2025-06-04 01:53Investment Rating - The report maintains an "In-Line" industry view for the global technology sector, particularly focusing on AI supply chains and server racks [6]. Core Insights - The ramp-up of NVL72 server racks is a significant relief for the global AI supply chain, with a target of "1k per week" still to be achieved. AI GPU shipments to China are also highlighted as a critical development to monitor [1][2]. - NVIDIA is reiterated as a top pick due to strong demand for Blackwell chips, while TSMC is also favored for its intact CoWoS-L orders and anticipated wafer price hikes [8]. Summary by Sections NVL72 Server Rack Ramp - The NVL72 server rack ramp is the primary focus, with an assumption of 30k NVL72 server racks in 2025. Post-Computex, estimates have been adjusted to 25-30k [2]. - NVIDIA's recent earnings call indicated that "multiple customers are taking 1k rack per week," suggesting a positive outlook for rack shipments [2][13]. - TSMC's CoWoS-L capacity for Blackwell chips is projected at a street-high estimate of 390k for 2025, contrasting with supply-chain chatter suggesting a reduction to 340k [2][14]. Key Events/Data Points from Asia Supply Chain - Monthly sales from Taiwan's ODMs (Hon Hai, Quanta, Wistron) are crucial, with a target of 3-3.5k NVL72 server racks per month to achieve over 10k per quarter [3]. - TSMC's forecast for 2026 CoWoS consumption is currently held at 90k, with expectations of 31% year-over-year growth in cloud AI semiconductors [4][15]. China GPU Demand and Supply - The new license for shipping AI GPUs to China has not yet been granted, but demand for the B30 China version GPU is growing, with initial volume estimates at 0.5 million units [5][17]. - AMD is reportedly planning to introduce a gaming-graphic-based GPU to replace the MI308 GPU for shipments to China [5]. Global AI Capex and Market Dynamics - The report forecasts that the top four US hyperscalers will generate $550 billion in operating cash flow in 2025, supporting ongoing investments in AI data centers [47]. - AI server capital expenditure is expected to grow significantly, with a projected 40% year-over-year increase in 2025 [58][61].