Financing of AI build out raises risks in US, researcher says
A new study reveals that the AI build-out in the U.S. is projected to require a larger share of the nation's output than past infrastructure expansions, consuming 3.6% of GDP annually through 2032. This massive investment, including 183 gigawatts of new data center capacity, is increasingly reliant on complex financial structures that pose systemic risks. Columbia Business School professor Stijn Van Nieuwerburgh presented these findings, drawing parallels to past economic bubbles and warning about untested revenue streams.
Columbia Business School professor Stijn Van Nieuwerburgh presented a new study at a Brookings Institution conference in Washington, detailing the unprecedented financial demands and potential systemic risks of the artificial intelligence build-out in the United States. The study estimates that the expansion of AI infrastructure, primarily data centers, will consume approximately 3.6% of the U.S. gross domestic product annually through 2032, totaling over $10 trillion. This expenditure surpasses the economic share of historical infrastructure rollouts like railroads, interstate highways, or the internet.
Van Nieuwerburgh highlighted a shift from major tech companies like Amazon, Meta, and Google funding their own expansions with cash stockpiles to relying on increasingly intricate outside financing arrangements involving banks, private credit lenders, and real estate firms. He conservatively estimates this will add 183 gigawatts of new data center capacity over the next seven years. He cautioned that the untested revenue streams for AI services, coupled with high leverage and rapid technological change, could lead to significant downside risk, drawing comparisons to the subprime mortgage crisis due to the opacity of special purpose vehicles.
The financial structure and risks associated with AI infrastructure have become a central point in U.S. political and economic discussions, with Federal Reserve officials examining the boom's impact on inflation and some localities expressing reluctance to host data centers due to concerns about local resource strain. The study suggests that the AI industry, exemplified by companies like OpenAI and Anthropic, would need to generate approximately $3.7 trillion in annual revenue by 2032 to justify the investment, requiring an 80% annual growth rate from current estimates of $100 billion. Despite the risks, Van Nieuwerburgh noted that strong growth and utilization of AI applications could still support the projected infrastructure.