半导体-未来数年超大规模数据中心资本支出仍有充足上行空间，支撑人工智能加速器 GPUXPU 半导体TAM实现 40 - 50% 的复合年增长率-Still Ample Room for Upside to Hyperscaler DC Capex in Coming Years, Supporting 40-50% CAGR for AI Accelerator GPUXPU Semiconductor TAM

Summary of J.P. Morgan's Semiconductor Research Call Industry Overview - Industry: Semiconductors, specifically focusing on AI-related infrastructure and hyperscaler data center capital expenditures (Capex) [1][4] Key Insights - Data Center Capex Growth: - Data center spending by the Top 4 U.S. hyperscalers is projected to grow approximately 60% year-over-year (Y/Y) in 2025, an increase from the previous estimate of 40% [1] - This growth translates to an incremental spend of $100 billion+ in 2025 compared to 2024, up from $75-80 billion in 2024 [1] - Data center Capex is expected to track around 30% growth in 2026, revised from 20% previously, indicating an incremental spend of $80-85 billion [1] - AI Accelerator Market: - The spending environment is anticipated to support a 40-50% compound annual growth rate (CAGR) for the AI accelerator total addressable market (TAM), starting from a base of approximately $200 billion in 2025 [1] - Hyperscaler Capex Intensity: - The capital intensity for the Top 5 U.S. hyperscalers is currently around 18% of gross revenue, expected to rise to 21% by 2026 [1] - Incremental compute demand and upgrade cycles are likely to maintain capital intensity at higher levels than historical averages [1] Company-Specific Insights - Key Beneficiaries: - Companies expected to benefit from strong AI server spending include: - NVIDIA (NVDA): Focused on AI compute acceleration processors - Broadcom (AVGO): Involved with Google TPU and Meta MTIA ASICs - Marvell (MRVL): Engaged in AI XPU ramps - Advanced Micro Devices (AMD): Positioned as the second-largest merchant GPU supplier - Micron Technology (MU): Notable for HBM memory and eSSD products [4] - Market Dynamics: - The demand for AI compute is driving a multi-year growth trajectory for NVDA's data center GPU business, with AMD poised to capture demand through its MI400/MI500 series [4] - Broadcom and Marvell are expected to benefit from continued cloud and hyperscale AI XPU ramps, while Micron is ramping its HBM3E 12-hi products [4] Additional Considerations - Emerging Spend Sources: - Significant spending is occurring outside traditional hyperscalers, including neoclouds and sovereign AI projects, which are becoming increasingly impactful as they scale [1] - Long-Term Outlook: - The expectation is that Capex growth forecasts will continue to increase as visibility and confidence in sustained spending growth solidify [1] Conclusion - The semiconductor industry, particularly in the context of AI infrastructure, is poised for substantial growth driven by hyperscaler spending and evolving market dynamics. Key companies in this space are well-positioned to capitalize on these trends, with expectations of significant revenue opportunities in the coming years [1][4]