Columbia model: up to 45% of US AI data center projects may not be built

A Columbia professor's model suggests that even 45% of US AI infrastructure projects, worth $9.3B, may not be realized, citing overcapacity or financial constraints. The claim is based on a single academic model, not yet corroborated by industry data. If accurate, this could signal a slowdown in AI capex, affecting tech supply chains and energy demand forecasts.