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

Columbia Business School Professor Stijn Van Nieuwerburgh noted that investment in AI infrastructure is unprecedented in scale. Capital expenditures by tech giants such as Meta and Microsoft have exceeded their operating cash flow, and more projects are turning to leases, joint ventures, special-purpose vehicles, and private credit for financing. As a result, some debt does not appear directly on tech companies’ balance sheets.

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

He warned that if AI demand, revenue, or project progress falls short of expectations, highly leveraged projects could face the risk of their equity value being wiped out or creditors suffering losses, with those risks spreading through banks, insurance companies, and private credit institutions into the financial system.

However, he emphasized that it is not yet possible to conclude that AI has formed a bubble, and that there is currently no comprehensive framework for measuring how related risks are distributed across the financial system.

Van Nieuwerburgh recommends strengthening disclosures of off-balance-sheet debt and contingent liabilities, and tracking project asset ownership, debtors, leases, and guarantee obligations to improve financing transparency. He believes regulatory efforts should focus not on preventing AI investment, but on identifying potential concentrations of risk and avoiding the AI investment boom evolving into systemic financial risk during a downturn.