The AI Buildout Could Cost $10 Trillion. A Researcher Is Warning It Could Create Systemic Risks.

The U.S. artificial intelligence boom is rapidly becoming one of the largest infrastructure investments in American history. Still, the increasingly complex financial machinery funding it could create risks that extend well beyond Silicon Valley.

A new study by Columbia Business School professor Stijn Van Nieuwerburgh, prepared for a Brookings Institution conference ​this week, estimates that investment in AI infrastructure could reach roughly $10.3 trillion between 2025 and 2032, averaging about 3.6% of U.S. gross domestic product each year.

That would make the AI expansion larger relative to the economy than the historic buildouts of innovative technologies like railroads, electricity, interstate highways, and telecommunication lines.

The railroad expansion of the late 1800s, previously one of the largest U.S. infrastructure booms, absorbed about 2.2% of annual GDP. Construction of the interstate highway system beginning in the 1950s and the telecommunications expansion associated with the internet each accounted for roughly 1% annually.

Van Nieuwerburgh argued that how the AI boom is being financed may be just as important as its scale. Early investment could largely be funded directly by cash-rich technology giants such as Amazon, Meta and Google‘s parent Alphabet.

As spending accelerates, however, the industry is increasingly turning to outside capital and more complicated financial structures involving hyperscalers, data-center developers, banks, private-credit firms, real estate companies and other investors. “This is freaking complicated,” Van Nieuwerburgh told reporters while describing the emerging financing network.

His study estimates the U.S. could add about 183 gigawatts of data-center capacity through 2032, compared with roughly 57 gigawatts currently installed. The infrastructure needed goes far beyond server buildings. AI facilities require enormous investments in specialized chips, networking equipment, cooling systems, and electricity generation and transmission.

To fund those projects, companies are increasingly relying on joint ventures, private credit, securitizations, leases, loan guarantees and special-purpose vehicles, or SPVs. Those arrangements can expand the amount of capital available, but they can also make it harder for investors and regulators to determine where financial risks ultimately reside.

Van Nieuwerburgh compared that opacity with one of the most consequential financial episodes of recent history. “This opacity of all these special purpose vehicles is somewhat reminiscent of what happened in the subprime mortgage crisis,” he said.

However, the comparison does not mean another financial crisis is inevitable. Van Nieuwerburgh explicitly cautioned that it would be premature to conclude that AI infrastructure already represents a systemic threat comparable with previous credit booms. Strong demand for AI services, high data-center utilization and continued improvements in AI models could ultimately produce enough revenue to support the investments.

But the financial assumptions behind the expansion are becoming increasingly demanding. Van Nieuwerburgh estimates that the AI industry would need to generate approximately $3.7 trillion in annual revenue by 2032 to produce the expected returns on the infrastructure being built.

With combined annual revenue estimates for OpenAI and Anthropic currently around $100 billion, that would require revenue to grow at roughly 80% annually, according to his analysis.

If demand falls short, technology changes faster than expected, or some data centers become obsolete before generating projected returns, losses could travel through the increasingly interconnected network of lenders, developers and investors financing the boom.

“These developments do not imply that financial distress is imminent,” Van Nieuwerburgh wrote. But he warned that uncertain demand, technological change, construction bottlenecks and high leverage create “meaningful downside risk” if expectations are revised.