Key Highlights

  • A new study estimates the AI buildout will require more than USD 10 trillion and consume about 3.6% of annual US GDP through 2032, exceeding the relative cost of prior technology rollouts including railroads and the interstate highway system.
  • The financing structure behind AI infrastructure has shifted from company cash reserves to increasingly complex arrangements involving banks, private Credit lenders and real estate firms.
  • The study's author compared the opacity of financing vehicles used in the buildout to arrangements seen before the 2007-2009 financial crisis.
  • The AI industry would need to generate about USD 3.7 trillion in annual Revenue by 2032 to justify the projected investment, versus roughly USD 100 billion in combined revenue today for two of its largest developers.

A Buildout Bigger Than Prior Technology Waves

The artificial intelligence buildout is on track to consume a larger share of US economic output than the rollout of electricity, railroads, interstate highways or the internet, according to a new study, and its increasingly complex financing structure could pose risks to the broader financial system. The research, prepared by Columbia Business School finance and real estate professor Stijn Van Nieuwerburgh for a Brookings Institution conference this week, estimates the buildout will consume around 3.6% of US gross domestic product annually through 2032, or more than USD 10 trillion in total investment.

That estimate exceeds the 2.2% of annual GDP absorbed by the railroad expansion of the late 1800s, previously the most costly rollout of a general-purpose technology, and is well above the roughly 1% annual GDP share tied to the construction of the US interstate highway system or the telecommunications buildout of the mid-1990s.

From Corporate Cash to Complex Financing

What began as an expansion funded largely out of the cash reserves of large technology companies has evolved into a buildout reliant on increasingly intricate financing arrangements involving banks, private credit lenders, real estate firms and other participants. Van Nieuwerburgh estimated the current pipeline at roughly 183 gigawatts of new data-center capacity planned over the next seven years, compared with about 57 gigawatts currently installed.

"This is freaking complicated," Van Nieuwerburgh said in a briefing with reporters, describing the web of financing arrangements now underpinning the industry's expansion.

Echoes of the Subprime Mortgage Crisis

The study drew a direct parallel between the opacity of special purpose financing vehicles now used to fund AI infrastructure and the complex mortgage financing arrangements that preceded the 2007-2009 financial crisis. Van Nieuwerburgh cautioned that the shift toward outside financing has increased Leverage and redistributed risk across the economy, making the buildout dependent on revenue streams that remain largely unproven. He stressed that a downturn is not inevitable, noting that continued strong growth in AI applications and improving model capability could support the Investment and generate stable cash flows over time.

"The combination of uncertain demand, rapid technological change, execution bottlenecks, and high leverage creates meaningful downside risk if expectations are revised," the study found.

A Steep Revenue Bar to Clear

To justify the scale of the projected investment, the study estimates the AI industry will need to generate about USD 3.7 trillion in annual revenue by 2032. Given current combined annual revenue of roughly USD 100 billion for two of the industry's largest developers, that would require sustained growth of approximately 80% per year for several years running.

A Growing Political and Economic Flashpoint

Data center construction and the risks tied to the AI buildout have become an increasingly prominent issue in US political and economic debate. Some local communities have grown reluctant to host new facilities amid concerns about strain on electricity and water resources, while Federal Reserve officials have begun weighing whether the construction boom is contributing to broader Inflation pressure. Some AI industry executives have separately suggested that a slower pace of development might carry lower risk.

Outlook

Whether the AI buildout's financing structure proves resilient will likely hinge on how quickly revenue from AI applications scales relative to the enormous Capital already committed, a gap the study's author describes as historically wide at this stage of a technology rollout. The research is scheduled to be presented in full on Friday.