Big Tech's $635 Billion AI Spending Plans Face Energy Supply Test, S&P Global Warns

The $635 billion in AI infrastructure spending planned by major technology companies faces a significant stress test from energy supply constraints, S&P Global has warned in an analysis reported by Reuters on Tuesday. The scale of capital expenditure being deployed by Amazon, Microsoft, Google, and Meta on data centers, chips, and AI infrastructure has strained power grid capacity in key markets. S&P Global's analysis flags that energy availability, not capital, is increasingly the binding constraint on the AI build-out timeline. Data centers supporting large language model training and inference require substantial and reliable power, typically between 100 megawatts and one gigawatt per hyperscale campus. The pace of new facility construction has outrun the ability of utilities to bring new generation capacity online, leading to multi-year queues for power interconnection in regions including Northern Virginia, Texas, and parts of the U.K. and Ireland. S&P Global's warning adds to a growing body of analysis suggesting that the AI infrastructure cycle, while real, faces physical constraints that financial capital cannot immediately solve. Grid upgrades, new transmission lines, and additional generation capacity from nuclear, natural gas, and renewables all involve regulatory processes and construction timelines measured in years. Energy costs also directly affect the economics of AI services. As power prices rise in data center markets due to demand pressure, the cost of inference is elevated, which bears on the profitability of AI products sold at flat per-token or subscription rates. The energy constraint dynamic has accelerated corporate investment in dedicated power arrangements, including Microsoft's reactivation of Three Mile Island's nuclear reactor and Amazon's direct power purchase agreements with nuclear operators.