Analysis projects cumulative US AI infrastructure investment reaching $10 trillion USD through 2032 and identifies systemic financial risks from capital concentration and debt service.
The scale of investment required to build out AI infrastructure in the US could exceed $10 trillion by 2032, according to research by Stijn Van Nieuwerburgh, a professor of finance and real estate at Columbia Business School. This magnitude of spending, coupled with increasingly complex financing arrangements, poses significant risks to the financial system.
Annual spending on AI infrastructure is projected to reach approximately 3.6% of US GDP by 2032—a share that exceeds historical precedent. Railroad construction in the second half of the 19th century accounted for approximately 2.2% of annual GDP. The interstate highway system, built from the 1950s onward, required just over 1% of GDP per year, as did telecommunications development from the mid-1990s onward.
The financing structure underlying this expansion has fundamentally shifted. Initially, technology companies funded infrastructure investments from their own revenues. Today, spending has outpaced even the largest companies' capacity to self-fund, forcing the industry to rely increasingly on external financing from banks, private credit funds, real estate firms, and cloud providers. This distributed financing approach adds to overall debt burdens and spreads financial risk across multiple sectors of the economy.
Over the next seven years, the industry will need 183 GW of data center capacity—more than three times the approximately 57 GW currently installed in the US. The complexity of the financing structures required to deliver this capacity poses particular concerns. Van Nieuwerburgh warned that such arrangements can obscure how risks are actually distributed, drawing parallels to the mortgage crisis, when complex financial products contributed to the 2007–2009 recession and ensuing turmoil in global financial markets.
Van Nieuwerburgh noted that a financial crisis is not inevitable under current trends. Sustained demand for AI, widespread adoption of its systems, and continued improvements to models could provide infrastructure projects with stable revenues. However, significant risks remain. Uncertainty about AI demand, rapid technological change, project completion difficulties, and high debt levels create the potential for substantial losses if investor expectations are not met.
For current infrastructure investments to deliver expected returns, the AI industry must generate approximately $3.7 trillion in annual revenue by 2032. OpenAI and Anthropic currently have combined annual revenues of about $100 billion, according to estimates cited in the research. Reaching the required revenue level would demand approximately 80% annual growth.
The construction boom has emerged as a subject of political and economic debate. Some communities are reluctant to approve new data center facilities due to concerns about strain on local resources. Federal Reserve officials are assessing whether the construction surge is contributing to inflation. Meanwhile, some executives within the AI industry have suggested that a slower pace of growth could be safer for the broader economy.