A U.S. data-center buildout could need about $5.5 per installed GPU-hour at full utilization
A post summarizing Columbia Business School’s “Financing the AI Buildout” puts required mature annual revenue at $3.725 trillion for a central scenario adding 182.7 GW of U.S. data-center capacity between 2025 and 2032.
TLDR
A post describing Columbia Business School’s “Financing the AI Buildout” says its central scenario involves 182.7 GW of additional U.S. data-center capacity between 2025 and 2032. Using the GB300 NVL72 configuration, the post puts that at roughly 1.07 million racks and 77 million GPUs. The required $3.725 trillion in mature annual revenue translates to about $5.5 per installed GPU-hour at full utilization. The post compares that requirement with high-end Nvidia capacity pricing it cites at $6–$10+ per GPU-hour, arguing that the scenario does not require radically higher compute prices to make the arithmetic work.
