Workflow
GB200/300机架
icon
Search documents
2025年出货量下调至2.73万台
傅里叶的猫· 2025-12-14 12:37
Core Viewpoint - The article discusses the AI industry chain, focusing on infrastructure, algorithms, and applications, while providing insights from recent reports by Morgan Stanley and JP Morgan regarding ODM manufacturers' performance and shipment forecasts. ODM Manufacturers' Performance and Shipment Analysis - Morgan Stanley ranks ODM manufacturers for GPU AI servers as Wistron > Hon Hai > Quanta [2][18] - Morgan Stanley's latest forecast for GB200/300 rack shipments is adjusted to 27,300 units, down from 28,000 units, primarily due to updates following Quanta's Q3 earnings call [2] - Quanta's management indicates a conservative outlook for AI revenue growth in Q1 2026, leading to a downward adjustment of their Q4 2025 rack shipment forecast from approximately 3,500 to 2,500 units [7] - Despite Quanta's adjustment, Wistron shows strong growth, leading to a slight increase in overall rack shipment forecasts for Q4 2025, from 8,000-8,500 to 13,500-14,000 units [7] Company-Specific Revenue Insights - Quanta reported November revenue of approximately NT$193 billion, with a month-on-month increase of 11% and a year-on-year increase of 36%, driven by GB200/300 rack shipments expected to reach 1,000-1,100 units [13] - Wistron achieved a record revenue of NT$281 billion in November, with a month-on-month increase of 52% and a year-on-year increase of 195%, attributed to significant increases in L10 computing tray shipments [14] - Hon Hai's November GB200 rack shipments remained stable at approximately 2,600 units, with expectations of a decline in December due to year-end holidays, maintaining a forecast of 7,200 units for Q4 2025 [15] 2026 Preliminary Outlook - The forecast for rack shipments in 2026 is challenging, but Morgan Stanley has adjusted its estimate to 70,000-80,000 units, up from 60,000-70,000 units, based on anticipated inventory carryover of approximately 2 million Blackwell chips [17] - Morgan Stanley maintains the ranking of ODM manufacturers as Wistron > Hon Hai > Quanta, noting that actual deliveries may be lower than predicted due to assembly and testing times for L11 racks not being included in the estimates [18]
GB200出货量上修,但NVL72目前尚未大规模训练
傅里叶的猫· 2025-08-20 11:32
Core Viewpoint - The article discusses the performance and cost comparison between NVIDIA's H100 and GB200 NVL72 GPUs, highlighting the potential advantages and challenges of the GB200 NVL72 in AI training environments [30][37]. Group 1: Market Predictions and Performance - After the ODM performance announcement, institutions raised the forecast for GB200/300 rack shipments in 2025 from 30,000 to 34,000, with expected shipments of 11,600 in Q3 and 15,700 in Q4 [3]. - Foxconn anticipates a 300% quarter-over-quarter increase in AI rack shipments, projecting a total of 19,500 units for the year, capturing approximately 57% of the market [3]. - By 2026, even with stable production of NVIDIA chips, downstream assemblers could potentially assemble over 60,000 racks due to an estimated 2 million Blackwell chips carried over [3]. Group 2: Cost Analysis - The total capital expenditure (Capex) for H100 servers is approximately $250,866, while for GB200 NVL72, it is around $3,916,824, making GB200 NVL72 about 1.6 to 1.7 times more expensive per GPU [12][13]. - The operational expenditure (Opex) for GB200 NVL72 is slightly higher than H100, primarily due to higher power consumption (1200W vs. 700W) [14][15]. - The total cost of ownership (TCO) for GB200 NVL72 is about 1.6 times that of H100, necessitating at least a 1.6 times performance advantage for GB200 NVL72 to be attractive for AI training [15][30]. Group 3: Reliability and Software Improvements - As of May 2025, GB200 NVL72 has not yet been widely adopted for large-scale training due to software maturity and reliability issues, with H100 and Google TPU remaining the mainstream options [11]. - The reliability of GB200 NVL72 is a significant concern, with early operators facing numerous XID 149 errors, which complicates diagnostics and maintenance [34][36]. - Software optimizations, particularly in the CUDA stack, are expected to enhance GB200 NVL72's performance significantly, but reliability remains a bottleneck [37]. Group 4: Future Outlook - By July 2025, GB200 NVL72's performance/TCO is projected to reach 1.5 times that of H100, with further improvements expected to make it a more favorable option [30][32]. - The GB200 NVL72's architecture allows for faster operations in certain scenarios, such as MoE (Mixture of Experts) models, which could enhance its competitive edge in the market [33].