公众号记得加星标⭐️，第一时间看推送不会错过。 近日，闪迪发布了一篇博客文章，采访了韩国科学技术院（KAIST）电子电气工程系高带宽内存 （HBM）先驱金钟浩教授。KAIST是韩国国立研究型大学，金钟浩教授在HBM的研发中发挥了关键 作 用 。 他 目 前 致 力 于 高 带 宽 闪 存 （ HBF ） 技 术 的 研 究 ， 闪 迪 认 为 这 项 技 术 可 以 解 决 GPU HBM 的"墙"问题；即AI工作负载上下文内存（键值缓存）溢出HBM容量，导致向量重复计算耗时过长。 我们在去年底曾撰文讨论过这个问题，并指出其开发过程将十分复杂。作为HBM最大买家的英伟达 尚未公开表示对这项技术有任何兴趣。 此后，英伟达开发了上下文内存扩展技术（ICMSP），该技术利用与DPU连接的NVMe SSD来存储 来自HBM和GPU服务器DRAM的溢出键值缓存数据。ICMSP（推理上下文内存存储平台）的带宽和 延 迟 均 高 于 标 准 SSD ， 因 为 它 所 连 接 的 BlueField-4 DPU 本 身 就 是 一 个 存 储 加 速 器 ， 并 通 过 Spectrum-6以太网连接到Vera Rubin ...

Investment Rating - The industry investment rating is "Buy" [3] Core Insights - The report highlights the potential of HBF (High Bandwidth Flash) technology to meet the memory capacity requirements of AI large models, which have been increasing due to the growth in model parameters and context lengths [13][20] - Companies with a database technology foundation are well-positioned to develop data infrastructure software based on HBF, with major players including Huawei and Alibaba, as well as independent firms like StarRing Technology and PingCAP [29][32] - The maturity of HBF technology is expected to drive the application of related data infrastructure software, particularly in AI inference tasks [33] Summary by Sections HBF Technology Development - HBF is anticipated to become the optimal solution for meeting the memory capacity demands of AI large models, as current HBM (High Bandwidth Memory) solutions are insufficient [20][23] - The first batch of HBF samples is expected to be delivered in the second half of 2026, with integrated AI inference hardware projected for early 2027 [23] Data Infrastructure Software Opportunities - The report identifies a growing need for data processing software that can handle the high memory requirements of various AI models, particularly large language models and video generation models [28][30] - Companies with existing database technology capabilities are likely to develop software optimized for HBF, enhancing their competitive edge in the market [29][32] Key Company Valuations and Financial Analysis - StarRing Technology (688031.SH) has a current price of CNY 97.99 with a rating of "Increase" [8] - Cambricon (688256.SH) is rated "Buy" with a target value of CNY 1480.92, reflecting strong growth potential [8] - Inspur Information (000977.SZ) is rated "Increase" with a current price of CNY 62.85 [8] - Unisplendour (000938.SZ) is rated "Buy" with a target value of CNY 39.18 [8]

SanDisk FY Conference Summary Company Overview - **Company**: SanDisk (NasdaqGS: SNDK) - **Event**: Barclays Global Tech Conference - **Date**: December 10, 2025 Key Industry Insights - **Market Dynamics**: The NAND market is experiencing a significant restructuring post-2023 downturn, with a shift from a focus on price to a focus on supply. Demand is currently prioritized over pricing, indicating a dynamic market environment [6][10][11]. - **Data Center Growth**: The data center market is projected to become the largest consumer of NAND by 2026, surpassing mobile demand, which has been the largest for the past 15 years. This shift is expected to drive substantial growth in NAND consumption [12][18][19]. - **Long-Term Agreements (LTAs)**: There is an increasing interest from major customers in securing long-term agreements to ensure supply stability, reflecting the structural importance of NAND in their business models [25][27]. Financial Performance and Projections - **Gross Margin Outlook**: SanDisk aims for a through-cycle gross margin of 35%, but acknowledges the need to exceed this target to deliver returns for investors. The company has experienced three consecutive quarters below this margin [10][13]. - **Capital Investment Strategy**: The company is committed to making long-term capital investments in fabs and R&D, with a focus on sustainable growth rather than short-term gains. The industry is expected to grow at a mid-teens percentage rate, with SanDisk planning to align its capacity with this growth [17][26][29]. Technology and Product Development - **BiCS8 Transition**: SanDisk is on track with the transition to BiCS8 technology, which is expected to constitute 40-50% of its portfolio by the end of the fiscal year. This technology is crucial for maintaining competitive advantage in the NAND market [34][37]. - **HBF Technology**: The company is developing HBF technology, which aims to address the growing demand for higher bandwidth in AI applications. The first memory die is expected to be available in late 2026, with a controller following in early 2027 [38][42][43]. Market Challenges and Considerations - **Supply Constraints**: There are concerns about potential supply limitations for PCs and smartphones due to the prioritization of data center demands. SanDisk is committed to maintaining a balanced approach across its three key markets: consumer, PC, and data center [32][33]. - **Industry Evolution**: The NAND market is undergoing profound changes, with new demand drivers emerging. The company is focused on understanding these dynamics and adapting its strategies accordingly [24][44]. Conclusion - SanDisk is navigating a transformative period in the NAND industry, characterized by shifting demand dynamics, a focus on long-term supply agreements, and significant technological advancements. The company remains committed to prudent capital management while positioning itself for future growth in a rapidly evolving market landscape [44].

Investment Rating - The report maintains an "Overweight" rating for the computer industry [1] Core Insights - The emergence of AI inference necessitates a new storage architecture centered around GPU directly connected to SSD, which will replace the CPU-dominated era [9][14] - The shift from CPU-centric to GPU-centric architecture will drive significant changes in database software design, optimizing for GPU's data processing capabilities [18][19] - The industry is witnessing accelerated advancements in both hardware and software, with notable collaborations and innovations aimed at enhancing performance for AI workloads [22][24] Summary by Sections AI Inference Era and New Storage Architecture - AI inference requires different I/O demands compared to training, with a focus on small data blocks and high concurrency, leading to the need for a new storage architecture [9][10] - The traditional CPU-centric data loading architecture is becoming a bottleneck for AI workloads, necessitating a shift to GPU as the primary controller for data access [11][14] Changes in Database Architecture - The transition to GPU-centric architecture will require a complete redesign of database software, with GPU taking on the role of the main computing unit [18][19] - Key components such as storage engines and query execution engines will need to be restructured to optimize for GPU capabilities and direct SSD connections [19][21] Industry Progress - Hardware advancements include the development of High Bandwidth Flash (HBF) technology, which is expected to play a crucial role in the future of AI storage solutions [22] - Collaborations between companies like SanDisk and SK Hynix aim to standardize HBF technology, with initial products expected by 2027 [22] - Software improvements are being made to enhance data orchestration and performance, such as Hammerspace's advancements in metadata reading and Cloudian HyperStore's object storage capabilities [24] Investment Recommendations - The report suggests that as AI inference grows, the importance of GPU will increase, leading to new opportunities in the database industry as software architectures adapt to these changes [25][26]

Core Insights - The article discusses the trends and predictions for the semiconductor and storage industries, emphasizing the impact of AI on market dynamics and technology advancements [4][10][17]. Group 1: AI and Semiconductor Industry Trends - The wafer foundry industry is expected to grow by 19% in revenue in 2026, with AI-related demand driving advanced process markets to a remarkable 28% growth [10]. - TSMC has begun production using 2nm technology and plans to advance to 1nm processes, indicating a significant technological shift in the semiconductor sector [10]. - The demand for AI chips is expected to surge, with major cloud providers launching their own AI chips, highlighting the competitive landscape in the semiconductor market [10]. Group 2: AR Glasses and AI Integration - AI and AR glasses are forming a symbiotic relationship, enhancing user interaction and accelerating data accumulation for large language models [12]. - The global shipment of AR glasses is projected to exceed 10 million units by 2030, driven by major brands like Google and Apple entering the market [13]. - China plays a crucial role in the AR glasses market, with significant contributions in manufacturing and supply chain integration [14]. Group 3: Memory Market Dynamics - AI servers and general servers are driving a new super cycle in the memory market, with AI applications expected to account for 66% of DRAM capacity by 2026 [17]. - The DRAM market is anticipated to face severe shortages, with prices expected to rise significantly, leading to a projected 56% increase in DRAM revenue in 2026 [17]. - The competition for limited DRAM capacity is intensifying, particularly as AI server demands begin to outstrip supply for consumer devices [17]. Group 4: Server Market Outlook - Global server shipments are expected to grow by over 9% in 2026, with AI servers leading this growth at over 20% [20]. - The competition in the AI server market is intensifying, with major players like NVIDIA and AMD dominating the GPU AI market while Chinese companies pursue self-developed ASIC solutions [20]. Group 5: Power Semiconductor Transformation - The demand for power semiconductors is shifting due to AI, with SiC and GaN technologies becoming critical for high-voltage power supply architectures [22]. - SiC technology is establishing a leadership position in high-voltage applications, while GaN is entering a rapid growth phase across multiple applications, including AI data centers [22][24]. Group 6: NAND Flash Market Innovations - The AI boom is creating storage bottlenecks, leading to a shift towards high-density QLC eSSD due to limited supply in traditional HDD markets [27]. - New NAND Flash technologies are emerging, such as HBF and AI SSD, which aim to enhance performance and address storage challenges in AI applications [27]. - The NAND Flash industry is expected to thrive amid ongoing supply shortages, with innovations reshaping its value proposition [27].

Core Viewpoint - The article discusses the future of high bandwidth memory (HBM) and high bandwidth flash (HBF) technologies, emphasizing their importance in the AI era and the anticipated market dynamics surrounding these technologies by 2030 [2][3]. Group 1: HBM and HBF Market Predictions - The demand for HBM is expected to increase significantly, with projections indicating that the amount of training data for AI will grow a thousandfold in the next decade, leading to a corresponding increase in HBM demand [4]. - Intel's CEO has committed to supplying 260,000 GPUs to South Korea, which will include 2.08 million HBM memory modules, highlighting the scale of investment in this sector [4]. - The total budget for cooling, power facilities, land, and other components is estimated to be around 100 trillion KRW, with the final GPU count potentially exceeding 100 million, indicating a fierce competition dubbed the "money war" [5]. Group 2: Strategic Collaborations and Supply Chain - CEO Huang Renxun's meetings with SK Hynix and Samsung Electronics are aimed at securing a stable supply of HBM, as the demand for HBM memory is expected to outstrip supply from any single manufacturer [6]. - The collaboration between SanDisk, SK Hynix, and Samsung Electronics is crucial for advancing HBF technology, which is seen as a necessary evolution from traditional DRAM due to physical limitations in memory capacity [7]. Group 3: Future Development and Research Directions - The article outlines plans for the HBF roadmap to be released in early next year, with expectations that memory revenue will surpass GPU revenue by the mid-2030s [7][8]. - The establishment of research centers focused on HBM and HBF technologies is proposed, with a call for government support to foster talent and foundational research in this field [9]. Group 4: Vision for AI and Memory Integration - The integration of HBM and HBF alongside GPUs is proposed, where HBM handles encoding tasks and HBF manages decoding tasks, reflecting a shift towards memory-centric computing [7]. - The future vision includes HBM evolving into a system semiconductor, with diverse customer needs leading to the development of AI wafer foundry and AI solution businesses [8].

Group 1 - The importance of NAND products in the AI memory market is rapidly increasing, with demand for data center NAND flash expected to surpass mobile applications for the first time next year [2] - Major NAND suppliers, including Samsung and SK Hynix, have reported that customer demand will exceed production capacity, leading to a significant supply shortage [2] - The price of NAND flash products has risen sharply, with the fixed trading price of 128Gb MLC NAND flash increasing by 10.6% in September and 14.9% in October, marking the largest increase in a decade [2] Group 2 - The next-generation memory "HBF" is being developed to combine the advantages of high bandwidth memory (HBM) with NAND characteristics, aiming for mass production in 2027 [3]

Group 1 - The core viewpoint of the articles highlights the significant price increase in NAND flash memory contracts by SanDisk, driven by surging demand from AI data centers and severe wafer supply constraints, with a price hike of up to 50% [1] - Major storage companies like SanDisk, Seagate, and Micron Technology are reaching new highs, with SanDisk rising by 12% on November 10, and Nvidia, a leader in AI chips, increasing by 5.8%, marking its largest single-day gain since April [1] - The semiconductor storage industry is entering a structural prosperity phase, with rapid advancements in high-bandwidth memory (HBM) technologies, leading to suppliers gaining pricing power due to tight supply of traditional DRAM and NAND [1] Group 2 - The management fee rate for the Science and Technology Innovation Index ETF by Jianxin (589880) is 0.15%, and the custody fee rate is 0.05%, resulting in a comprehensive fee rate of 0.2%, which is considered low in the industry [2] - The Jianxin Science and Technology Innovation Index ETF (589880) serves as a core tool for capturing China's technological innovation and industrial upgrading, thanks to its tracking of the Science and Technology Innovation Board Composite Index and coverage of hard technology sectors [2] Group 3 - Related products include Jianxin Science and Technology Innovation Index ETF (589880), Jianxin Shanghai Stock Exchange Science and Technology Innovation Board Composite Connection A (023743), and Jianxin Shanghai Stock Exchange Science and Technology Innovation Board Composite Connection C (023744) [4]

Core Insights - The storage chip industry is entering a "super cycle" driven by the unprecedented demand for AI computing power, particularly for large models and high bandwidth requirements, leading to a fundamental restructuring of the industry logic and cycle characteristics [2][3][17] - The humanoid robot industry is accelerating its industrialization process, with significant developments from companies like Tesla, XPeng, and Leju Intelligent, showcasing advancements in AI and robotics integration [2][3][41] Storage Chip Industry Trends - The storage chip sector is experiencing a structural prosperity period, with rapid advancements in high bandwidth memory (HBM) and high bandwidth flash (HBF) technologies, leading to a supply-demand imbalance and rising prices [3][16] - Major players like SK Hynix and Samsung Electronics are strengthening their profitability in the HBM market, which is driving growth in upstream equipment, materials, and packaging manufacturers, as well as downstream AI servers and cloud computing [3][19] - The current cycle is characterized by a shift from traditional inventory management to a demand-driven model, with AI's insatiable need for computing power redefining the value of memory [17][19] Humanoid Robot Industry Trends - The humanoid robot industry can be divided into three main segments: "brain," "body," and "system integration," with the core value concentrated in the midstream components [3][59] - Companies like Tesla are planning large-scale production of humanoid robots, with Tesla's Optimus expected to have a market capacity of billions of units, indicating a significant shift towards AI-driven manufacturing platforms [41][44] - XPeng's new humanoid robot, IRON, showcases advanced features and aims for mass production by the end of 2026, highlighting the competitive landscape in the humanoid robotics sector [50][52] Investment Recommendations - The report suggests focusing on sectors with marginal improvements, including AI applications, AI hardware, power equipment, non-ferrous metals, and innovative pharmaceuticals in the short term [4] - In the medium to long term, it emphasizes the importance of technological cycles, supply-demand dynamics, and the progress of intelligent infrastructure and AI ecosystems [5]

Investment Rating - The storage industry is expected to enter a "prosperity cycle" with a recommendation to focus on specific companies such as Demingli, Jiangbolong, and Zhaoyi Innovation [4][72][74]. Core Insights - The demand for storage is driven by AI, with a significant shift from HDD to SSD/DRAM due to the increasing data volume and the need for faster access [2][18][72]. - The supply side is innovating with CBA and HBF technologies to overcome the limitations of traditional memory, enhancing storage density and performance [3][40][52]. - The semiconductor equipment market is projected to benefit from increased capital expenditure by storage manufacturers, with NAND equipment market size expected to reach $13.7 billion in 2025 [4][61][72]. Summary by Sections Storage Cycle - The storage industry exhibits clear cyclical characteristics, with the current cycle driven by AI demand and supply optimization from major manufacturers [9][12]. Demand Side - The transition from cold data to warm data is accelerating the replacement of HDDs with SSDs, as AI applications require more frequent data access [2.2][28]. - The data generated by AI applications is expected to grow exponentially, with projections indicating a shift from megabytes to exabytes and zettabytes [2.1][21]. Supply Side - CBA technology is crucial for achieving high density and performance in storage solutions, with applications in next-generation DRAM and NAND technologies [3.1][40]. - HBF technology offers significant advantages in capacity and energy efficiency, positioning it as a key solution for AI applications [3.2][52]. Semiconductor Equipment - The semiconductor equipment market is anticipated to grow significantly, with NAND equipment sales projected to increase by 42.5% in 2025 [4.1][61]. - Innovations in etching and deposition equipment are essential for advancing storage density and performance [4.2][68]. Investment Recommendations - The report suggests focusing on companies that are well-positioned to benefit from the storage industry's growth, including Demingli, Jiangbolong, and Zhaoyi Innovation [4][72][74].