A September 2026 Brookings conference draft by Columbia Business School economist Stijn Van Nieuwerburgh estimates that a 1 GW AI campus costs about $41 billion. It projects 183 GW of U.S. AI data-center capacity completed by 2032, with another 118 GW completed later, and estimates roughly $10.3 trillion in investment over 2025–2032, averaging about 3.63% of GDP per year.
The paper describes a financing shift as hyperscaler capital expenditure outgrows internal cash flow: leases, joint ventures, project finance, private credit and structured-finance vehicles increasingly complement direct ownership. Such arrangements can expand funding capacity, while placing leverage and exposure at the asset or project level. The author identifies risks tied to uncertain AI demand, technological obsolescence, power and execution delays, residual values, and concentration among a small number of tenants.
These figures are the paper’s estimates and projections, not a confirmed buildout or a forecast that every planned project will be completed. The draft analyzes the financing architecture and its risks; it does not establish whether future AI revenues will be sufficient to justify the investment.