超节点OEM-被低估的中国AI核心资产

Summary of Key Points from the Conference Call Industry Overview - The conference call focuses on the Chinese AI industry, particularly the emerging "super node" OEM market, which is expected to see significant growth starting in the second half of 2026. [1][2] Core Insights and Arguments - Token Consumption Growth: By 2026, China's average daily token call volume is projected to reach 140 trillion, reflecting over a thousand-fold increase in two years. This growth indicates a substantial demand potential in AI inference. [1][2] - Super Node OEM Year: 2026 is defined as the "Super Node OEM Year" for China, with domestic AI chips and self-developed internet chips expected to be released in large quantities. This new server form factor will significantly impact OEM manufacturers' business and profitability. [1][3] - Investment Logic Shift: The investment logic is shifting from "domestic substitution" to "total growth," with previously undervalued segments like servers and switches expected to experience both performance and valuation boosts. [1][4] - Supply Side Explosion: The supply of AI computing power in China is anticipated to experience three major growth points: the release of domestic AI chip capacity, the approval of NVIDIA's H200 special supply version, and a doubling of computing power leasing procurement year-on-year. [1][2][7] Important Developments - NVIDIA's Role: NVIDIA's special supply version of the H200 chip is expected to significantly contribute to China's AI computing power supply, with potential orders reaching hundreds of thousands to millions of units. [7] - Alibaba's Super Node: Alibaba's "Pan Jiu" 128 card super node is expected to achieve mass production in the second half of 2026, supporting over 5,000 units annually, which will notably boost ODM business. [1][5] - Capital Expenditure Gap: There is a significant capital expenditure gap between Chinese and North American cloud service providers, with Chinese CSPs projected to spend around $100 billion compared to $870 billion for North American CSPs from 2023 to 2025. This gap indicates a strong future demand for AI assets in China. [6][7] Investment Opportunities - Focus Areas: Key areas for investment include: - Servers and Super Node OEM: Companies like Inspur Information, Sugon, and Hon Hai Precision Industry. - AI Chips: Companies such as Cambricon, Haiguang Information, and others. - Network Connectivity: Firms like Zhongji Xuchuang and Shengke Communication. - Cloud Computing and Computing Power Services: Companies like Xiechuang Data and Hongjing Technology. - Large Models and Applications: Companies like Zhipu AI and iFlytek. [8][9] Risks to Consider - Market Risks: Potential risks include macroeconomic fluctuations affecting downstream demand, slower-than-expected advancements in AI model technology, intensified market competition squeezing profit margins, and policy uncertainties arising from the US-China tech rivalry. [8][9]