本报告援引哥伦比亚商学院Stijn Van Nieuwerburgh教授的研究，分析美国AI投资热潮及由此带来的金融风险。报告指出，训练和部署前沿AI模型需要快速积累数据中心、电力系统、冷却系统、网络设备及专用芯片等实物资本，当前AI投资热潮更像历史上的铁路、电气化、电信等基础设施投资热潮，而非科技公司的常规资本支出。五大AI公司合计资本支出从2020年约970亿美元增至2025年超4000亿美元，2026年预计将超8000亿美元，首次超过其合计经营性现金流，融资来源转向租赁、合资、项目债、私人信贷、证券化及特殊目的载体（SPV）。报告测算，2025—2032年建设合计183GW的AI园区成本约9.61万亿美元，2032年需运营现金流1.86万亿美元、年销售收入3.725万亿美元（约占当年美国GDP的9.2%）；
According to a recent paper by Columbia Business School Professor Stijn Van Nieuwerburgh, training and deploying frontier AI models requires the rapid accumulation of physical capital, such as data centers, power systems, cooling systems, network equipment, and specialized chips. Today's AI investment boom is more similar to past infrastructure investment booms in railroads, electrification, and telecommunications than to typical capital investments by high-tech companies. Professor Nieuwerburgh projects that the aggregate capital expenditures by the five major AI companies, which expanded from about $97 billion in 2020 to over $400 billion in 2025, will exceed $800 billion in 2026, surpassing their combined operating cash flow for the first time. The main sources of financing are shifting to leasing, joint ventures, project debt, private credit, securitization, and special-purpose vehicles (SPVs), thus increasing reliance on external capital. Building AI campuses with an aggregate cap