Core Viewpoint - The article discusses the emergence of supernodes in high-performance computing, emphasizing their role in enhancing the efficiency of large-scale model training and inference through optical technology [1][2][21]. Group 1: Supernode Architecture and Performance - Supernodes provide a new solution for large-scale model training and inference, significantly improving efficiency by optimizing resource allocation and data transmission [1]. - The architecture of supernodes can be categorized into single-layer and two-layer designs, with single-layer architecture being the ultimate goal due to its lower latency and higher reliability [4][6]. - The demand for GPU power has surged with the exponential growth of model sizes, necessitating thousands of GPUs to work in tandem, which supernodes can facilitate [1][2]. Group 2: Challenges in Domestic Ecosystem - Domestic GPUs face significant performance gaps compared to international counterparts, requiring hundreds of domestic GPUs to match the power of a few high-end international GPUs [6][8]. - The implementation of supernodes in the domestic market is hindered by limitations in manufacturing processes, such as the 7nm technology [6]. Group 3: Development Paths for Supernodes - Two main development paths are proposed: increasing the power capacity of individual cabinets to accommodate more GPUs or increasing the number of cabinets while ensuring efficient interconnection [8][10]. - Optical interconnect technology is crucial for multi-cabinet scenarios, offering significant advantages over traditional copper cables in terms of transmission distance and flexibility [10][12]. Group 4: Optical Technology Advancements - The transition to higher integration optical products, such as Co-Packaged Optics (CPO), enhances system performance by reducing complexity and improving reliability [14][16]. - CPO technology can save 1/3 to 2/3 of power consumption, which is significant even though communication power is a smaller fraction of total GPU power [16][17]. Group 5: Reliability and Flexibility - The use of distributed optical switching technology enhances the flexibility and reliability of supernodes, allowing for dynamic topology adjustments in case of node failures [18][19]. - Optical interconnect technology simplifies the supply chain, making it more controllable compared to advanced process-dependent components [19][21]. Group 6: Future Outlook - With advancements in domestic GPU performance and the maturation of optical interconnect technology, the supernode ecosystem is expected to achieve significant breakthroughs, supporting the rapid development of artificial intelligence [21].

Summary of Key Points from the Conference Call Industry Overview - The conference call focuses on the AI computing power industry, particularly the demand for AI computing clusters and the implications for major tech companies like Microsoft, Meta, and Amazon [1][2][3]. Core Insights and Arguments 1. **AI Computing Demand Trends**: There is a significant expected growth in AI computing demand, particularly in training and inference. The market has shown a discrepancy in expectations, especially before the earnings reports of major companies [2][3]. 2. **Optimistic Outlook for AI Computing Clusters**: The outlook for AI computing clusters is optimistic, with anticipated increases in inference demand in the first half of 2025 and training demand in the second half [1][3]. 3. **U.S.-China AI Development Gap**: The gap in AI development between the U.S. and China may widen, depending on the evolution of large model iterations over the next year. The U.S. is expected to continue advancing parameter optimization, while China may rely on software algorithm innovations [1][5][8]. 4. **Role of Clusters in AI Model Iteration**: Clusters play a crucial role in AI model iterations, especially for large-scale computational tasks. The emergence of technologies like DeepSpeed indicates a shift towards reduced dependency on large clusters [7][9]. 5. **Impact of DeepSpeed**: The introduction of DeepSpeed marks the end of the computing inflation logic and initiates a new deflation logic, reducing the overall reliance on large clusters [9][10]. 6. **Market Focus on Optical Interconnect Technology**: There has been a notable increase in market attention towards optical interconnect technologies and related companies due to the growing demand for large clusters [11][12]. 7. **Changes in Major Tech Companies' Cluster Needs**: Major tech companies have shifted their needs away from large clusters, with many opting for strategies that do not require significant investments in large-scale computing resources [12][24]. 8. **Future Model Iteration Paths**: The next year is expected to see a return to pre-training phases, which will require substantial computational resources. Different companies will adopt varied strategies for this transition [14][15]. 9. **Meta's Data Strategy**: Meta's strategy involves leveraging its vast data resources, but merely increasing data volume has not significantly improved model performance. The acquisition of Skillz AI aims to enhance data quality [16][18]. 10. **Challenges in Large-Scale Cluster Construction**: The construction of large clusters faces various bottlenecks, including data and storage walls, which require hardware upgrades or algorithm optimizations to overcome [32][37]. Other Important but Potentially Overlooked Content - **Market Expectations for 2025**: The A-share market is expected to experience fluctuations in AI computing, with downward expectations in the first half of 2025 and upward expectations in the second half, driven by actual demand and supply chain recovery [40]. - **Technological Innovations**: Innovations in communication technologies, such as Broadcom's "Fat Cat" technology, are crucial for enhancing data synchronization and load balancing in training processes [36]. - **Scalability Trends**: There is an anticipated increase in the demand for scale-up solutions, which enhance the computational capacity of individual nodes, as opposed to scale-out solutions [38][39]. This summary encapsulates the key points discussed in the conference call, highlighting the trends, challenges, and strategic directions within the AI computing power industry.