Investment Rating - The report maintains a "Recommended" rating for several key companies in the semiconductor and AI infrastructure sectors, including Chipone Technology, Industrial Fulian, and Huakong Technology [4][45]. Core Insights - Oracle has signed a significant cloud service agreement expected to generate over $30 billion annually starting from FY2028, which will account for approximately 52% of its total revenue for FY2025 [1][8]. - The demand for AI computing power is driven by three main application scenarios: third-party large language model (LLM) training, sovereign AI infrastructure development, and customized private cloud solutions for enterprise clients [2][33]. - The global AI computing landscape is evolving with both GPGPU and ASIC technologies advancing rapidly, indicating a dual-track growth in the market [3][12]. Summary by Sections Oracle's Major Contract and GPU Demand - Oracle's recent contract is a record-breaking deal that significantly impacts its revenue structure, highlighting the rapid growth in AI model and cloud service demand [1][8]. - To meet this demand, Oracle has procured approximately 400,000 NVIDIA GB200 high-end computing cards, making it the second-largest holder of NVIDIA's high-end computing cards globally [1][9]. Global AI Computing Landscape - The AI computing market is bifurcating into two main technology camps: GPGPU, led by NVIDIA, and ASIC, driven by companies like Google and Amazon [3][12]. - GPGPU technology is particularly suited for large model training and general AI applications, while ASIC technology focuses on optimizing specific tasks such as AI inference and cost efficiency [3][22]. New Growth Drivers for NVIDIA GPGPU Demand - The demand for NVIDIA's GPGPU is primarily fueled by three areas: third-party LLM training, sovereign AI initiatives, and enterprise-level private cloud deployments [33][34]. - The training of large models, such as GPT-3, requires substantial computational power, which NVIDIA's GPUs provide efficiently [34][35]. Investment Recommendations - The report suggests focusing on companies with strong core technologies and competitive advantages in the AI computing supply chain, including Chipone Technology, Industrial Fulian, and Huakong Technology [4][43]. - The long-term demand for computing power is expected to be robust, driven by sovereign AI, accelerated large model training, and enterprise private cloud deployments [4][44].