source its models, releasing DeepSeek-V3.2 and DeepSeek-Math-V2, the latter being the first mathematical model to reach international Olympiad gold medal standards [9].

Group 1 - DeepSeek has released a new paper proposing a novel mHC architecture, with CEO Liang Wenfeng listed as one of the authors [1] - Anker Innovation has responded to rumors of a 30% layoff, stating that the reported figure is significantly exaggerated and that the adjustments are part of a strategic upgrade [2] - Tesla has launched a HarmonyOS version of its app in the Huawei app market, supporting features like remote vehicle control and mobile key [3] Group 2 - Xiaomi has announced a limited-time offer for the YU7 model, allowing customers to choose between a tax subsidy and a three-year interest-free option for orders placed before December [5] - The Redmi Note 15 series has been officially launched, starting at 999 yuan, with various color options available [6] - Huawei has released the Smart Screen V6, with prices ranging from 7999 to 14999 yuan, and is offering a limited-time discount on its high-end ADS feature package [7] Group 3 - Apple has updated its list of "vintage products," including the iPhone 11 Pro and the last Intel MacBook Air [8] - Seres Group announced that the AITO car deliveries exceeded 57,000 units in December, setting a new monthly record, with total deliveries surpassing 420,000 units for the year [9] - Li Auto plans to focus on adjusting its models in the 300,000 to 400,000 yuan price range while continuing to iterate on its pure electric i8 series [10] Group 4 - Li Auto has achieved a cumulative delivery milestone of over 1.5 million vehicles, becoming the first new force brand in China to reach this figure [12] - Huawei's Enjoy series has met its annual challenge goals for 2025, with plans to introduce more diverse products in 2026 [13] - TrendForce reports that Samsung is rigorously executing its production halt plan, which may lead to a significant increase in DDR4 memory prices in 2026 [14]

Core Viewpoint - DeepSeek has introduced a new architecture called mHC (Manifold-Constrained Hyperconnection) aimed at addressing the instability issues in traditional hyperconnections during large-scale model training while maintaining significant performance gains [1][3]. Summary by Sections Introduction of mHC - DeepSeek's new paper presents mHC, which projects the hyperconnection's residual connection space onto a specific manifold to restore the identity mapping property and ensure operational efficiency through rigorous infrastructure optimization [3][4]. Performance and Scalability - Empirical results indicate that mHC effectively supports large-scale training, with an additional time overhead of only 6.7% when the expansion rate is set to 4 [4][6]. Research Directions - mHC opens up several important research directions, including compatibility with various manifold constraints tailored for specific learning objectives and potential new methods for balancing plasticity and stability through in-depth studies of differential geometric constraints [7]. Recent Developments - DeepSeek has been active, releasing two official model versions, DeepSeek-V3.2 and DeepSeek-V3.2-Speciale, with the former achieving performance comparable to GPT-5 in benchmark tests [8]. - The DeepSeek-V3.2-Speciale model combines enhanced reasoning capabilities with mathematical proof abilities, performing well in mainstream reasoning benchmarks [8]. - Additionally, the release of DeepSeek-V3.2-Exp introduces a sparse attention mechanism aimed at improving training and inference efficiency for long texts, with a significant reduction in API costs for developers [9]. Recognition in the Scientific Community - DeepSeek's research paper on the DeepSeek-R1 reasoning model was featured on the cover of the prestigious journal Nature, marking a significant milestone for Chinese AI technology in the international scientific community [9][10].