Changjiang Securities·2025-04-17 12:05Investment Rating - The report maintains a "Positive" investment rating for the industry [7]. Core Viewpoints - The U.S. government has intensified export controls on high-end AI chips to China, specifically requiring export licenses for NVIDIA's H20 and AMD's MI308 chips, which could lead to significant financial impacts for these companies [2][4]. - The Chinese AI computing market is vast, and if strict export limitations are enforced without successful license applications, domestic high-end AI chip manufacturers are expected to rapidly increase their market share [2][4]. - The demand for AI computing power is entering a new expansion phase, driven by increased capital expenditures from major Chinese companies, with Alibaba planning to invest over 380 billion yuan in cloud and AI hardware infrastructure over the next three years [10]. Summary by Sections Event Description - NVIDIA announced that its H20 chip and any other high-end chips meeting similar specifications must apply for export licenses indefinitely starting April 14, 2025, with potential costs related to this estimated at up to $5.5 billion [4]. - AMD has also completed its assessment of the new export license requirements for its MI308 product, expecting to apply for the license but uncertain of approval, with potential costs up to $800 million [4]. Market Dynamics - The report highlights that the U.S. has implemented multiple rounds of export controls and restrictions on AI chip access for China, which has led to a significant market share for NVIDIA and AMD's customized products in China despite their reduced performance metrics [10]. - The Chinese accelerated server market is projected to reach $22.1 billion in 2024, growing 134% year-on-year, with GPU servers maintaining a dominant position [10]. Recommendations - The report emphasizes the acceleration of technological self-reliance in China amid external disruptions, recommending investment in leading domestic high-end AI chip manufacturers such as Cambricon and those excelling in CPU and DCU technologies [10].