Core Insights - Nvidia continues to innovate with the launch of new systems and architectures, including Groq LPX, Vera ETL256, and STX, as well as updates to the Kyber rack architecture and the introduction of the CPO [3][4]. Group 1: Groq Acquisition and LPU Architecture - Nvidia's acquisition of Groq involved a payment of $20 billion for intellectual property rights and team integration, effectively streamlining the process to avoid regulatory hurdles [4]. - The LPU architecture from Groq is designed with single-purpose units called "slices," which enhance data flow and processing efficiency compared to traditional multi-core architectures [5][6]. - The first generation of LPU utilizes a 14nm process, allowing for a focus on architecture validation rather than raw performance, with the ability to be fully manufactured in the U.S. [6][7]. Group 2: SRAM and Memory Hierarchy - SRAM is utilized in Groq's LPU for its low latency and high bandwidth, although it comes with limitations in memory density and total throughput compared to GPUs [8][9]. - The LPU's SRAM can achieve rapid token processing times, but its limited capacity restricts overall throughput, necessitating a hybrid approach with GPUs for memory-intensive tasks [8][9]. Group 3: Integration of LPU and GPU - The integration of LPU into Nvidia's inference architecture aims to enhance performance in high-interaction scenarios by leveraging LPU's low-latency characteristics [19]. - The attention mechanism and feedforward neural networks (FFN) are decoupled to optimize GPU utilization for dynamic workloads while assigning static workloads to LPU [25][27]. - The use of AFD (Attention Feedforward Network Decoupling) allows for efficient token routing between GPUs and LPUs, although it may introduce bottlenecks under strict latency constraints [27][29]. Group 4: LPX Rack System - The LPX rack system features 32 1U LPU compute trays and is designed for high-density configurations, with Nvidia planning modifications before mass production [35][38]. - Each LPX compute tray includes 16 LPUs, Altera FPGAs, and Intel Granite Rapids CPUs, facilitating high-performance computing in data centers [38][43]. - The LPU network architecture is designed for high bandwidth and low latency, with a total vertical bandwidth of 640TB/s per rack [44][46]. Group 5: Future Roadmap and Innovations - Nvidia's roadmap includes the introduction of the LP40, which will utilize TSMC's N3P process and incorporate new IP for enhanced performance [15][54]. - The upcoming Feynman generation will feature a large-scale NVL1152 supercomputer, utilizing CPO technology for inter-rack connectivity while maintaining copper connections within racks [56][60]. - The Kyber rack architecture has evolved to support higher density and performance, with each rack capable of housing 144 GPUs and utilizing advanced NVLink technology for interconnectivity [66][70].

英伟达在GPU领域深耕多年，自1999年发布首款GPU至今，已有约27年时间。其芯片制程从220纳米迭 代至4纳米左右，未来还将向1.6纳米推进，这也是我们较为期待的投资价值。 本轮AI浪潮始于2023年，当时市场主流GPU为A100与H100。截至目前，市场主流GPU已更新为 Blackwell架构芯片。A100与H100的核心技术特征是什么？其中H100芯片性能强劲，在2023年AI浪潮爆 发后，迅速成为市场炙手可热的GPU产品。 H100由中国台湾台积电采用4纳米工艺代工生产，单芯片集成800亿个晶体管，还专门内置了 Transformer模型引擎。为什么要专门针对Transformer做硬件适配？当前国内外我们耳熟能详的各类大模 型，其底层架构基本都是基于Transformer基础架构针对性优化发展而来。 英伟达极具前瞻性地在Hopper架构中，从硬件层面对Transformer做了专项优化，也就是引入了对应的专 用引擎。英伟达也凭借这一核心优势，在短短两年多的时间里，从一家规模中等的企业，成长为全球市 值第一的科技巨头，由此也能充分看到AI产业的强劲爆发力。 英伟达在2023年前后发布的Blackw ...

Summary of Conference Call Records Industry Overview - The AI Cloud IaaS sector is entering a price increase cycle driven by surging demand from applications like OpenCL, with significant price upside potential for AIDC, computing power leasing, and CDN services [1][2] Key Insights and Arguments - **Price Increases**: Major cloud computing companies, including Tencent Cloud, Alibaba Cloud, and Baidu Cloud, have announced price hikes primarily due to a surge in Token usage, indicating a bullish outlook for the AI Cloud IaaS sector [2][3] - **Token Economics**: Industry giants like NVIDIA and Alibaba are emphasizing "Token Economics," with NVIDIA introducing the concept of "Token Power Plants" at its GTC conference, suggesting that future data centers will focus on producing Tokens as a core revenue driver [3][4] - **Tencent's Strategic Shift**: Tencent's recent financial report indicates a strategic pivot towards AI investments, with plans to reduce stock buybacks to fund AI initiatives, reflecting a strong confidence in the future of AI [4] Market Dynamics - **Supply and Demand**: The domestic AIDC and computing power leasing markets are experiencing extreme demand, with prices for H100 and 5,090 servers doubling in the past month. Some companies expect a sixfold increase in B card orders by 2026 [7] - **Market Expectations**: There is a discrepancy in market expectations regarding the sustainability of AI applications, with some skepticism about the longevity of current trends. However, the demand for AI products continues to rise, indicating a shift from a buyer's market to a seller's market [7] Investment Logic and Recommendations - **AIDC Investment Logic**: AIDC serves as a leading indicator for AI infrastructure, with high-power domestic computing cards driving demand and revenue. Recommended companies include Guanghui New Network, which is seen as a core supplier for ByteDance and has significant growth potential [8][9] - **Valuation Model**: AIDC resources are estimated to correspond to a market value of approximately 50 billion, based on a model that considers the investment required and expected EBITDA [9] - **Emerging Trends**: The global shift towards liquid cooling technology is expected to benefit companies like Invec, while CDN services are also poised for growth due to the rise of AI applications and edge computing [9][10] Additional Noteworthy Content - **Core Investment Targets**: Key investment targets in the AI infrastructure space include companies involved in liquid cooling, AIDC power supply, CDN, and AI computing networks. Specific companies mentioned include Shenyang Environment, Zhongheng Electric, and NetEase Technology [9][10] - **Future Infrastructure Upgrades**: The 2026 super node is identified as a critical technological direction that will drive upgrades across servers, liquid cooling, switches, optical modules, and power supplies [10]

Core Viewpoint - Nvidia has raised its revenue guidance for the Blackwell and Rubin series chips at the GTC conference, extending the forecast to 2027, indicating a significant increase in AI demand visibility [1] Group 1: Nvidia's Revenue Guidance - Nvidia expects the Blackwell and Rubin series chips to generate $1 trillion in revenue by 2027, an increase from the previous guidance of $500 billion by the end of 2026 for data center equipment [1] - The revenue guidance extension to 2027 reflects a clear improvement in AI demand visibility [1] Group 2: LPX Architecture and Chip Production - The LPX architecture is set to begin shipping in the second half of 2026, featuring the Groq 3 LPX rack with 256 LPU processors, 128GB on-chip SRAM, and 640TB/s expansion bandwidth [2] - The combination of LPX with the Vera Rubin platform is expected to enhance inference throughput/power ratio by 35 times [2] - LPU chips will be manufactured by Samsung, with rack shipments anticipated to start in the latter half of this year [2] Group 3: Copper Foil Market Dynamics - The demand for high-frequency and high-speed copper foil is increasing due to severe signal attenuation, leading to higher performance requirements for copper-clad laminate materials [3] - Major players in the copper foil market, such as Mitsui and Taiwanese companies, are negotiating price increases for ultra-thin copper foil used in AI servers, with an average price increase of about 15% expected [3] - Domestic manufacturers are likely to follow suit with price increases, benefiting from the tight supply-demand situation in high-end electronic circuit copper foil [3] Group 4: Investment Recommendations - Recommended stocks include Defu Technology (301511.SZ), which is well-positioned to benefit from copper foil price increases; Cuprum Copper Foil (301217.SZ), with extensive experience in electronic circuit copper foil; and Jiyuan Technology (688388.SH), which collaborates with CATL and has acquired Endatong for optical modules [4] - Other notable mentions are Nord Shares (600110.SH), leading in lithium battery 4.5-micron products, and Zhongyi Technology (301150.SZ) [4]

Group 1 - The A-share market saw a strong performance in PCB concept stocks, with notable gains from companies such as Aoshikang, which hit the daily limit, and Huhua Electronics, which approached the limit as well [1] - Key stocks included Jinhua Electronics with a rise of over 9%, Aohong Electronics up 8%, and Jin'an Guoji increasing by over 5% [1] - Other companies like Guanghe Technology, Xiehe Electronics, and Xinqi Microelectronics also experienced gains exceeding 3% [1] Group 2 - Aoshikang's stock price increased by 10%, with a total market value of 16.2 billion and a year-to-date increase of 50.92% [2] - Jinhua Electronics rose by 9.24%, with a market value of 544.9 million and a year-to-date increase of 36.05% [2] - Huhua Electronics saw an increase of 4.81%, with a market value of 172.6 billion and a year-to-date increase of 89.71% [2] Group 3 - Nvidia's recent GTC 2026 conference introduced the Feynman architecture, which sets extreme requirements for PCB layers (32-44 layers), thermal resistance, and signal transmission rates [1][3] - This development is expected to significantly increase the value of high-end multilayer boards (HLC) and high-density interconnect boards (HDI) [1] - The new architecture indicates a substantial increase in the number of trays per server, which translates to an increase in PCB demand [3]

Core Insights - Nvidia is set to unveil a new AI inference system at the upcoming GTC conference, featuring a chip optimized specifically for inference tasks, with OpenAI as its first major client [1][3][6] Group 1: New Chip Development - The new chip's architecture is based on the LPU (Language Processing Unit) designed by the former Groq team, marking Nvidia's first significant integration of external architecture into its core AI computing products [3][4] - This strategic move follows Nvidia's acquisition of Groq's core technology and team for approximately $20 billion, demonstrating a focus on rapid deployment of mature solutions [3][10] Group 2: Market Dynamics and Competition - The demand for inference capabilities is rapidly increasing, prompting Nvidia to provide targeted solutions more quickly, especially as competitors like Cerebras and Amazon are developing specialized chips for inference tasks [6][13][15] - The shift in focus from training to inference is reshaping the AI computing landscape, with companies like OpenAI and Meta exploring alternatives to Nvidia's GPUs for inference workloads [13][14][16] Group 3: Technical Advantages of LPU - The LPU architecture utilizes high-density on-chip SRAM, significantly reducing data movement latency and energy consumption, making it more suitable for low-latency inference scenarios compared to traditional GPUs [8][20] - The LPU is theoretically capable of achieving speeds up to 100 times faster than GPUs, addressing the bottlenecks associated with data access and movement during the inference process [8][21] Group 4: Future Outlook - Nvidia's introduction of the LPU chip is seen as a critical response to the evolving demands of the AI market, where inference is becoming a primary focus rather than a supplementary phase [10][21] - The upcoming GTC conference is anticipated to showcase not only the new LPU chip but also potentially other groundbreaking products, including the Rubin series GPUs and possibly new consumer-grade graphics cards [22][23]

Group 1: Macro Overview and Asset Performance - In February, the Shanghai Composite Index rose by 1.1%, while the CSI 300 increased by 0.1%. The ChiNext Index and the STAR 50 fell by 1.1% and 1.4%, respectively. The bond market saw a slight increase of 0.1%, with the 1-year government bond yield rising by 1.7 basis points, while the 5-year and 10-year yields fell by 3.4 and 3.6 basis points, respectively [1][20][11] - The Hang Seng Index dropped by 2.8%, and the three major US stock indices showed mixed results, with the Dow Jones Industrial Average up by 0.2%, while the Nasdaq and S&P 500 fell by 3.4% and 0.9%, respectively [1][20][11] Group 2: A-Share Market Analysis - The re-ignition of conflict in Iran, trade uncertainties, and rising inflation in the US are creating short-term upward pressure on the market. Attention should be focused on the upcoming Two Sessions and the mid-March release of the 14th Five-Year Plan outline, which may provide upward momentum for the market [2][21] - Current market expectations for China's GDP growth in Q1 are at 4.70%, with a slight decline to 4.63% in Q2. As the annual report disclosure period approaches in March, market risk appetite may face further adjustments [2][21] Group 3: Investment Strategy - The investment strategy emphasizes focusing on performance certainty and price-increasing varieties. During the annual report disclosure phase, attention should be paid to sectors with strong performance certainty such as CPO, PCB, and liquid cooling, as well as upstream price-increasing products like indium phosphate and copper foil [3][24] - In the technology sector, advancements in AI hardware are expected to open growth opportunities in the computing power supply chain. Continued focus on AI-related hardware and price-increasing products in the power sector is recommended [3][25] Group 4: Bond Market Strategy - With the upcoming Two Sessions in early March, which will clarify the new year's work objectives, the bond market is expected to enter a new pricing cycle. The 10-year government bond yield has shown a rebound demand, and if policy and economic data meet expectations, yields are likely to continue rising [4][27][28] - The performance of China's port container throughput in February remains strong, indicating resilient exports, which may support bond yields [4][28] Group 5: Economic Fundamentals - The RMB has appreciated against the USD, rising from 7.35 on April 9, 2025, to 6.84 on February 26, 2026, marking a 7% increase. This appreciation is attributed to both market forces and policy guidance [35][38] - The broad social financing (社融) remains stable, with significant government financing contributing to the economic support. However, the growth rate of social financing is expected to decline, which may weaken macro liquidity and its support for the economy and market [35][41][47]

Investment Rating - The report maintains a positive outlook on the AI-PCB and core computing hardware sectors, as well as the Apple supply chain and self-controlled beneficiary sectors [4][26]. Core Insights - NVIDIA's FY27Q1 revenue guidance is strong at $78 billion (±2%), indicating sustained demand for AI [1][26]. - The demand for AI-related products, particularly the new LPU chip, is expected to drive significant growth in the PCB industry, with potential increases in material requirements and production capacity [1][26]. - The report anticipates explosive growth in the ASIC numbers from major tech companies like Google, Amazon, Meta, OpenAI, and Microsoft between 2026 and 2027 [4][26]. Summary by Sections 1. Consumer Electronics - The report highlights the ongoing expansion of AI applications in consumer electronics, particularly within the Apple supply chain and smart glasses, driven by advancements in multi-modal interaction and AI model optimization [5]. - AI mobile applications are expected to see significant growth, with Apple focusing on innovations in chips, systems, and hardware [5]. 2. PCB - The PCB industry is experiencing high demand, particularly for copper-clad laminates, driven by the automotive and industrial sectors, as well as AI applications [6][26]. - The report notes that many AI-PCB companies are currently operating at full capacity and are expanding production to meet demand [4][26]. 3. Semiconductor and Components - The semiconductor sector is projected to benefit from increased demand for storage solutions, particularly DRAM, as cloud service providers expand their data center capacities [20][22]. - The report emphasizes the importance of domestic semiconductor equipment and materials in light of global supply chain challenges and export controls [23][25]. 4. Market Trends - The report indicates a robust upward trend in the semiconductor equipment sector, with significant growth expected in 2025 driven by advanced logic processes and increased demand for HBM applications [24]. - The overall electronic industry has shown a positive performance, with specific segments like PCB and passive components leading in growth [35][38]. 5. Company Highlights - Companies like NVIDIA, Micron Technology, and various PCB manufacturers are highlighted as key players benefiting from the current market dynamics [1][26][27]. - The report suggests that firms such as North Huachuang and Zhongwei Company are well-positioned to capitalize on the growing demand for semiconductor equipment and materials [28][29].

Investment Rating - The report maintains a positive outlook on the AI-PCB and core computing hardware sectors, as well as the Apple supply chain and self-controlled beneficiary sectors [4][26]. Core Insights - NVIDIA's FY27Q1 revenue guidance is strong at $78 billion (±2%), indicating sustained demand for AI [1][26]. - The demand for AI-driven products is expected to lead to significant growth in the PCB industry, with many AI-PCB companies experiencing strong orders and production [4][26]. - The report anticipates explosive growth in the number of ASICs from major tech companies like Google, Amazon, Meta, OpenAI, and Microsoft between 2026 and 2027 [4][26]. Summary by Sections 1. Industry Overview - The AI demand is driving a significant increase in PCB value, with expectations for material upgrades and increased production layers [1][26]. - The report highlights the strong performance of the AI sector, with NVIDIA's new LPU chip expected to create new demands in the PCB market [1][26]. 2. Semiconductor and PCB Sector - The PCB industry is maintaining high demand due to the growth in automotive and industrial control sectors, alongside AI expansion [6][26]. - The report notes that the price of copper-clad laminates is expected to rise, indicating a tightening supply situation [6][26]. 3. Consumer Electronics - The report emphasizes the ongoing expansion of consumer electronics applications, particularly in the Apple supply chain and smart glasses [5][26]. - AI applications are expected to continue their rapid growth, with significant advancements in mobile and wearable technologies [5][26]. 4. Storage and Memory - The storage sector is projected to enter an upward cycle, driven by increased demand from cloud service providers and consumer electronics [20][22]. - The report suggests that the DRAM market will see price increases due to supply constraints and rising demand [20][22]. 5. Semiconductor Equipment - The semiconductor equipment sector is expected to benefit from the ongoing trend of domestic production and self-sufficiency in the face of global supply chain challenges [23][25]. - The report highlights the strong demand for advanced packaging and the need for domestic semiconductor equipment [23][25]. 6. Key Companies - Companies such as North Huachuang, Zhongwei Company, and Sanhua Group are highlighted for their strong positions in the semiconductor and PCB markets [27][28][33]. - The report notes that these companies are well-positioned to benefit from the ongoing trends in AI and semiconductor demand [27][28][33].

Summary of Longjiang Electronics Conference Call Industry Overview - The conference call discusses advancements in the PCB (Printed Circuit Board) industry, particularly focusing on the upcoming launch of the Feynman architecture chip by Nvidia, which is expected to create new growth opportunities in the sector [1]. Key Points and Arguments - Nvidia plans to unveil the Feynman architecture chip at the GTC 2026 conference, with the release occurring earlier than market expectations [1]. - The core technological breakthrough of the Feynman architecture lies in its 3D stacking method, which integrates the LPU (Logic Processing Unit) chip optimized for inference tasks directly onto the GPU (Graphics Processing Unit) computing core, enabling deep physical integration of general and specialized computing [1]. - The new LPU chip is primarily designed for inference, utilizing a high multi-layer approach, with the value of a single chip PCB expected to reach between $300 to $500 [1]. - Key suppliers to watch in this space include Shenghong Technology, Huidian Co., Shenzhen South Circuit, and Jingwang Electronics [1]. Additional Important Insights - The CoWoP (Chip-on-Wafer-on-Panel) technology is anticipated to enter small-scale production by the end of 2027 and large-scale production in 2028, potentially increasing the PCB value per square meter by several times, up to tenfold [1]. - The orthogonal backplane project is progressing steadily, with a new round of sample testing planned for early March. This project is expected to enter mass production in the second half of 2027 [1]. - There is a strong recommendation to focus on undervalued leading companies in this sector, as their cost-performance ratio is expected to become more prominent [1].