Investment Rating - The industry investment rating is "Overweight" [6] Core Insights - The report highlights the launch of four generations of GPU architectures by the company, which includes solutions for AI computing, professional graphics acceleration, desktop graphics acceleration, and intelligent SoC products [1][2][25] - The company aims to provide a comprehensive computing acceleration infrastructure and one-stop solutions globally, focusing on AI computing support for various industries [17][25] - The company has achieved significant revenue growth, with AI computing products projected to contribute 3.36 billion yuan, accounting for 77.63% of total revenue in 2024 [1][2] Summary by Sections Section 1: Company Overview - The company has undergone three development phases, focusing on full-function GPUs to support digital transformation across industries [17] - The company has successfully launched four generations of GPU architectures from 2021 to 2024, with products catering to various market needs [1][28] Section 2: AI Computing Products - The company’s AI computing products are expected to generate significant revenue, with a projected 42.42% of total revenue coming from AI computing clusters in 2024 [1][2] - The company has developed a diverse product line, including AI computing boards and modules, with a focus on high-performance applications [1][2][25] Section 3: Technological Advancements - The company has advanced its manufacturing capabilities, achieving rapid iteration from 12nm to 7nm processes and enhancing its domestic supply chain [2][7] - The company has developed a heterogeneous computing architecture integrating GPU, CPU, NPU, and VPU, leading to the successful production of the "Yangtze" SoC chip [2][32] Section 4: Market Performance - The report provides insights into the performance of the electronic industry, comparing it with the CSI 300 index, indicating a competitive landscape [3][4] - The company has established a strong investor base, having completed six rounds of financing totaling over 4.5 billion yuan, with notable investors including Sequoia China and Tencent [8][9]

Core Insights - The surge in demand for AI computing power has led to a wave of IPO submissions from domestic GPU manufacturers, while a new paradigm of AI computing based on light is gaining traction among investors [1] - "Light-based Technology" has completed strategic financing rounds, indicating strong investor interest and confidence in its innovative approach to AI computing [1][3] Company Overview - "Light-based Technology" was established in 2022 and is the first company globally to commercialize light computing chips that integrate silicon photonics and phase change materials (PCM) [3] - The company has achieved significant milestones, including the development of a light computing chip with a matrix size of 128x128, surpassing the previous industry ceiling of 64x64 [3][5] Technology and Product Development - The technology path adopted by "Light-based Technology" allows for a tenfold increase in integration density, making it suitable for large-scale AI computing scenarios [3] - The company is currently working on the production of 256x256 light computing chips and has designs for 512x512 chips, which are expected to exceed the performance of current leading electronic chip products [3][5] Strategic Partnerships and Collaborations - In December 2024, "Light-based Technology" established a strategic partnership with a leading domestic internet company to collaborate on AI computing hardware [5] - The company is actively exploring product requirements with core users in high-performance computing fields, including large models and scientific calculations [5] Investor Sentiment - Investors view photon computing as a breakthrough technology that addresses the limitations of traditional computing paradigms, with "Light-based Technology" positioned as a leader in this field [6][7] - The unique "PCM + Crossbar" technology route of "Light-based Technology" is recognized for its potential to enhance AI inference speed and significantly reduce power consumption [7][9] Market Position and Future Outlook - "Light-based Technology" is seen as a key player in the AI computing landscape, with its technology aligning with national strategic priorities for computing infrastructure and artificial intelligence [8] - The company is expected to redefine the AI chip market landscape through its innovative solutions and strong execution capabilities [9]

Core Viewpoint - HSBC warns that Tesla's profitability will continue to be disappointing, facing three significant challenges in scaling its Robotaxi business [1][3] Group 1: Delivery and Financial Forecasts - HSBC maintains a "reduce" rating on Tesla with a target price of $120, indicating a 63% downside from the current stock price [1][4] - Based on actual sales data from April and May, HSBC predicts that Tesla's Q2 delivery volume will remain flat quarter-over-quarter, which is 15% lower than market expectations [2][3] - The firm expects operational revenue to be 8% lower than consensus due to weak delivery volumes, with free cash flow expected to be slightly positive [2][3] Group 2: Challenges Facing Robotaxi - Tesla's Robotaxi business must overcome three major challenges: proving the robustness of its pure camera solution compared to competitors' sensor combinations, changing consumer perceptions about vehicle ownership, and demonstrating profitability in the Robotaxi operation [3] - Early signs from the Austin pilot raise concerns about the reliability of Tesla's approach [3] Group 3: Valuation and Business Segments - HSBC uses a DCF valuation and peer multiples, each accounting for 50% of the valuation, maintaining a target price of $120 [4] - The DCF valuation covers six business segments, including automotive, energy storage, full self-driving, Dojo, Optimus, and services, yielding a fair value of $180 per share [5]

Core Insights - Snowcap Compute, a new player in the AI chip market, has raised $23 million in seed funding led by Playground Global, with former Intel CEO Pat Gelsinger joining its board [2][8] - The company aims to develop a commercially viable superconducting AI computing chip that significantly outperforms current systems while consuming minimal power [2][12] - Snowcap plans to launch its first foundational chip by the end of 2026, with a complete system to follow later [2] Company Overview - Snowcap's CEO, Michael Lafferty, has a background in superconducting and quantum technologies, having previously led a team at Cadence [3] - The founding team includes experts from leading tech companies, such as former NVIDIA, Google, and Tesla executives, enhancing its credibility and expertise [5][8] Technology and Innovation - Snowcap's chip architecture is designed for extreme performance and energy efficiency, targeting AI, quantum, and high-performance computing workloads [2][12] - The company utilizes Josephson junctions instead of traditional transistors, achieving a power efficiency that is five orders of magnitude lower than current chips [12][13] - The superconducting technology allows for zero electrical resistance, drastically reducing energy consumption during operations [13][14] Market Position and Future Outlook - Snowcap is positioned to address the limitations of current CMOS technology, paving the way for a post-CMOS era in computing [13] - The company aims to tackle the energy consumption challenges faced by AI computing, which has become critical due to the increasing demand for computational power [15] - By overcoming engineering challenges related to scalability and compatibility with existing semiconductor processes, Snowcap is injecting fresh innovation into the AI chip industry [15]

Core Insights - The article discusses the technological evolution of NVIDIA's GPU architecture, particularly focusing on the advancements in tensor cores and their implications for AI and deep learning performance [2]. Performance Fundamentals - The Amdahl's Law provides a framework for understanding the limitations of performance improvements through parallel computing, indicating that the maximum speedup is constrained by the serial portion of a task [3][4]. - Strong scaling and weak scaling describe the impact of scaling computational resources on performance, with strong scaling focusing on reducing execution time for fixed problem sizes and weak scaling addressing larger problem sizes while maintaining execution time [6]. Tensor Core Architecture Evolution - The Volta architecture marked the introduction of tensor cores, addressing the energy imbalance between instruction execution and computation in matrix multiplication, with the first tensor core supporting half-precision matrix multiply-accumulate (HMMA) instructions [9][10]. - Subsequent architectures, such as Turing, Ampere, Hopper, and Blackwell, introduced enhancements like support for INT8 and INT4 precision, asynchronous data copying, and new memory architectures to optimize performance and reduce data movement bottlenecks [11][12][13][17][19]. Data Movement and Memory Optimization - Data movement is identified as a critical bottleneck in performance optimization, with modern DRAM operations being significantly slower than transistor switching speeds, leading to a "memory wall" that affects overall system performance [8]. - The evolution of memory systems from Volta to Blackwell has focused on increasing memory bandwidth and capacity to meet the growing computational demands of tensor cores, with Blackwell achieving a bandwidth of 8000 GB/s [19]. MMA Instruction Asynchronous Development - The evolution of Matrix Multiply-Accumulate (MMA) instructions from Volta to Blackwell highlights a shift towards asynchronous execution, allowing for overlapping data loading and computation, thereby maximizing tensor core utilization [20][24]. - Blackwell's architecture introduces single-threaded asynchronous MMA operations, significantly enhancing performance by reducing data movement delays [23][30]. Data Type Precision Evolution - The trend towards lower precision data types across NVIDIA's architectures aligns with the needs of deep learning workloads, optimizing power consumption and chip area while maintaining acceptable accuracy levels [25][27]. - Blackwell architecture introduces new micro-scaled floating-point formats (MXFP8, MXFP6, MXFP4) and emphasizes low-precision types to enhance computational throughput [27]. Programming Model Evolution - The programming model has evolved to focus on strong scaling optimization and asynchronous execution, transitioning from high occupancy models to single Cooperative Thread Array (CTA) tuning for improved performance [28][29]. - The introduction of asynchronous data copy instructions and the development of distributed shared memory (DSMEM) in Hopper and Blackwell architectures facilitate more efficient data handling and computation [29][31].

Core Viewpoint - The report from Morgan Stanley highlights a significant increase in the global production of GB200 NVL72 racks, driven by the surging demand for AI computing, particularly in cloud computing and data center sectors [1][2]. Group 1: Production Forecast - The global total production of GB200 NVL72 racks is estimated to reach 2,000 to 2,500 units by May 2025, a notable increase from the previous estimate of 1,000 to 1,500 units in April [1]. - The overall production for the second quarter is expected to reach 5,000 to 6,000 units, indicating a robust supply chain response to market demand [1]. Group 2: Company Performance - Quanta shipped approximately 400 GB200 racks in May, a slight increase from 300 to 400 units in April, with monthly revenue reaching about 160 billion New Taiwan Dollars, a year-on-year increase of 58% [2]. - Wistron demonstrated a strong growth trajectory, shipping around 900 to 1,000 GB200 computing trays in May, a nearly sixfold increase from 150 units in April, with revenue growth of 162%, reaching 208.406 billion New Taiwan Dollars [2]. - Hon Hai shipped nearly 1,000 GB200 racks in May, with a forecast of delivering 3,000 to 4,000 racks in the second quarter, despite some decline in its cloud and networking business due to traditional server shipment slowdowns [2]. Group 3: Market Dynamics - The actual delivery volume of GB200 racks may be lower than the reported shipment figures due to the need for further assembly of Wistron's L10 computing trays into complete L11 racks, which involves additional testing and integration time [3]. - Morgan Stanley ranks the preference for downstream AI server manufacturers as Giga-Byte, Hon Hai, Quanta, Wistron, and Wiwynn, with Giga-Byte being favored for its potential in GPU demand and the server market [3]. - A report from Tianfeng Securities indicates that major hyperscale cloud providers are deploying nearly 1,000 NVL72 cabinets weekly, with the shipment pace continuing to accelerate [3].

点击蓝字，关注我们 今日上午港股 1. 信义能源(03868)再涨超5%。 消息面上，信义能源、信义光能联合公布，建议中国上市的申请材料 已递交予发改委。太阳能发电场基础设施REIT拟于中国一家证券交易所上市，仍需获得监管机构批 准及豁免。 2. 石药集团(01093)涨超3%。 6月5日，石药集团发布公告，集团开发的腺苷钴胺胶囊(0.5mg)("该产 品")已获得中国国家药品监督管理局颁发的药品注册批件，并视同通过仿制药质量和疗效一致性评 价。 3. 东岳集团(00189)涨超5%。 据百川盈孚数据，第三代主流制冷剂R32（主要用于家用和商用空 调），目前其国内均价每吨已突破5万元大关，同比增长42%；截至6月5日，R32行业毛利润达3.5万 元/吨，较去年同期每吨毛利润增加超过1万元。 4. 康诺亚-B(02162)涨超3%。 近日，康诺亚在药物临床试验登记与信息公示平台官网上登记了一项 CDH17ADC药物CM518D1治疗晚期实体瘤的I/II期临床研究。据公开资料显示，这是该药首次启动临 床。 5. 药师帮(09885)再涨超6%。 药师帮公告，与越疆科技(02432)订立战略合作与业务部署协议。 6 ...

Core Viewpoint - Intel and SoftBank are collaborating to develop a stacked DRAM as a substitute for high bandwidth memory (HBM), aiming for prototype completion by 2027 and commercialization by 2030 [1][2]. Group 1: Collaboration and Development - The joint venture, named Saimemory, will leverage Intel's technology and patents from Japanese universities to create prototype chips [1]. - The new stacked DRAM is expected to reduce power consumption by half compared to existing HBM chips, addressing the high power demands of AI processors [1][2]. Group 2: Market Context and Competition - Currently, only Samsung, SK Hynix, and Micron can mass-produce the latest generation of HBM chips, leading to a supply shortage amid rising AI chip demand [2]. - Saimemory aims to capture the Japanese data center market and re-establish Japan's presence in the mainstream memory chip supply chain after over two decades [2]. Group 3: Industry Trends - The focus of Saimemory is on reducing power consumption, which is a critical need for data centers, especially as AI computing power demands continue to rise [2]. - Other companies, such as Samsung and NEO Semiconductor, are also exploring 3D stacked DRAM, but their emphasis is on increasing chip capacity rather than power efficiency [2].

如果您希望可以时常见面，欢迎标星收藏哦~ 北京奕斯伟计算技术股份有限公司（Beijing ESWIN Computing Technology Co., Ltd.，简称"奕 斯伟计算"）今日向香港联交所递交招股书，计划在主板挂牌上市。这标志着这家以RISC-V架构为 核心的中国科技企业，正迈向资本市场新阶段，或有望成为港股RISC-V"第一股"。 奕 斯 伟 计 算 成 立 于 2019 年 ， 是 奕 斯 伟 集 团 在 AI 与 计 算 架 构 领 域 的 重 要 布 局 。 公 司 聚 焦 新 一 代 RISC-V计算架构的研发与商业化，围绕智能终端与具身智能两大核心应用场景，打造高效、开 放、具竞争力的系统级软硬件平台。 根据咨询机构弗若斯特沙利文的数据，奕斯伟计算是截至2024年中国RISC-V主控量产解决方案数 量最多的提供商，亦是中国最大的RISC-V全定制解决方案供应商。 截至2024年底，公司已成功商业化超100款系统级解决方案，服务客户超百家，覆盖消费电子、汽 车、机器人等领域，客户包括多家全球头部企业。 随着AI计算、物联网与智能制造快速发展，业界对可控、开放架构的计算平台需求日益增长。 ...

Domestic News - Beijing encourages foreign enterprises to participate in high-level autonomous driving demonstration zones and key application scenarios, providing support for technology testing and demonstration applications [3] - Changan plans to launch over 7 high-end automotive products in the next 5 years, aiming for total revenue of 300 billion yuan and sales of 3 million vehicles, including 1 million new energy vehicles [4] - Geely is set to launch multiple satellites, including the "Qianli Haohan," to establish a global real-time satellite communication service [5] - BYD's blade battery and fast-charging blade battery have passed the new national standard testing ahead of schedule, with the new standards set to take effect in July 2026 [6] - Wuling's Shenlian battery has passed the latest national standard certification, showcasing its leading position in battery technology [7] - GAC Group has published a patent for an AI computing device for smart cockpits, enhancing AI capabilities without modifying existing hardware [9] - Chery has announced a patent for an automatic follow robot system that utilizes facial and voice recognition for improved user convenience [10] - Pony.ai has reached a strategic cooperation with Dubai's Roads and Transport Authority to launch a Robotaxi fleet, with plans for testing in 2025 and full commercial operation in 2026 [11] International News - The European Council has approved the relaxation of carbon dioxide reduction targets for 2025-2027, allowing manufacturers to be evaluated based on average performance over three years [13] - Italy's new car sales in April increased by 2.71% year-on-year, totaling 139,084 vehicles, marking the second consecutive month of growth [14] - BMW announced a major overhaul of its product line, planning to implement a new design language across all models within three and a half years [15][16] - General Motors will invest $888 million in its Tonawanda plant in New York to produce the next generation of V8 engines, reflecting a strategic adjustment in response to electric vehicle market demands [17] Commercial Vehicles - Foton Motor is expanding its new energy heavy truck business by establishing a joint venture with EVE Energy to provide diverse battery leasing solutions [19] - XCMG has launched its new generation of "Han Jie" light trucks, focusing on high safety, reliability, long endurance, and low energy consumption [20] - JAC Motors has unveiled 25 strategic new products across various powertrain categories, emphasizing innovation in new energy commercial vehicles [21] - Chery's new pickup has been officially registered, marking a significant step towards its domestic launch [22][23]