The Brookings Institution projects that investment in AI-related infrastructure in the United States will reach $10,3 trillion between 2025 and 2032. This spending is expected to average 3,63% of US GDP annually, exceeding historical investments in railroads, the interstate highway system, and early electrification.
The research, authored by Stijn Van Nieuwerburgh from Columbia University, allocates these funds across data centers, power systems, network infrastructure, and specialized hardware. Hyperscaler capital expenditures have grown from $97 billion in 2020 to over $400 billion in 2025, with 2026 projections reaching $800 billion.
Amazon, Alphabet, Microsoft, Meta, and Oracle are identified as the primary drivers of this growth, with their combined capital expenditures estimated at $4,2 trillion through 2029. The study anticipates an increase in compute power capacity of 183 to 509 gigawatts by 2032, depending on demand and deployment speed.
Financing models are shifting from traditional on-balance-sheet expenditures toward complex off-balance-sheet structures, including joint ventures, private credit, and securitizations. The report notes that this transition introduces new operational risks and uncertainties regarding future AI demand.
