高盛：全球经济评论：为何人工智能支出未能推动 GDP 增长

Investment Rating - The report does not explicitly provide an investment rating for the industry but highlights significant revenue growth in AI infrastructure-related companies, indicating a positive outlook for investment opportunities in this sector [2][3]. Core Insights - There is a notable divergence between the annualized revenue growth of public companies involved in AI infrastructure, which increased by over $340 billion from 2022 to Q4 2024, and the real investment in AI-related categories in the US GDP, which only rose by$42 billion during the same period [2][3]. - The report suggests that a substantial portion of the revenue increase is attributed to cost inflation and foreign revenue, which do not contribute to real GDP growth [6][9]. - Methodological issues in the US national accounts may lead to an underestimation of the impact of AI-related investments on real GDP, with an estimated potential understatement of around $100 billion [2][10]. Summary by Sections Revenue Growth - Annualized revenue for public companies exposed to AI infrastructure build-out increased by over$340 billion from 2022 through Q4 2024, with projections suggesting a further increase to $580 billion by the end of 2025 [2][3]. - The increase in real investment in AI-related categories in the US GDP accounts was only$42 billion during the same period, raising questions about the apparent disconnect between revenue and GDP growth [3][4]. Cost Inflation and Foreign Revenue - A significant portion of the revenue increase is driven by cost inflation, particularly in semiconductors, and foreign sales, which account for nearly half of the reported AI spending surge [6][9]. - Margin expansion is estimated to explain around $30 billion of the overall revenue increase, with foreign revenue contributing approximately$130 billion [9][17]. Methodological Considerations - The report discusses the commodity-flow approach used in US national accounts, which may misclassify semiconductor purchases as intermediate inputs rather than investments, leading to an underrepresentation of actual investment in GDP [11][12]. - The surge in cloud services used for AI model training is also likely underreported in GDP calculations due to their classification as intermediate inputs [14][15]. - The report concludes that while there is potential for AI-related investment to provide a moderate boost to real US GDP in 2025, much of the investment in semiconductors and cloud computing will likely remain unmeasured unless there are changes in national accounting methodologies [22].