Source: Marginal REVOLUTION
U.S. tech giants have shifted data-center infrastructure from a routine operational expense into a multi-hundred-billion-dollar annual capital commitment, driven by AI model training and deployment demands that dwarf previous generations of compute needs. This spending concentration—where a handful of companies (Microsoft, Google, Amazon, Meta) control the bulk of new capacity buildout—is affecting real estate markets, power grids, semiconductor supply chains, and geopolitical semiconductor policy. Whoever builds the most efficient data centers at scale will control the compute bottleneck for the next decade of AI applications.