摩根大通：明年数据中心资本开支增长将超50%！AI相关股票盈利预期被低估了

Core Viewpoint - JPMorgan has significantly raised its forecast for data center capital expenditure growth, expecting it to exceed 50% in 2026, up from a previous estimate of 30%, indicating an additional spending of over $150 billion next year, which has not yet been factored into Wall Street's AI revenue projections [1][4][5]. Group 1: Capital Expenditure Forecast - The capital expenditure growth rate for data centers in 2025 has been revised upward from 55% to approximately 65%, driven by large cloud service providers increasing their investments in AI infrastructure due to a supply-demand mismatch in AI computing power [5][6]. - For 2025, the incremental spending is expected to exceed $115 billion compared to 2024, significantly higher than the $75-80 billion increase projected for 2024 [5]. - The total capital expenditure for the four major U.S. cloud service providers is projected to reach approximately $363 billion in 2025, representing a year-on-year growth of about 65%, and is expected to rise to around $447 billion in 2026 [6]. Group 2: Underestimated Revenue Potential - Analysts have not fully accounted for the upcoming $150 billion to $175 billion in new capital expenditure when forecasting revenues for core AI infrastructure suppliers like Nvidia, Broadcom, AMD, and Marvell [7]. - There is a strong and urgent demand for AI computing, and if the data center capital expenditure reaches a growth rate of over 60% in 2026, it would lead to upward revisions in earnings expectations for these chip giants [7]. Group 3: Order Backlogs and Emerging Buyers - The report highlights that the market has misinterpreted the backlog value of companies like Broadcom and Nvidia, underestimating the speed at which these backlogs will convert into actual revenue [9]. - Investors often focus solely on the top four or five U.S. cloud providers, neglecting significant spending from emerging players such as neoclouds and sovereign AI projects, which are becoming new pillars of demand for AI chips [9].