AI infrastructure investment accounts for an average of 3.6% of U.S. GDP annually, with off-balance-sheet financing hiding $1.4 trillion in debt risks.

Beating AI News Flash: According to the latest research from the Brookings Institution, from 2025 to 2032, total investment in U.S. AI data center construction is projected to reach $10.3 trillion, with average annual investment accounting for approximately 3.63% of U.S. GDP — about 1.6 times the peak of 19th-century railway construction and more than 3 times the internet and fiber-optic buildout boom.

Stijn Van Nieuwerburgh, a professor at Columbia Business School, pointed out that the scale of AI infrastructure investment is unprecedented. Capital expenditures by tech giants such as Meta and Microsoft have exceeded operating cash flow, with a growing number of projects turning to leasing, joint ventures, special purpose vehicles, and private credit financing. As a result, some of this debt is not directly reflected on tech companies' balance sheets.

Taking Meta's Hyperion data center project as an example, approximately $30 billion in project assets correspond to $27 billion in debt, bringing the project-level debt-to-asset ratio to 90%. Van Nieuwerburgh estimates that related financing structures have issued approximately $1.4 trillion in debt, of which about $700 billion to $800 billion is held by banks.

He warned that if AI demand, revenue, or project timelines fall short of expectations, highly leveraged projects could face equity values dropping to zero or even debt impairment, transmitting risks to the financial system through banks, insurance companies, and private credit institutions. However, he emphasized that it is currently premature to conclude that AI has formed a bubble, and there is also a lack of a complete framework at this stage for measuring how related risks are distributed across the financial system.

Van Nieuwerburgh recommended strengthening disclosure of off-balance-sheet debt and contingent liabilities, and tracking project asset ownership, debt obligors, lease agreements, and guarantee obligations to improve financing transparency. He believes the regulatory focus should not be on blocking AI investment, but rather on identifying potential risk concentrations to prevent the AI investment boom from turning into systemic financial risk during a downturn.