据产业链调研反馈，谷歌TPUv8p有望引入内存池化，用以拓展内存容量，该方案下，每张TPU有望额外增加数百 GB的内存拓展；内存池化有望逐步迎来规模商业化、放量可期。 【长江电子】#澜起科技第四call：在产业趋势上做配置 #rambus自身因封测品质因素指引不及预期、但不改产业趋势 rambus因自身封测品质因素影响一季度产品交付，Q1业绩指引不及预期至盘后大跌。 但公司#对服务器CPU内存条未来增长指引、MRDIMM等产品放量均给予正面回应、产业趋势持续向上。 #谷歌TPUv8p或引入内存池化、CXL放量可期 据产业链调 【长江电子】#澜起科技第四call：在产业趋势上做配置 #rambus自身因封测品质因素指引不及预期、但不改产业趋势 rambus因自身封测品质因素影响一季度产品交付，Q1业绩指引不及预期至盘后大跌。 但公司#对服务器CPU内存条未来增长指引、MRDIMM等产品放量均给予正面回应、产业趋势持续向上。 #谷歌TPUv8p或引入内存池化、CXL放量可期 伴随着CXL拓展芯片、MRCD/MDB、PCIe Retimer、PCIe Switch的规模起量；我们认为公司远期利润有望达100- 15 ...

Rambus确认行业高景气，自身交付问题利好澜起科技 Rambus确认行业高景气，自身交付问题利好澜起科技 1. 产品营收同比增长32%至9700万美元，全年增长41%至3.5美元，因为封测品质问题一季度营收影响2000-3000万 美元（lowdouble-digitmillion）。 2. 美国投资者质疑，Rambus份额提升至45%的说法，并担心本次品质问题影响产品声誉。 3. 服务器需求稳固，处于更新换代周期，AI智能体和推理应用也推动需求，单通道的平均内存模组数量在增长。 4. MRDIMM将在26年底左右开始量产，产生初步贡献，主要贡献将在27年。 1. 产品营收同比增长32%至9700万美元，全年增长41%至3.5美元，因为封测品质问题一季度营收影响2000-3000万 美元（lowdouble-digitmillion）。 2. 美国投资者质疑，Rambus份额提升至45%的说法，并担心本次品质问题影响产品声誉。 3. 服务器需求稳固，处于更新换代周期，AI智能体和 ...

Core Viewpoint - MRDIMM is a new memory module architecture designed for high-performance computing and data-intensive applications, achieving a data transfer rate of 8800 MT/s, which is nearly a 40% increase in peak bandwidth compared to standard DDR5 RDIMM [1][2] Group 1: MRDIMM Technology Overview - MRDIMM is suitable for high-performance computing and artificial intelligence applications, addressing the bandwidth decline per core due to the increasing number of processor cores [2] - The first generation of MRDIMM supports a data transfer rate of 8800 MT/s, with the third generation expected to reach 14000 MT/s [2] - The architecture of MRDIMM includes 1 MRCD (Memory Register Clock Driver) and 10 MDB (Memory Data Buffer) chips, indicating significant growth potential in the market [3] Group 2: Market Demand Projections - By 2030, the demand for MRCD and MDB is projected to increase significantly, with estimates under pessimistic, neutral, and optimistic scenarios suggesting demand for MRCD could reach 274 million, 456 million, and 593 million units respectively, while MDB demand could reach 2.736 billion, 4.560 billion, and 5.928 billion units [4] - The penetration rates for MRDIMM are estimated to be 30%, 50%, and 65% under pessimistic, neutral, and optimistic scenarios respectively [4] Group 3: Industry Implications - The adoption of MRDIMM technology is expected to drive demand for MRCD and MDB chips, particularly as it becomes compatible with Intel's sixth-generation CPUs [1][3] - Key companies in the market include 澜起科技 (Lanke Technology) and 聚辰股份 (Juchen Technology), which are involved in the production of MRCD/MDB chips and DDR5 memory modules respectively [1]

C套内存模组向 MRDIMM MRCD RHS MDB 麻 . Peter The new State of the research and a line of the state of the 行业专题报告 / 2026.02.01 ■ 证券研究报告 因 投资评级:看好(维持) 最近 12 月市场表现 359 分析师 唐佳 SAC 证书编号: S0160525110002 tangjia@ctsec.com 联系人 周勃宇 zhouby@ctsec.com 相关报告 1.《招股说明书梳理系列(一):盛合晶微》 2026-01-29 2.《先进封装涨价与扩产共振，强周期与成 长共舞》 2026-01-29 3. 《AI Agent 沙箱化有望带来 CPU 新增 量空间》 2026-01-25 看好 MRCD 芯片与 MDB 芯片环节 核心观点 服务器级 CPU 配套内存模组向 MRDIMM 发展:服务器中 CPU 与内 * 存模组是两大核心部件,内存模组主要作为 CPU 与硬盘的数据中转站,用于 临时存储数据。目前服务器中三种主流的内存模组技术分别为 UDIMM、 RDIMM、LRDIMM，其中 UDIMM ...

Macroeconomic and International - The market expects the Federal Reserve to maintain interest rates unchanged before Powell's term ends in May [1] - Trump has softened his stance on Greenland, while U.S. Treasury yields continue to decline [2] - The U.S. is increasing military presence in the Middle East, indicating ongoing geopolitical risks [3] - European pension funds (Denmark, Sweden) continue to reduce their holdings in U.S. Treasuries [4] - The U.S. proposed more trade talks with China before Trump's visit in April [5] Semiconductor - Long-term supply contracts (LTA) for memory products from companies like Winbond and Nanya have been extended to over two years, with some frameworks extending to 2030, reflecting optimism about long-term industry prospects [5] - Sellers indicate that Agentic AI will drive demand for high-bandwidth memory such as MRDIMM [6] - Overseas manufacturers are raising prices for analog chips, driven by structural shortages and growth in AI servers [7] - Related ETFs include Semiconductor Equipment ETF (on-market: 561980, off-market: 020464) [7] Artificial Intelligence - Nvidia's CEO Jensen Huang plans to visit China in late January to seek to reopen the AI chip market; OpenAI is reportedly set to launch ChatGPT advertisements in early February [8] - China has initiated a national AI industry investment fund with a scale of 60 billion yuan [9] - Sellers estimate that the demand for CPUs from active AI agents will grow exponentially [10] - Tengjing Technology has signed a large order worth nearly 90 million yuan [11] - Related ETFs include Cloud Computing ETF (on-market: 159890, off-market: 021716), Software Leaders ETF (on-market: 159899, off-market: 018385), TMT50 ETF (on-market: 159909, off-market: 004409) [11] Autonomous Driving - Guangdong has released policies to support the orderly expansion of high-level autonomous driving road testing and application areas [12] Solid-State Batteries - The third China All-Solid-State Battery Innovation Development Summit Forum will be held on February 7-8 to assess industry strategic directions and technological paths [13] Consumer Policy - The Ministry of Finance has announced the establishment of duty-free shops at 41 ports, including Wuhan Tianhe Airport, to boost consumption [14] Market Dynamics and Funds - The CSI 300 ETF faced significant selling pressure towards the end of trading [15] - Sellers report that Huijin's holdings in the Sci-Tech 50 ETF have decreased, while significant holdings remain in the CSI 300 ETF [15] - Abnormal trading regulations may tighten further, potentially limiting buying permissions for volatile stocks [15] - A Bank of America survey shows that global fund managers' cash holdings have reached a historic low, with risk appetite at a high level, but the bull-bear indicator has entered the "sell" zone [15] - Related ETFs include A500 Index ETF (on-market: 560610, off-market: 022455), Sci-Tech 50 ETF Enhanced (on-market: 588450), CSI 300 ETF (on-market: 561930, off-market: 022504) [15] Strategy Observation - On Wednesday, trading volume was 2.6 trillion yuan, indicating further contraction [15] - Despite a decline in financing data and external market downturns, the market attempted to rally in the morning but fell back in the afternoon under pressure from heavyweight stocks, showing a clear oscillation around the 4100-point mark [15] - The market structure is diverging, with trend-driven sectors led by institutions performing prominently [15] - Sectors such as non-ferrous metals (commodity bull market), electronics (semiconductor packaging, CPU, PCB), and machinery (AI equipment, robotics) are leading the gains [15] - Current trends include semiconductor (especially Sci-Tech attributes), AI hardware, robotics, and chemical cycles, with a relatively healthy rhythm [15] - The CPO sector (especially small and mid-cap stocks) is breaking through following overseas trends; the logic of rising prices for storage and CPUs continues to strengthen [15] - The satellite internet sector has entered a left-side observation interval [15] - The overall market is showing characteristics of "stable index, active structure" under regulatory guidance, focusing on industry prosperity and performance trends [15]

Core Insights - The supply of NAND flash wafers is tightening, with some products seeing contract price increases of over 60% in November, significantly raising the material costs for consumer electronics and leading to reduced shipments of smartphones, laptops, and gaming consoles [1] - The demand for NAND flash is being driven by AI infrastructure, with the total revenue of the top five NAND flash suppliers increasing by 16.5% quarter-over-quarter, approaching $17.1 billion [1] - SK Hynix has announced a strategic vision to become a "full-line AI storage creator" at the SKAI Summit 2025, detailing a product roadmap from 2026 to 2031 [1] Group 1: SK Hynix's Ambitions in AI - SK Hynix is focusing on three main areas: Custom HBM, AI DRAM, and AI NAND [3] - In the HBM segment, SK Hynix plans to launch HBM4 16-layer stacked products and custom HBM solutions starting with HBM4E, which will significantly reduce interface power consumption [3][4] - The company aims to fully enter the HBM5 generation by 2029-2031, with plans for next-generation GDDR7, DDR6, and 3D DRAM, as well as NAND products with over 400 layers [3] Group 2: Custom HBM Developments - SK Hynix is supplying HBM for Google's seventh-generation TPU, with a projected market share of 56.6% in 2025 [6] - The shift towards custom HBM is seen as a key competitive factor, with major tech companies like Apple, Microsoft, Google, Amazon, Nvidia, Meta, and Tesla requesting custom HBM solutions [8] - The market is transitioning from general-purpose products to customized solutions based on client specifications, starting with HBM4e [9] Group 3: AI DRAM Innovations - Capital expenditure for DRAM is expected to reach $53.7 billion in 2025, growing to $61.3 billion in 2026, a year-on-year increase of approximately 14% [12] - SK Hynix is focusing on three directions for AI DRAM: optimization, breakthrough technologies to overcome the "memory wall," and expansion into new use cases beyond data centers [16][19] - The MRDIMM technology is highlighted for its efficiency improvements, with performance metrics showing significant gains over traditional DRAM solutions [15] Group 4: AI NAND Strategies - Capital expenditure for NAND Flash is projected at $21.1 billion in 2025, with a slight increase to $22.2 billion in 2026, reflecting a year-on-year growth of about 5% [19] - SK Hynix is developing AI NAND solutions focused on performance, bandwidth, and density, with plans to introduce new architectures and technologies by 2026 [19][20] - The collaboration with SanDisk on HBF technology aims to enhance bandwidth and meet the growing demands for high-performance storage solutions [20] Group 5: Market Trends and Future Outlook - The storage market is entering a new upward cycle, with recent price increases for new devices ranging from 100 to 300 yuan, indicating a trend of further price hikes [21] - The structural and long-term shortages driven by AI applications are exceeding expectations, necessitating a rapid upgrade in storage capabilities to keep pace with computational advancements [21][22] - The integration of storage and AI is expected to deepen, with increasing demands for bandwidth and capacity across various applications [22]

Core Insights - The report from GF Securities highlights the significant growth potential in the AI storage sector driven by advancements in AI models and capital expenditures (CAPEX) [1] - AI-driven storage prices are on the rise, leading to improved gross margins for manufacturers, with DRAM and NAND architecture upgrades creating new equipment demand [1] - The storage foundry model is poised for transformative opportunities within the industry, with new interface chips like MRDIMM and VPD opening up additional market space [1] Group 1: AI and Storage Growth - Storage serves as the "tokens" for AI, with inference driving rapid growth in AI storage [2] - AI servers utilize storage types such as HBM, DRAM, and SSD, characterized by decreasing performance, increasing capacity, and lowering costs [2] - The demand for storage capacity is expected to surge to hundreds of exabytes (EB) due to the rapid growth in AI inference requirements [2] Group 2: Market Opportunities - The eSSD market for AI and storage servers is expanding due to the increasing demand for high bandwidth and large capacity driven by long-context inference and RAG databases [2] - MRDIMM is anticipated to enhance large model inference by providing deterministic gains and optimizing CPU-GPU memory orchestration [2] - The DDR5 SPD market is rapidly developing, driven by higher specifications and prices compared to the DDR4 generation, while SSD upgrades are expected to create growth opportunities for VPD [2] Group 3: CXL Technology Impact - CXL technology facilitates storage pooling, significantly enhancing computational efficiency and providing a total cost of ownership (TCO) advantage in KVCache-intensive inference [3] - Companies like NVIDIA and Alibaba Cloud are investing in CXL capabilities to improve inference throughput and performance [3] - The CXL protocol is increasingly penetrating the AI sector, driving demand for chips and positioning CXL interconnect technology as a key player in AI applications [3]

Investment Rating - The industry investment rating is "Buy" with a previous rating of "Buy" as well [2]. Core Viewpoints - The storage sector is crucial for AI inference, driving rapid growth in storage demand, particularly for HBM, DRAM, and SSD, characterized by decreasing costs and increasing capacities [5][13]. - AI-driven storage demand is expected to surge, with projections indicating a need for hundreds of exabytes (EB) of storage capacity in the near future [5][24]. - The report emphasizes the broad space within the industry chain, highlighting opportunities in eSSD, MRDIMM, SPD, and VPD chips, as well as CXL storage pooling [5][79]. Summary by Sections 1. Storage as Tokens for AI Inference - AI servers utilize various storage types, including HBM, DRAM, and SSD, with a focus on high bandwidth and large capacity to support efficient data processing [13][17]. - The demand for SSD and HDD is projected to grow significantly, with estimates suggesting a requirement of 49 EB for ten Google-level inference applications by 2026 [24]. 2. AI-Driven Storage Demand Growth - eSSD is identified as a core demand area for AI and storage servers, with increasing needs for high bandwidth and large capacity due to long-context inference and RAG databases [25][26]. - The market for AI server eSSD is expected to expand, with theoretical maximum capacities of 59 EB, 89 EB, and 120 EB for 2024, 2025, and 2026 respectively [27][34]. 3. MRDIMM Applications - MRDIMM is anticipated to enhance performance in large model inference, providing significant bandwidth improvements and capacity expansions [38][39]. 4. SPD and VPD Chip Opportunities - The transition to DDR5 memory modules presents growth opportunities for SPD and VPD chips, driven by increased specifications and demand [45][46]. 5. CXL Storage Pooling - CXL technology facilitates storage pooling, enhancing computational efficiency and enabling better resource allocation for AI applications [53][54]. - The report notes significant TCO advantages in KV Cache performance when utilizing CXL in high-concurrency, long-context workloads [56][59]. 6. Investment Recommendations - The report suggests focusing on storage industry chain-related entities, as AI-driven storage prices are expected to rise, leading to improved profit margins for manufacturers [79].

Rambus FY Conference Summary Company Overview - Rambus has over 35 years of experience in high-performance memory subsystems, providing leading ICs and Silicon IP solutions that enhance data center connectivity, addressing the bottleneck between memory and processing [3][4] Financial Performance - Patent licensing business generates approximately $210 million annually with a 100% margin, stable but not expected to grow long-term [4] - Silicon IP business generated $120 million last year, growing at 10%-15% annually [5] - ICs business projected to reach about $340 million this year, with a growth rate of 40% year-over-year, driven by data center demand [5] Market Dynamics - The market for interface chips is estimated at $800 million annually, with additional opportunities from companion chips adding $600 million, and further expansion into high-end client systems adding another $200 million [6][7] - Transition to DDR5 technology has created new chip opportunities, increasing the total addressable market (TAM) from $800 million to $1.4 billion [9] AI and Server Market - AI servers are driving demand for traditional servers, as they require both AI and traditional processing capabilities [10] - AI inference is expected to be a significant growth driver, as it is more cost-effective and simpler than AI training [11] MRDIMM Technology - MRDIMM technology doubles memory capacity and bandwidth on existing infrastructure, significantly increasing Rambus's content opportunity [12][14] - Expected rollout linked to next-generation platforms from Intel and AMD by late 2026 to early 2027 [15] CXL Opportunities - Rambus has a CXL offering as part of its Silicon IP business, but the market is fragmented, and the company sees MRDIMM as a more elegant solution for memory expansion [16][17] Silicon IP Business Strategy - Focused on security and high-speed interfaces, with a projected growth of 10%-15% annually [20][21] - Minimal exposure to China, with less than 5% of business from that market [22] Patent Licensing Insights - Patent licensing provides a stable revenue stream and insights into future technology trends, with contracts typically lasting 3 to 10 years [23][24] Financial Model and Capital Allocation - Patent licensing has a 100% gross margin, Silicon IP at 95%, and product business between 61%-63% [28] - Rambus aims to return 40%-50% of free cash flow to investors, having generated $300 million in cash from operations over the last 12 months [30] Competitive Landscape - Rambus maintains a strong position in hardware-based security against fast followers and internally developed solutions [34][35] - The company is developing quantum-safe security solutions in anticipation of future challenges posed by quantum computing [35] Conclusion - Rambus is well-positioned for growth in the evolving data center and AI markets, leveraging its strong patent portfolio, innovative technologies, and strategic focus on high-performance memory solutions [1][2]

Group 1: Market Trends and Pricing - The demand for CSP in high-performance TLC eSSD and QLC eSSD is expected to rise due to AI applications, while HDD supply remains tight, supporting a continued increase in NAND Flash prices in Q4 [2][3] - Major manufacturers are maintaining a cautious capacity expansion strategy, which may limit the contribution to bit output growth in 2026 due to the lag in capacity construction cycles [2] Group 2: Supply Chain and Partnerships - The company has established long-term direct cooperation with major global storage wafer manufacturers, providing a competitive edge in the supply chain [3] - Supplier information is disclosed in public documents such as the prospectus [3] Group 3: Technology and Product Development - The company's self-developed controller chip technology offers significant performance and power consumption advantages, utilizing advanced foundry processes and proprietary firmware algorithms [3] - The company has launched several cutting-edge high-performance storage products, including MRDIMM and CXL2.0 memory expansion modules, and is focused on market demand for future product development [3] Group 4: Clientele and Market Engagement - The company actively participates in technology and new product bidding for major clients, with enterprise storage products integrated into the supply chains of leading internet companies, including telecom operators and server manufacturers [3] Group 5: Disclosure Compliance - The activity does not involve any undisclosed significant information [3]