Core Insights - The Chinese computing power industry is experiencing rapid growth in capital operations, highlighted by significant IPOs and market enthusiasm for domestic semiconductor companies [1][2] - The focus of competition in the domestic computing power sector is shifting from hardware specifications to system stability, software ecosystem usability, and cost-effectiveness [3][4] Capital Market Activity - TianShuZhiXin Semiconductor Co., Ltd. went public on January 8, 2026, with over 400 times subscription, indicating strong market interest [1] - Other domestic GPU companies, such as MoEr Thread and MuXi Co., saw their stock prices surge on their debut, with MoEr Thread's market cap exceeding 305.5 billion yuan and MuXi's reaching 330 billion yuan [1] - ChangXin Technology submitted its IPO application on December 30, 2025, reporting revenue of 32.084 billion yuan for the first three quarters of 2025, showcasing the scale of domestic DRAM production [1] Technological Developments - The "Ten Thousand Card Cluster" concept is becoming a benchmark for evaluating domestic computing power, but it also presents challenges in reliability as system scale increases [3][4] - The introduction of the scaleX Ten Thousand Card Super Cluster by ZhongKe Shuguang, featuring 10,240 AI accelerator cards, represents a significant advancement in system architecture [3][4] - The need for high-quality, low-latency data transmission networks is critical for supercomputing, with domestic products now matching international standards [5][6] Storage Solutions - ChangXin Technology and ChangChun Group are positioned in the core areas of DRAM and NAND Flash, respectively, with ChangXin reporting a compound annual growth rate of over 70% in revenue from 2022 to 2024 [6][7] - The introduction of advanced technologies like Xtacking in NAND Flash production by ChangChun Group marks a significant technological breakthrough [7] Software Ecosystem - The transition to a robust software ecosystem is complex, with developers facing high costs in switching from established platforms like NVIDIA's CUDA [10][11] - MoEr Thread is addressing this by launching the MTT AIBOOK, which includes development tools to facilitate easier adoption of its platform [10] - Cloud service providers are playing a crucial role in integrating various hardware brands to create a unified software environment, addressing compatibility issues [11][12] Market Dynamics - The industry is witnessing a shift towards collaborative ecosystems, with companies recognizing the need for specialization rather than attempting to cover the entire supply chain independently [9][12] - The emergence of customized products from companies like Haiguang is aimed at meeting the specific needs of large enterprises, reflecting a trend towards more open architectures [15] Future Outlook - The domestic computing power industry is expected to face challenges related to global supply chain fluctuations, particularly in DRAM and NAND supply [13] - The successful integration of domestic computing solutions in high-stakes environments, such as the National High Energy Physics Data Center, indicates growing confidence in local technologies [14] - The potential easing of export restrictions on NVIDIA's H200 chip could impact the domestic ecosystem, but the established supply chain and customer preferences for security are likely to mitigate risks [17]

Core Viewpoint - The article discusses the increasing demand for optical modules in AI clusters, particularly in relation to the architecture and scale of the networks used in semiconductor and AI applications [2][5][10]. Group 1: Optical Module Requirements - In Huawei's CM384 super node, the ratio of NPU to optical modules is calculated to be 1:18, requiring a total of 6,912 optical modules for 384 NPUs [4]. - The comparison between Huawei and NVIDIA's server optical module usage reveals that CM384 has a significantly higher optical module requirement, indicating a trend towards "full optical interconnection" [5]. - The demand for optical modules increases non-linearly with the scale of AI clusters, with larger clusters requiring more complex network architectures [6][10]. Group 2: Network Architecture Impact - In a small cluster of 1,024 GPUs, the ratio of optical modules to GPUs is approximately 2.5, but this jumps to 3.5 when scaling to 4,096 GPUs due to the introduction of a third layer of core switches [6][8]. - For ultra-large clusters (e.g., 100,000 GPUs), the ratio of optical modules to GPUs can reach up to 4, indicating a significant increase in network complexity [6][10]. Group 3: Cost Differences Among Solutions - Different interconnect solutions exhibit notable cost differences; for instance, NVIDIA's InfiniBand solution is the most expensive at approximately $3.9 billion, with a ratio of 3.6 optical modules per GPU [11]. - Broadcom's Ethernet solution is the most cost-effective at around $3.5 billion, with a similar optical module ratio of 2.6, saving approximately $400 million compared to InfiniBand [11]. Group 4: Future Trends - As GPU clusters continue to grow, the network architecture may evolve to four or even five layers, potentially increasing the optical module to GPU ratio from 3.5 to 4.5 [10]. - Broadcom's Ethernet solution is expected to gain traction due to its cost advantages, particularly in large-scale deployments where budget constraints are a concern [10].

Report Summary 1. Report Industry Investment Rating - The rating for the global economy in the macro and financial sector is (Bullish) [1] 2. Core View of the Report - The report presents a series of global economic and financial news and analyzes their potential impacts. It shows that although there are certain risks in the global economic situation, such as the escalating Japanese debt crisis and potential trade disputes, there are also positive factors, including the resilience of China's exports and the improvement of domestic demand, the strong performance of the US economy, and the expansion of the European economy. 3. Summary by Relevant Catalog Important Information - The Japanese debt crisis has escalated, with the 10 - year yield approaching 1.6%, the highest since 2008. Market concerns about a change in fiscal policy may trigger a "bond vigilante" sell - off [1] - UK retail sales in June increased 3.1% year - on - year, exceeding the 1% growth in May and setting the second - largest monthly increase this year [1] - Meta has launched two giant AI clusters, Prometheus and Hyperion, to break through the computing power bottleneck. Prometheus has a scale of up to 1 GW [1] - A well - known asset management institution believes that the investment opportunity in emerging market bonds is "once - in - a - generation" [1] - Citi believes that China's exports in June showed resilience, and imports had their first year - on - year positive growth, reflecting improved domestic demand [1] - The EU is formulating a retaliatory plan and will propose a new tariff list covering about 72 billion euros of US imports [1] - Trump said that if the Russia - Ukraine conflict is not resolved within 50 days, the US will impose "very severe, about 100%" tariffs on Russia [1] Global Economic Logic - China's GDP grew 5.3% in the first half of the year. Asian exports are strong, and the US inventory has not increased, indicating strong end - demand [1] - The market expects the Fed to cut interest rates in September 2025 and accelerate rate cuts in 2026 [1] - The US Markit manufacturing PMI in June was 52.0, continuing to expand [1] - China's PMI production index continued to expand, and the new order index resumed expansion in June [1] - China's comprehensive rectification of involution - style competition is expected to boost listed company performance [1] - The European Central Bank has cut interest rates 8 times, and Germany is large - scale arming with a 30% military expansion [1]

Core Viewpoint - NVIDIA has officially launched two CPO (Co-Packaged Optics) switch products, Quantum-X Photonics for InfiniBand and Spectrum-X Photonics for Ethernet, with the InfiniBand CPO expected to debut in the second half of 2025 and the Ethernet CPO in the second half of 2026 [1][2]. Group 1: CPO Technology and Market Position - CPO will be an optional configuration, and NVIDIA will continue to offer traditional switch systems with pluggable modules [2]. - The primary driver for NVIDIA's investment in CPO technology is power optimization, with a significant reduction in power consumption from 30W to 9W for a 1.6T port, achieving a 70% decrease [2]. - NVIDIA's CPO solution utilizes new micro-ring modulators (MRM) for enhanced energy efficiency, contrasting with Broadcom's CPO solution that achieved a 50% power reduction using traditional Mach-Zehnder modulators (MZM) [2]. Group 2: Ecosystem and Partnerships - The technology involves a multi-component ecosystem, integrating electronic and photonic chips through 3D stacking, with TSMC's compact optical engine technology playing a key role [3]. - NVIDIA's CPO partners include major industry players such as Browave, Coherent, Corning, Fabrinet, Foxconn, Lumentum, Senko, SPIL, Sumitomo, Tianfu Communication, and TSMC [3]. Group 3: Strategic Implications and Future Developments - LightCounting notes that NVIDIA's first CPO product is an InfiniBand switch, which has been overshadowed by Ethernet in NVIDIA's AI strategy, indicating that the initial deployment will primarily serve NVIDIA's own clusters [4]. - The Spectrum-X platform aims to elevate Ethernet performance to match that of InfiniBand, with potential for significant GPU interconnectivity in future architectures [4]. - NVIDIA's entry into the CPO market is expected to invigorate the ecosystem, with both NVIDIA and Broadcom projected to release single-channel 200G CPO switches by 2027, leading to a more mature industry landscape [4]. Group 4: Technical Challenges and Innovations - Scale-out is seen as a low-risk entry point for CPO, while scale-up is critical for success, particularly for the mixed expert (MoE) model requiring rapid response times across GPUs [5]. - NVIDIA has previously announced fiber-based NVLink plans, with at least one cluster built internally, but large-scale deployment has been hindered by high power consumption of timing modules [6]. - The NVLink CPO is scheduled for 2028, allowing NVIDIA to validate technology feasibility over two product cycles, significantly reducing future integration risks [6].