NV Rubin 新架构&Agent 存储最强方向 - GPU Native 数据库【星环科技】20260106 摘要 英伟达推出 BlueField 4 与 Context Memory 平台，通过 DPU 连接 150TB 上下文内存，为每个 GPU 提供高速访问 16TB 分布式存储，显 著提升 AI 推理速度，解决 HBM 容量瓶颈及远程访问时延问题。 英伟达发布 NVIDIA Demo 和 NXL 等软件组件，推动软件生态向 GPU 为核心的新型存储架构迁移。VAST Data 和 Wakanda 等厂商正将软件 从 CPU x86 架构迁移至 GPU 中心架构，以提升推理速度并降低延迟。 传统基于 CPU 的数据库软件面临重构，需开发 GPU native 数据库，进 行引擎变化、算子优化和寻址宽度增加等改造，以适应新型硬件环境， 实现更高效的数据处理。 英伟达推动新型存储架构旨在应对 AI 推理阶段高频次、小数据块的数据 访问需求，传统 CPU 调度效率低下，GPU 直连 SSD 或 HBM 成为趋势， 需软硬件协同优化。 新型 GPU 中心系统中，数据库将深度重构，从 AI Ready 演变为 ...

| 指数 | | 收盘 | 涨跌幅% | | --- | --- | --- | --- | | 上证指数 | | 3,968.84 | 0.09 | | 深证成指 | | 13,525.02 | -0.58 | | 沪深 | 300 | 4,629.94 | -0.46 | | 中小板指 | | 8,263.86 | -0.46 | | 创业板指 | | 3,203.17 | -1.23 | | 科创 | 50 | 1,344.20 | -1.15 | 资料来源：常闻 研究早观点 2026 年 1 月 5 日 星期一 资料来源：常闻 国内市场主要指数 分析师： 彭皓辰 执业登记编码：S0760525060001 邮箱：penghaochen@sxzq.com 市场走势 【今日要点】 【行业评论】通信：周跟踪（20251222-20251228）-英伟达投资 Groq 重点布局推理 ASIC，科创板促进商业航天加速发展 【行业评论】化学原料：新材料周报（251222-1226）-政策赋能商业航 天高速发展，碳纤维底部确认景气有望回升 【行业评论】2026 年元旦小长假旅游市场整体表现 【行业评论】PDE4B、 ...

Investment Rating - The report maintains an "Outperform" rating for the communication industry, indicating expected growth exceeding the benchmark index by over 10% [1][40]. Core Insights - Nvidia's acquisition of Groq for $20 billion highlights the rapid growth in AI inference demand, driven by low-latency inference technology [2][14]. - Groq's revenue is projected to reach $500 million in 2025, with estimates of $1.2 billion and $1.9 billion for 2026 and 2027 respectively, and a recent valuation of $6.9 billion [3][14]. - The report emphasizes the importance of deterministic computing and networking in Groq's technology, which will synergize with Nvidia's capabilities, enhancing the demand for high-speed interconnects and optical modules [4][15]. Summary by Sections Industry Trends - Nvidia's acquisition of Groq signifies a strategic move towards enhancing AI inference capabilities, with Groq's unique chip innovations expected to drive market growth [2][14]. - The report notes that the optimal architecture for inference is still evolving, providing opportunities for both GPGPU and ASIC markets, which is encouraging for domestic companies entering the market [4][15]. Market Performance - The overall market saw an increase during the week of December 22-26, 2025, with the Shenwan Communication Index rising by 4.07% and the ChiNext Index by 3.90% [8][18]. - The satellite communication sector led the gains with a weekly increase of 29.13%, followed by liquid cooling and connectors [8][18]. Company Recommendations - The report suggests focusing on companies within the Nvidia ecosystem, including Zhongji Xuchuang, Xinyi Sheng, and Tianfu Communication, as well as those in the commercial aerospace sector like Shanghai Hanyun and XinKe Mobile [6][18].

Investment Rating - The report assigns a rating of "Buy" to the company with a target price of 103.01 CNY, compared to the current price of 77.43 CNY [5]. Core Insights - The company aims to create a "Chinese version of TPU" to seize opportunities in AI inference, leveraging its subsidiaries to penetrate the AI consumer electronics market. The consumer segment is expected to realize its potential soon, while enterprise orders are rapidly increasing, indicating a broad future growth space [2][10]. - The company is projected to achieve revenues of 12.51 billion CNY, 16.07 billion CNY, and 19.54 billion CNY from 2025 to 2027, respectively, with a focus on optimizing its revenue structure through both consumer and enterprise segments [10][17]. Financial Summary - Total revenue is expected to grow from 506 million CNY in 2023 to 1.95 billion CNY in 2027, reflecting a compound annual growth rate (CAGR) of approximately 21.6% [4][17]. - The net profit attributable to the parent company is forecasted to improve from -383 million CNY in 2023 to -208 million CNY in 2027, indicating a reduction in losses over the forecast period [4][17]. - The company’s gross margin is expected to increase from 20.94% in 2024 to 29.84% in 2027, driven by improvements in both consumer and enterprise segments [19]. Revenue and Profit Forecast - Consumer segment revenue is anticipated to grow at a rate of 20% to 21% from 2025 to 2027, reaching approximately 4.87 million CNY, 5.89 million CNY, and 7.07 million CNY, respectively [14][15]. - Enterprise segment revenue is expected to see significant growth, with projections of 99.49%, 50.23%, and 30.11% year-on-year increases from 2025 to 2027, resulting in revenues of 4.96 million CNY, 7.45 million CNY, and 9.69 million CNY [15][16]. - Industry segment revenue is forecasted to grow modestly at 2% annually, reaching 2.59 million CNY, 2.64 million CNY, and 2.69 million CNY from 2025 to 2027 [16]. Valuation Analysis - The report utilizes both PS and PB valuation methods, concluding a reasonable valuation of 369.62 billion CNY, corresponding to a target price of 103.01 CNY [21][22]. - The company is expected to maintain a PS ratio of 23 times by 2026, reflecting its growth potential in the AI inference chip market [21][22]. Market Position and Strategy - The company is recognized as a leading player in the AI sector, focusing on AI inference chips and aiming to capitalize on the rapid growth of the AI market, which is projected to expand significantly in the coming years [25][44]. - The strategic acquisition of Qiancheng Technology is expected to enhance the company's capabilities in integrating AI technology with consumer electronics, thereby accelerating its market penetration [14][45].

Investment Rating - The report maintains a "Positive" investment rating for the communication industry [10]. Core Insights - The communication sector saw a 4.21% increase in the 52nd week of 2025, ranking 8th among major industries in the Yangtze River region. Year-to-date, the sector has risen by 87.14%, ranking 1st [2][4]. - Nvidia has entered a non-exclusive licensing agreement with AI inference chip leader Groq, with a transaction value of $20 billion, significantly above Groq's September valuation of $6.9 billion. This acquisition enhances Nvidia's capabilities in inference processing [5]. - Alibaba's Tongyi Laboratory has launched a self-developed layered architecture image model, Qwen-Image-Layered, which optimizes computing power and enables precise image editing, facilitating the international expansion of domestic multimodal large models [6]. - The Long March 12 rocket successfully completed its first flight, marking a milestone for China's commercial aerospace sector, with new regulations facilitating the listing of rocket companies on the Sci-Tech Innovation Board [7]. Summary by Sections Market Performance - In the 52nd week of 2025, the communication sector increased by 4.21%, with a year-to-date rise of 87.14%, leading among major industries [2][4]. - Top gainers in the sector include Tongyu Communication (+30.7%), Yongding Co. (+28.1%), and Beidou Star (+27.3%), while China Telecom (-5.3%), Dekeli (-3.6%), and Shijia Photon (-3.6%) faced declines [4]. Key Developments - Nvidia's acquisition of Groq focuses on Groq's LPU architecture, which uses on-chip SRAM to enhance memory access speed, improving the overall inference processing ecosystem [5]. - Alibaba's Qwen-Image-Layered model breaks traditional diffusion model limitations, achieving element-level decoupling and precise editing, thus supporting the export of domestic multimodal models [6]. - The successful flight of the Long March 12 rocket represents a significant advancement in commercial aerospace, with new guidelines supporting the listing of rocket companies [7]. Investment Recommendations - Recommended operators include China Mobile, China Telecom, and China Unicom. - For optical modules, recommended companies are Zhongji Xuchuang, Xinyi Sheng, Tianfu Communication, and Shijia Photon, with attention to Taicheng Light and Yuanjie Technology. - In liquid cooling, the recommendation is for Yingweike, while for hollow core fibers, Fenghuo Communication, Hengtong Optic-Electric, and Zhongtian Technology are suggested, with a focus on Changfei Optical Fiber. - For domestic computing power, recommended companies include Runze Technology, Guanghuan New Network, Aofei Data, Huafeng Technology, Guangxun Technology, ZTE, and Unisplendour. - In AI applications, recommended firms are Boshi Jie, Heertai, Tuobang Co., Yiyuan Communication, Meige Intelligent, and Guanghetong. - For satellite applications, recommended companies include Huace Navigation, Haige Communication, and Canqin Technology [8].

Core Insights - Nvidia has acquired a non-exclusive license for technology from AI chip startup Groq, which includes key personnel joining Nvidia [1][2] - The deal is valued at $20 billion, significantly higher than Groq's previous valuation of $6.9 billion in September 2024 [1][5] Group 1: Technology and Product Details - The technology license covers Groq's inference technology, specifically its flagship Language Processing Unit (LPU), which claims to be ten times more efficient than Nvidia's and AMD's GPUs in terms of energy consumption [3] - Groq's LPU can process inference workloads at ten times the speed of GPUs while consuming only one-tenth of the energy [3] - The LPU's efficiency is attributed to its deterministic design, which allows precise control over computation timing, addressing issues like "crystal-based drift" that can slow down AI inference processes [4] Group 2: Strategic Implications - Nvidia's CEO Jensen Huang stated that integrating Groq's low-latency processors into Nvidia's AI factory architecture will expand capabilities for AI inference and real-time workloads [2] - The collaboration is seen as a way to maintain competitive dynamics in the AI processing market, especially as major clients are developing their own AI processors [7] - Despite losing key leadership, Groq will continue to operate independently, with its CFO Simon Edwards stepping in as CEO [4] Group 3: Financial Outlook - Groq anticipates reaching $500 million in revenue by the end of 2024, supported by its GroqCloud platform, which offers chip usage services and an open-source AI model library [6]

Core Insights - Nvidia has completed its largest acquisition to date, acquiring AI chip startup Groq for $20 billion (approximately 140.5 billion RMB) [1] - The deal involves Nvidia obtaining licensing rights to Groq's inference technology, with Groq's key executives joining Nvidia to facilitate the implementation of this technology [1][3] - Groq will continue to operate as an independent company, with its new CEO being the current CFO Simon Edwards, while Nvidia's CEO Jensen Huang indicated a nuanced approach to the acquisition [3][5] Company Background - Groq was founded in 2016 by Jonathan Ross, who previously designed the core architecture of Google's first TPU chip [5][7] - The company specializes in ASIC chips designed specifically for AI inference, known as LPU (Language Processing Unit), which outperform Nvidia's GPUs in certain scenarios regarding speed and energy efficiency [7][9] - Groq recently completed a $750 million funding round, achieving a valuation of $6.9 billion, with revenue targets of $500 million for the year [9] Market Dynamics - The AI industry is shifting focus from training to inference, with increasing demand for low-latency and efficient response capabilities [9][11] - Nvidia's acquisition strategy reflects a growing trend among tech giants to secure talent and technology through licensing agreements rather than full acquisitions, allowing for quicker integration of core capabilities [14][16] - The competitive landscape indicates that smaller AI startups may find better opportunities through partnerships with larger companies rather than struggling for market share independently [11][19] Strategic Implications - Nvidia aims to integrate Groq's low-latency processors into its AI infrastructure, enhancing its platform's capabilities for a broader range of AI inference and real-time workloads [17] - The company has been aggressively investing in various AI infrastructure firms, indicating a strategy to dominate the entire AI ecosystem [18][19] - Groq's situation serves as a lesson for emerging companies in the AI sector, suggesting that being acquired by a major player may be a more favorable outcome than pursuing an independent path to market [19]

Macro Strategy - The report highlights that export control measures on key metals have become a crucial advantage for China in the context of major power competition, particularly against the US and EU [10][11] - China's leading position in key metals is attributed not only to resource advantages but also to a complete industrial system that is difficult for developed economies to replicate [10][11] Fixed Income - The report indicates that the bond market is expected to face challenges in 2026, with less likelihood of a unilateral decline in interest rates as seen from 2022 to 2024, suggesting a need for a more flexible approach to bond trading [12][13] - Recent meetings have emphasized a continuation of proactive fiscal and moderately loose monetary policies, which are expected to influence bond market sentiment positively [12][13] Industry Analysis - The report discusses the merger of CICC with Dongxing and Xinda, which is expected to create a company with total assets exceeding one trillion yuan, ranking fourth in the industry [21][23] - The merger is seen as a strategic move to enhance scale and comprehensive strength in the securities industry, with large brokers expected to fill gaps through acquisitions while smaller brokers may achieve rapid growth through external mergers [21][23] Recommended Stocks - The report recommends investing in StarRing Technology, which is positioned as a leader in AI and big data infrastructure, with expected revenues of 426 million, 488 million, and 583 million yuan for 2025, 2026, and 2027 respectively [23][24] - The company is noted for its innovative GPU-Native database, which is anticipated to benefit significantly from the growing demand for AI inference [23][24]

Macro Strategy - The report highlights that export control measures on key metals have become a crucial advantage for China in the geopolitical competition with major economies like the US and EU. This advantage is rooted in China's resource superiority and a complete industrial system that is difficult for developed economies to replicate in the short term [10][11]. - China leads in the production of 30 out of 44 key minerals, with significant advantages in rare earths, gallium, germanium, and antimony. For instance, China holds nearly half of the world's rare earth reserves and dominates the processing technology, with a 92% global market share in high-performance neodymium-iron-boron magnets [10][11]. Fixed Income - The report indicates that the bond market is influenced by recent important meetings, with a shift in sentiment following the Central Political Bureau and Central Economic Work Conference. The 10-year government bond yield rose from 1.8285% to 1.8425% during the week, reflecting market reactions to policy expectations [12][13]. - The issuance of green bonds saw a significant increase, with 29 new bonds issued totaling approximately 36.75 billion yuan, up 16.01 billion yuan from the previous week. The secondary market saw a total transaction volume of 64.4 billion yuan, a decrease of 1.7 billion yuan [16]. Industry - The automotive industry is at a pivotal point in 2025, with expectations for electric vehicle penetration to reach 50%-80% over the next three years. Major players in intelligent driving have successfully implemented complex scenarios in urban environments, enhancing user experience [21]. - The report evaluates the intelligent driving experiences of seven major manufacturers, noting that Huawei and Xiaopeng lead in handling complex scenarios, while the gap in capabilities among manufacturers is narrowing compared to earlier quarters [21].

Summary of Conference Call for He Sheng New Material and Yi Zhi Electronics Company and Industry Overview - **Company**: He Sheng New Material - **Industry**: AI Chip and Electronics Key Points and Arguments 1. He Sheng New Material achieved steady growth through high capacity utilization and optimization of downstream customers, including Panasonic, Hitachi, LG, Samsung, and Midea, focusing on cost reduction and efficiency rather than large-scale expansion to support the long-term development of the chip industry [2][3] 2. The company strategically invested in Yi Zhi Electronics to deepen its involvement in the AI chip sector, planning to empower Yi Zhi Electronics and establish a foothold in the chip industry [2][3] 3. Yi Zhi Electronics specializes in server and CPU development, having mass-produced two generations of chips, with the third generation design nearing completion and entering the tape-out and mass production phase [2][4] 4. The CPU from Yi Zhi Electronics integrates NPU functionality, targeting three main customer categories: major telecom operators and financial institutions (Category A), leading internet companies (Category B), and inspection robots and factory AI applications (Category C), with approximately 8,000 computing nodes delivered this year for cloud gaming, cloud mobile, and video transcoding [2][4] 5. Yi Zhi Electronics has open-sourced a complete computing framework compatible with the CUDA ecosystem and major domestic GPU manufacturers, achieving compatibility with around 40 hardware and software vendors without requiring special versions [2][5][6] 6. The chips from Yi Zhi Electronics offer unique advantages in the AI inference era, integrating neural network computing with general computing to enhance data flow processing efficiency and reduce power consumption, suitable for electric vehicle factories and large-scale automation industries [2][7] 7. The third-generation product is expected to be mass-produced in 2026, with a projected performance improvement of 30% [2][7] 8. Yi Zhi Electronics' future strategy focuses on generalization and AI integration, aiming to create a broadly compatible ecosystem and develop multi-functional processors to adapt to large models and changing computing frameworks [2][8][9] 9. HiSilicon Technology plans to shift from the x86 ecosystem to the ARM ecosystem, emphasizing collaboration with Yi Zhi Electronics to support vertical applications and overall business development [2][10] 10. Yi Zhi Electronics has received three repeat orders from last year to this year, indicating strong market competitiveness and customer recognition, with further growth potential anticipated for the next year [2][11] 11. The company maintains its competitive edge through technological innovation, including self-developed NPU units and an open-source computing framework, ensuring user-friendly product compatibility [2][12][13] 12. The next-generation chip is expected to launch in mid-2026, utilizing advanced technology and targeting the high-end market [2][14] 13. Among the three customer categories (A, B, C), the internet customer segment is currently experiencing the fastest growth, prompting significant resource investment for product optimization [2][14] 14. Yi Zhi Electronics plans to continue promoting AI integrated machines while not ruling out the possibility of standalone chip sales [2][15] 15. Following the change in control, the company will increase investments in the chip industry, particularly in Yi Zhi Electronics, and will comprehensively arrange capital, talent, and strategic direction towards AI chips as a key development focus [2][16] Additional Important Information - The strategic investment in Yi Zhi Electronics reflects He Sheng New Material's commitment to transforming its business model and adapting to the evolving technology landscape [2][3][10] - The collaboration with major internet companies and telecom operators indicates a strong market presence and potential for future growth in the AI chip sector [2][4][15]