巴克莱：市场最大“黑天鹅”，AI资本支出放缓，三大“巨雷”会是美股噩梦

Core Viewpoint - Barclays' stock strategy team indicates that a 20% decline in data center capital expenditures over the next two years could lead to a 3-4% earnings pressure on the S&P 500 index, with a more severe potential valuation drop of 10-13% [4][24]. Group 1: Risks Identified - The report identifies three major risks that could trigger a crisis: 1. Efficiency Risk: Rapid improvements in AI model efficiency may lead to overbuilding of computing facilities, reminiscent of the "dark fiber" tragedy during the dot-com bubble [5][11]. 2. Physical Limitation Risk: Increasing electricity shortages are becoming a hard constraint on data center construction, potentially cooling capital expenditures [6][12]. 3. Liquidity Risk: If capital expenditures continue to outpace cash flow generation, financing pressures may arise, exacerbated by dwindling venture capital [7][18]. Group 2: AI Investment Landscape - Despite the potential slowdown in data center capital expenditures, the foundation for AI investment remains solid, with a projected annual growth of 30% in capital expenditures [8][9]. - The demand for computing power continues to exceed supply, driven by the proliferation of advanced reasoning models and AI agents [9][11]. Group 3: Economic Impact - Data center-related investments are projected to contribute approximately 1 percentage point to the 1.4% GDP growth in the first half of 2025 [19]. - If the U.S. economy enters a recession due to other factors, the simultaneous slowdown in AI capital expenditures could act as an "accelerator," worsening the situation [21][22]. Group 4: Valuation and Earnings Impact - The impact on earnings (EPS) from a 20% decline in data center capital expenditures is expected to be relatively mild, with a 3-4% drag on the S&P 500 index's EPS for the fiscal year 2026 [24]. - However, the valuation (P/E) impact is projected to be severe, leading to a 10-13% compression in the overall S&P 500 index [25]. - For industries directly benefiting from AI infrastructure, the average P/E compression could reach 15-20% [27].