A report from Columbia Business School professor Stijn Van Nieuwerburgh suggests that US AI infrastructure spending will reach $10.3 trillion between 2025 and 2032. To justify these investments, the sector must generate approximately $3.7 trillion in annual revenue by 2032, which would account for about 9.2% of the projected US GDP.
The analysis projects the addition of 182.7 gigawatts of data center capacity by 2032. Achieving profitability requires $5.5 in revenue per GPU-hour at full utilization, rising to $6.9 at 80% utilization. These targets assume a 10% unlevered return and a 50% cash-flow margin, necessitating an 80% compounded annual revenue growth rate.
Current combined revenue for OpenAI and Anthropic is estimated at $100 billion. The study warns that hyperscalers like Microsoft, Amazon, and Google are increasingly relying on external debt to fund infrastructure, creating potential systemic risks if demand for AI services fails to keep pace with the rapid expansion of capacity.
Physical constraints, including grid capacity and power generation, remain significant bottlenecks. The report notes that permitting timelines for new energy infrastructure often exceed the requirements of the AI investment schedule, potentially slowing the overall buildout of data centers.
