Stijn Van Nieuwerburgh, an economist at Columbia Business School, published a paper in the Brookings Papers on Economic Activity regarding the future of AI infrastructure. The research estimates that US investment in AI infrastructure will reach $10.3 trillion between 2025 and 2032, averaging 3.63% of annual GDP.
To justify this level of spending with a 10% unlevered return benchmark, the AI sector must generate approximately $3.725 trillion in annual revenue by 2032. This figure would represent about 9% of the projected US GDP for that year. Currently, OpenAI and Anthropic report a combined annual revenue of $100 billion.
Achieving the required revenue target necessitates a sustained annual growth rate of roughly 80% through 2032. The research also projects an addition of 183 GW of computing capacity by 2032, which is equivalent to doubling current residential electricity consumption in the United States.
The paper highlights that much of the current infrastructure expansion relies on off-balance-sheet financing. This structure obscures the distribution of risk and creates potential vulnerabilities. The study notes that the scale of investment is large enough to impact overall US economic growth figures.
