此外，华为也在近期宣布，自昇腾950PR开始，昇腾AI芯片将采用华为自研的HBM。其中，昇腾950搭载自研的HBM HiBL 1.0；昇腾 950DT升级至HBM HiZQ 2.0。 为什么是HBM？ AI时代， 存储芯片已从配角跃升为核心瓶颈与突破口。 随着大模型参数规模与训练数据量的爆炸式增长，传统内存技术已成为制约算力发挥 的"内存墙"， 而HBM凭借其超高带宽、低功耗和小体积特性，正成为AI芯片的主流选择。 全球科技巨头纷纷将HBM作为战略要地。 在日前的财报会上，美光特别强调， 预计半导体芯片、特别是HBM的供不应求情况将会加剧。 公司CEO Sanjay Mehrotra透露，当前半导体 存储领域，DRAM库存已低于目标水平，NAND库存持续下滑；而HBM产能需求增长显著，产能已被锁定，预计2026年HBM出货量增速将 超过整体DRAM水平，成为半导体存储板块的核心增长驱动力。 具体HBM产品中，美光表示， 已与几乎所有客户就2026年绝大部分HBM3E产能达成定价协议；公司正在与客户就HBM4进行讨论，供应"非常紧 张"。 券商指出，人工智能技术的蓬勃发展推动大模型训练走向规模化，但真正创造持续 ...

Group 1: Stock Market Activity - 31 stocks were listed on the institutional trading leaderboard today, with 17 seeing net purchases and 14 experiencing net sales [1] - The top three stocks with the highest net purchases by institutions were Huahong Semiconductor (6.84 billion), Dofu Dofu (3.32 billion), and Xingye Yinxin (3.2 billion) [1][2] Group 2: Company Insights - Huahong Semiconductor is recognized as a global leader in specialty process wafer foundry, with a comprehensive platform for specialty processes [2] - The company has completed the development of its 40nm logic/RF platform, with 12 products from Alpha customers entering mass production and a shipment volume exceeding 45,000 pieces [2] Group 3: Industry Trends - The "storage computing" technology is emerging as a disruptive paradigm to overcome computing power bottlenecks, shifting vector data from expensive DRAM and HBM memory to cost-effective SSDs [2] - This shift is expected to significantly reduce computing power consumption and accelerate AI inference, leading to a higher growth rate in SSD demand compared to traditional trends [2] Group 4: HDD Market Outlook - Despite the ongoing impact of SSDs on the consumer HDD market, there remains substantial demand for HDDs in data centers, enterprise storage, and big data processing due to their advantages in capacity and cost [3] - The sales volume of the company's base plate business is expected to see significant growth compared to the same period last year by mid-2025 [3] Group 5: Nonferrous Metals Industry - The Ministry of Industry and Information Technology and other departments have issued a work plan for the nonferrous metals industry, targeting an average annual growth of around 5% in value added from 2025 to 2026 [4] - The production of ten nonferrous metals is projected to grow at an average annual rate of about 1.5%, with positive developments in domestic resource development for copper, aluminum, and lithium [4] Group 6: Copper Market Performance - The international copper market saw a night trading increase of 2.07%, while the Shanghai copper futures rose by 1.27%, reaching a new high for the year [5]

Group 1 - The Shanghai Stock Exchange Sci-Tech Innovation Board Chip Index (000685) has seen a strong increase of 2.14%, with constituent stocks such as Baiwei Storage (688525) rising by 7.66%, Yandong Micro (688172) by 7.10%, and Lexin Technology (688018) by 5.24% [1] - The Sci-Tech Chip ETF Index (588920) also rose by 2.30%, with the latest price reported at 1.65 yuan [1] - Tianfeng Securities highlights that an AI storage revolution is underway, driven by "computing through storage," which significantly reduces computing power consumption and accelerates AI inference, leading to a higher growth rate in SSD demand compared to traditional curves [1] Group 2 - The Sci-Tech Chip ETF Index closely tracks the Shanghai Sci-Tech Innovation Board Chip Index, which selects securities related to semiconductor materials and equipment, chip design, manufacturing, packaging, and testing from listed companies on the Sci-Tech Board [2] - As of August 29, 2025, the top ten weighted stocks in the Sci-Tech Chip Index include Cambricon (688256), Haiguang Information (688041), SMIC (688981), and others, with these ten stocks accounting for a total of 62.02% of the index [2]

Group 1 - HSBC predicts that by 2026, the Shanghai Composite Index will reach 4500 points, the CSI 300 Index will reach 5400 points, and the Shenzhen Component Index will reach 16000 points, representing an increase of 17-20% [1] - Morgan Stanley reports that over 90% of roadshow clients expressed willingness to increase exposure to Chinese assets, marking the highest interest since early 2021 [1] - Fidelity International notes a significant increase in global investors' interest in Chinese assets, with hedge funds showing the highest activity in China's stock market in recent years [2] Group 2 - Barclays states that gold prices do not appear overvalued, with gold ETF holdings at their highest since 2022, and prices have surged over 40% this year [2] - Nomura expects the Reserve Bank of Australia to maintain its cash rate, with a shift towards a less dovish communication stance [3] - Nomura also indicates that volatility in the USD/JPY exchange rate may increase due to upcoming data and events [4] Group 3 - CICC suggests that the credit cycle in both China and the US may be approaching turning points, impacting market directions [9] - Guotai Junan emphasizes the importance of the fourth quarter for cyclical industries and high-growth sectors, with a historical tendency for cyclical industries to perform well [11] - Huatai Securities predicts that PPI year-on-year and industrial profits are likely to continue their recovery trend [14]

Core Insights - The report from Tianfeng Securities highlights the emergence of "storage-based computing" as a disruptive technological paradigm to overcome computing bottlenecks and storage limitations [1] - This technology shifts vector data (such as KVCache) from expensive DRAM and HBM memory to large-capacity, cost-effective SSDs, representing a strategic expansion of the storage layer rather than a simple replacement [1] - The core value of this approach lies in significantly reducing token latency, enhancing inference throughput, and optimizing end-to-end inference costs, providing a feasible path for large-scale AI deployment [1] - The AI storage revolution is underway, with "storage-based computing" creating core opportunities that substantially reduce computing power consumption and accelerate AI inference, leading to a demand growth for SSDs that exceeds traditional growth curves [1]

Core Viewpoint - The "storage instead of computing" paradigm emerges as a disruptive technology to overcome computing bottlenecks and storage limitations in AI inference, significantly reducing latency and costs while enhancing throughput [1][2]. Development Background - AI inference has become a key measure of the commercial value of large models, facing challenges such as slow processing and high costs. The "storage instead of computing" technology addresses these issues by migrating vector data from expensive DRAM and HBM to cost-effective SSDs, enabling strategic expansion of storage layers [1]. Core Technology - The "storage instead of computing" Cached Attention technology caches historical KVCache data across HBM, DRAM, and SSD, achieving an 87% reduction in first token latency and a 7.8-fold increase in prefill throughput, leading to a 70% decrease in end-to-end inference costs [2]. Hardware Breakthroughs - Under the "storage instead of computing" paradigm, SSDs evolve from mere data storage to core components in AI inference, requiring high capacity, throughput, and low latency. The AISSD technology will develop in three directions: transitioning to QLC particles, adopting PCIe 5.0/6.0 interfaces with NVMe protocols, and upgrading functionalities towards intelligent solutions [4]. Industry Layout - Major industry players are actively engaging in the core practices of "storage instead of computing," with companies like Huawei and Inspur optimizing storage architectures and cache management for efficient AI inference [5][6]. International Developments - Global giants such as Kioxia, Micron, and Solidigm are pushing for technological iterations and product innovations in AISSD, with QLC+PCIe/NVMe+CXL expected to form the foundation for the next generation of AISSD, transforming SSDs into long-term memory carriers for AI inference [10]. Investment Recommendations - The AI storage revolution is underway, with "storage instead of computing" creating significant opportunities. Companies in storage module manufacturing, storage chips, and distribution/testing are recommended for attention, including Jiangbolong, Demingli, and Zhaoyi Innovation [10].

Core Insights - The article discusses the explosive growth potential of AI storage, driven by technological breakthroughs such as Universal Storage architecture and hardware decoupling, which enable distributed storage to become the mainstream choice for data infrastructure in the AI era [1][2]. Group 1: Market Dynamics - Vast Data, a Silicon Valley AI storage company, is in talks for a new funding round with a valuation reaching $30 billion, reflecting a 3.3x increase in valuation over 18 months [2]. - The global data volume is expected to grow at a rate of 36% annually, reaching YB scale by 2030, highlighting the need for efficient and secure data storage solutions [2]. Group 2: Technological Evolution - The shift in focus for large models has transitioned from training to inference optimization, with the emergence of Agents facilitating complex decision-making and interaction [3][4]. - The core requirements for storage in AI have evolved to include extreme throughput, low latency, and high concurrency, necessitating a shift from traditional storage solutions [5]. Group 3: Storage Requirements - AI inference requires significant throughput, with multi-agent collaboration demanding up to 1TB/s aggregate bandwidth and sub-millisecond latency for optimal performance [5]. - The need for unified management of multimodal data and version traceability is critical, as AI applications increasingly rely on diverse data types [6]. Group 4: Architectural Innovations - The Universal Storage architecture aims to integrate various storage types (block, file, object) into a single platform, addressing the inefficiencies of traditional storage systems [23][26]. - Vast Data's approach eliminates data silos and migration overhead, allowing seamless access to data across different protocols, significantly simplifying management and reducing total cost of ownership [26][27]. Group 5: Future Outlook - The article emphasizes the investment value of Universal Storage technology, driven by the increasing demand for storage performance and scalability in the AI era [32][33]. - The potential for disruptive opportunities in the enterprise storage market is highlighted, as AI applications lead to a surge in data volume and storage needs [32].

Group 1: Macro Strategy and Market Overview - The A-share market saw significant gains, with the ChiNext Index rising by 8.58% and the CSI 500 and Shenzhen Component Index both increasing over 3.5% [1] - The central bank injected a net of 85.1 billion yuan into the market, maintaining stable liquidity [1][28] - The U.S. dollar index fell to 97.84, down 0.43% week-on-week, while the RMB remained stable at 7.19 [1][29] - The report anticipates continued policy stability and flexibility in the second half of the year, with a focus on gold and convertible bonds [1][30] Group 2: Fixed Income Market Insights - The current market shows a divergence between stock and bond performance, driven by policy expectations and market sentiment [2][31] - The central bank's timely interventions have provided support to the bond market, especially during periods of rising interest rates [2][33] - The report suggests that the 10-year government bond yield may reach a temporary peak around 1.80% [2][34] Group 3: Banking Sector Analysis - The banking sector is experiencing a trend of "credit pre-positioning," with a focus on early-year lending [4] - There is a notable divergence in credit growth between large state-owned banks and smaller banks, with the latter facing negative growth [4] - The report indicates that 2025 may see the smallest decline in loan rates since the LPR reform, with corporate and mortgage rates stabilizing around 3.2% and 3.1% respectively [4] Group 4: Cement Industry Overview - The necessity for "anti-involution" in the cement industry remains, with average prices down 43.7 yuan/ton year-on-year [7] - The previous supply-side reforms have led to a significant recovery in industry profits, with profits rising from 51.8 billion yuan in 2016 to 186.7 billion yuan in 2019 [7] - The report anticipates a continued decline in cement demand, with a potential drop of 18%-34% from 2024 levels [7] Group 5: Oil and Gas Sector Insights - The IEA has revised down its oil demand growth forecast by 350,000 barrels per day for the year, citing weak consumer confidence [8] - The IEA has increased its supply growth forecast for 2025 by 370,000 barrels per day, driven by OPEC's easing of production cuts [8] - Oil inventories have risen for five consecutive months, reaching a 46-month high of 783.6 million barrels [8] Group 6: Semiconductor Industry Trends - The "storage instead of computing" approach is expected to significantly enhance AI inference efficiency, driving rapid growth in SSD demand [17] - The semiconductor industry is experiencing stable growth in equipment and materials, with improved orders in wafer foundries and packaging [17] - The report maintains an optimistic outlook for global semiconductor growth driven by AI applications [17] Group 7: Home Appliance Sector Performance - Ecovacs reported a revenue of 8.68 billion yuan for H1 2025, a year-on-year increase of 24.4%, with a net profit of 980 million yuan, up 60.8% [35] - The company has seen strong growth in both domestic and overseas markets, particularly in Europe, where revenue increased by 66.6% [35][36] - The report highlights the company's strategic focus on optimizing its marketing investment model to improve profitability [36]

Investment Rating - Industry Rating: Outperform the Market (Maintained Rating) [1] Core Insights - The semiconductor industry is expected to continue its optimistic growth trajectory in 2025, driven by AI demand in downstream applications. The report highlights the ongoing push for domestic substitution and the positive performance forecasts for various segments in Q2 and Q3 [6][12]. - The introduction of Huawei's UCM technology is anticipated to significantly boost SSD demand by enhancing AI inference efficiency through innovative data management strategies [10][21]. Summary by Sections Semiconductor Market Overview - In July, major chip manufacturers experienced an increase in delivery times, with prices for storage, analog, power, and passive components showing signs of recovery. The report anticipates a continued upward trend in prices and improved order conditions in the distribution channel for Q3 [5][11][22]. Key Technologies and Innovations - The "Storage Computing" paradigm, initiated by Huawei's UCM, is expected to revolutionize the storage industry by optimizing AI inference processes and reducing reliance on HBM. This innovation is projected to enhance the efficiency of enterprise-level SSD products [10][17][21]. Market Segments and Recommendations - The report suggests focusing on sectors such as storage, power, foundry, ASIC, and SoC for their performance elasticity. It also emphasizes the importance of equipment materials and the domestic substitution of computing chips [6][12]. - Specific companies to watch include Jiangbolong, Zhaoyi Innovation, and Huahong Semiconductor among others, which are positioned to benefit from the anticipated market trends [7].

Investment Rating - The report rates the industry as "Better than the Trend" [7] Core Insights - The report emphasizes the importance of storage in enhancing the training and inference efficiency of large models, highlighting that storage optimization can significantly reduce training time and improve inference performance [3][17] - The shift towards inference as the core growth driver for computing power is noted, with increasing demand for diverse and long-context tasks [3][30] - Huawei's "Storage as Computation" approach is presented as a systematic solution to optimize performance through hardware and software integration [4][51] Summary by Sections 1. Storage Enhancements for Large Model Training and Inference - Storage plays a critical role in reducing data loading and checkpoint recovery times, potentially shortening training durations by 30% [18][21] - Inference performance can be significantly improved, with "Storage as Computation" reducing the first token latency by 90% and expanding context windows by over 10 times [24][27] 2. Transition to Inference-Centric Models - The report notes a surge in inference demand, with predictions indicating that by 2027, inference computing power will account for 70% of total demand [30][31] - The complexity of inference tasks is increasing, necessitating advanced storage solutions to manage longer contexts and higher concurrency [36][37] 3. Huawei's Systematic Approach - Huawei's AI SSDs are designed to handle both hot and cold data, with innovations in storage technology aimed at enhancing performance and capacity [4][52] - The UCM unified memory data manager is highlighted as a key component in optimizing inference efficiency [52] 4. Related Investment Opportunities - The report identifies several companies as potential investment targets, including Huawei storage agents and suppliers, as well as those involved in advanced packaging and SSD controller chips [5][6][4]