Key facts
- AI infrastructure buildout is projected to consume 3.6% of US GDP annually through 2032, exceeding $10 trillion.
- The cost is higher than previous technological rollouts like railroads (2.2% of GDP) and the internet (over 1% of GDP).
- The financing involves AI firms, hyperscalers, banks, private credit lenders, and real estate firms.
- An estimated 183 gigawatts of new data center capacity will be built over the next seven years.
- The AI industry needs to earn $3.7 trillion annually by 2032 for projected investment returns.
- Current annual revenues for OpenAI and Anthropic are estimated at around $100 billion.
The rapid expansion of artificial intelligence infrastructure in the US is on track to consume a significant portion of the nation's output, employing complex financing structures that could pose systemic risks, according to a new study by Stijn Van Nieuwerburgh, a finance and real estate professor at Columbia Business School. The buildout is projected to require around 3.6% of US gross domestic product annually through 2032, totaling more than $10 trillion. This is a larger share of GDP than previous major technological rollouts, including railroads (2.2% of GDP in the late 1800s), the US interstate highway system (about 1% annually), and the telecommunications expansion (about 1% annually).
Van Nieuwerburgh noted that the financing for this AI buildout has become increasingly intricate, involving AI firms, major tech companies like Amazon, Meta, and Google, banks, private credit lenders, and real estate firms. This shift from companies funding projects with their own cash reserves to relying on outside financing has increased leverage and redistributed risks across the economy. The venture is dependent on revenue streams that have yet to be proven, drawing comparisons to the opacity of special purpose vehicles seen during the subprime mortgage crisis.
While the researcher stated that financial distress is not imminent and strong AI growth could support the projected infrastructure, he cautioned that the combination of uncertain demand, rapid technological change, execution bottlenecks, and high leverage creates "meaningful downside risk" if expectations are not met. For instance, the AI industry would need to generate approximately $3.7 trillion in annual revenue by 2032 to achieve the expected return on investment. Given current estimates of combined annual revenues for OpenAI and Anthropic at around $100 billion, this would require revenues to grow at roughly 80% per year.