大中华区半导体全球人工智能供应链更新;亚洲半导体的关键机遇_57页_9mb
报告摘要
Morgan Stanley Greater China Technology Semiconductors Investor Presentation Summary
Date: September 19, 2025
Presented By: Charlie Chan, Daniel Yen, Daisy Dai, Tiffany Yeh
Key Points
-
Industry Outlook
- Uppgraded Industry View to "Attractive" for 2H25, with strong AI focus.
- Industry challenges: Tariffs, FX volatility; opportunities: AI-driven re-rating, memory shortages easing.
- Cloud spending growth (US$582B in 2026), AI semi TAM: US$235B in 2025e, US$1T in 2030.
-
AI Semiconductor Leadership
- TSMC: Top Pick (AI fabless, Foundry leader).
- Edge/Inference: Inferencing chips growing faster (55% CAGR).
- NVIDIA dominance, but China push to reduce reliance via TSMC (7nm) and local players (Huawei Ascend).
-
Investment Themes
- CoWoS/SoIC/HBM: Critical for AI compute density/bandwidth.
- Memory: DDR4/NOR Flash short supply through 2H26; NAND/LSD short in 2026.
- Lower Tier: China auto, consumer chips localization (e.g., MCU, auto semis).
-
Valuation Highlights
- TSMC: 31x P/E, strong revenue growth; ASE: Leading back-end play.
- China focus: SMIC (7nm progress), GigaDevice/Phison (NOR Flash).
-
China-Specific Opportunities
- Auto semi self-sufficiency: From 15% in 2024 to 28% in 2027.
- GPU: China self-sufficiency ratio ~34% in 2024; target 39% by 2027.
Analyst Recommendations
- Charlie Chan: Reiterate AI > Non-AI preference.
- Daniel Yen: Downside risks for Semiconductor/Power chips (FX, inventory cycles).
- Daisy Dai: Top Taiwanese plays: TSMC, ASE.
- Tiffany Yeh: Focus on logic vs memory decoupling, cloud capex drivers.
Key Metrics
- AI Chip Performance: NVIDIA L40S vs. Huawei Ascend910C + TSMC N4.
- TCO Analysis: ASICs vs GPUs competitive in training/inference with scaling.
- Demand Signals: Cloud capex, token processing, memory shortages.
Market Risks
- Inventory Cycles: Logic foundry utilization at 70-80%; memory oversupply risks.
- Geopolitical: U.S. Executive Order 14032 restrictions on semiconductor entities.
Visuals (Inferred from Report)
- Sector Re-rating Chart: AI semis leading logic performance.
- CoWoS Allocation Table: NVIDIA/AMD dominance.
- HBM Consumption: Up to 24B GB in 2026.
Appendices
- TCOs: Edge vs Inference growth vectors.
- Roll-ups: China auto/GPU localization plays.
(For full details, see Morgan Stanley Greater China Tech Semi Report)
展开完整摘要
试读结束,高清完整版pdf/doc/ppt,请点下载