Core Viewpoint - The article discusses the architecture and data flow of Huawei's Cloud Matrix 384, emphasizing the integration of optical and electrical interconnections in its network design [2][3][9]. Group 1: Data Transmission Layers - The Cloud Matrix 384 includes three main data transmission layers: UB Plane, RDMA Plane, and VPC Plane, each serving distinct roles in data processing and communication [5][7]. - The UB Plane connects all NPU and CPU with a non-blocking full-mesh topology, providing a unidirectional bandwidth of 392GB/s per Ascend 910C [7]. - The RDMA Plane facilitates horizontal scaling communication between supernodes using RoCE protocol, primarily connecting NPUs for high-speed KV Cache transfer [7]. - The VPC Plane connects supernodes to broader data center networks, managing tasks such as storage access and external service communication [7]. Group 2: Optical and Electrical Interconnections - Although the Cloud Matrix 384 is often referred to as a purely optical interconnection system, it also utilizes electrical interconnections for short distances to reduce costs and power consumption [9]. - The article highlights the necessity of both optical and electrical connections in achieving efficient data flow within the system [9]. Group 3: Scale-Up and Scale-Out Calculations - For Scale-Up, each server's UB Switch chip corresponds to a bandwidth of 448GBps, requiring 56 400G optical modules or 28 800G dual-channel optical modules per server [12]. - The ratio of NPUs to 400G optical modules in Scale-Up is 1:14, and to 800G modules is 1:7 [12]. - For Scale-Out, a Cloud Matrix node consists of 12 Compute cabinets, and the optical module demand ratio is approximately 1:4 for NPUs to 400G optical modules [14].

Core Viewpoint - The article discusses the strategic moves of NVIDIA and its CEO Jensen Huang in response to the evolving AI market in China, highlighting the importance of the Chinese market for NVIDIA despite U.S. export restrictions and competition from domestic companies [1][2][3]. Group 1: NVIDIA's Market Strategy - Jensen Huang has made multiple trips to China, emphasizing the significance of the Chinese AI market for U.S. companies [2][4]. - NVIDIA has received U.S. government approval to export its H20 chip to China, which is a modified version of its A800/H800 chips, with significantly reduced performance [7][8]. - Following the announcement of the export approval, NVIDIA's market value surged by over 1.16 trillion yuan, solidifying its position as the world's most valuable company [8]. Group 2: Challenges and Competition - Despite a significant reduction in market share in China, the country still accounted for 13% of NVIDIA's sales in the last fiscal year [16]. - Huang acknowledged that Chinese companies are rapidly advancing in AI applications, which poses a threat to NVIDIA's market dominance [17][20]. - The U.S. government's restrictions have led to substantial losses for NVIDIA, including a $5.5 billion inventory write-down due to the ban on the H20 chip, which was expected to generate $12 billion to $15 billion in revenue [24][25]. Group 3: Domestic Competitors - Chinese companies like Huawei are rapidly developing competitive AI chips, with Huawei's Cloud Matrix 384 reportedly outperforming NVIDIA's offerings [33][36]. - The domestic AI chip market is growing, with local companies capturing a significant share, as evidenced by Huawei's Ascend chips achieving a 24.8% market share in China [36]. - The increasing use of domestic chips in AI applications is reshaping the competitive landscape, with reports indicating that over 30% of AI model training is now done using domestic chips [38][43]. Group 4: Future Outlook - NVIDIA is reportedly preparing to launch a new AI chip based on the Blackwell architecture, priced significantly lower than the H20, to maintain its foothold in the Chinese market [47][48]. - The new chip's design reflects NVIDIA's attempts to comply with U.S. export regulations while still providing a viable product for Chinese customers [52][53]. - Huang's ongoing efforts to engage with U.S. officials suggest a strategic push to navigate the complex regulatory environment and secure NVIDIA's position in the Chinese market [57][65].

Group 1 - The short-term A-share market may lack strong upward catalysts, but the expectations for fundamental recovery and policy release have not been disproven, indicating limited downside risk [1][2] - The recent US-China Geneva trade talks resulted in a joint statement agreeing to significantly reduce bilateral tariff levels, boosting market confidence [2] - April's financial data showed that new social financing maintained a year-on-year increase trend, with the stock of social financing growing at a rate of 8.7%, suggesting an upward trend in fundamentals and A-share earnings [2][5] Group 2 - The recent US restrictions on high-end computing chips for China may temporarily impact Huawei's chip exports, but domestic demand for local computing chips is strengthening [26][30] - Huawei's Cloud Matrix 384 computing cluster has achieved performance metrics that surpass Nvidia's flagship product GB200 NVL72, marking a significant breakthrough in China's AI infrastructure [31][32] - The capital expenditure of major cloud service providers like Tencent and Alibaba has decreased significantly compared to the previous quarter, but remains above historical averages, indicating a potential shift in investment strategy [25][30] Group 3 - The recent US-China tariff negotiations have led to a recovery in industries closely related to exports, such as e-commerce, chemical fibers, and shipping ports [15] - The technology sector is showing signs of recovery, but the market consensus suggests a phase of consolidation and potential volatility ahead [17][21] - The overall industry scores indicate a high allocation recommendation for sectors like electronics, computers, and automation equipment, while sectors like real estate and coal are rated for lower allocation [33]