Workflow
RTX Pro Enterprise 服务器
icon
Search documents
COMPUTEX现场走访:主权AI、美国制造、NVLinkFusion是三大话题
HTSC· 2025-05-21 09:31
Investment Rating - The report maintains an "Overweight" rating for the technology sector [4] Core Insights - Sovereign AI investments are temporarily compensating for the sales restrictions in the Chinese market, but the slow development of AI applications poses a significant risk to the entire industry chain [1] - The challenge for the Taiwanese industry chain is to meet U.S. government requirements for domestic manufacturing while ensuring technological upgrades do not slow down [2] - NVLink Fusion, a new technology announced by NVIDIA, may further strengthen its position in the industry [3] Summary by Sections Sovereign AI Investment - Following the U.S. Department of Commerce's announcement to revoke the AI diffusion rules, NVIDIA and the global computing industry saw a significant rebound in stock prices. This regulatory adjustment opens up sales opportunities for AI chips in major Middle Eastern countries, potentially compensating for the estimated $5.5 billion loss from restricted sales to China. However, investors express concerns over the lack of substantial applications for generative AI to support over $300 billion in annual capital expenditures [1] U.S. Manufacturing Challenges - Taiwan hosts a complete data center industry chain, including TSMC and Hon Hai. The primary challenge is balancing the need for technological upgrades every two years with the U.S. government's domestic manufacturing requirements. TSMC has committed to investing $165 billion in six fabs in the U.S., while Hon Hai has pledged to produce servers domestically. Investors are worried about the impact of this transition on profitability [2] NVLink Fusion Technology - NVIDIA has introduced NVLink Fusion, which allows third-party chip companies to connect their ASIC and CPU chips to NVIDIA's ecosystem. Eight companies, including MediaTek and Qualcomm, have announced support for this technology. NVLink Fusion is seen as a solution for ASIC integration in AI chips, potentially reinforcing NVIDIA's industry position [3]