The gargantuan push to develop infrastructure to support a world driven by artificial intelligence is expected to be the most expensive build-out in American history. 

It will cost roughly $10.3T in investment capital between 2025 and 2032 — roughly 3.6% of U.S. gross domestic product per year — to fund the massive AI build-out already in the pipeline, a new paper from Columbia University economist Stijn Van Nieuwerburgh and the Brookings Institution found. 

Tech firms and developers are turning to increasingly complex and opaque financing mechanisms to get projects built, a practice that will only accelerate as the cycle continues. The run-up in financing from Big Tech firms is ballooning at a pace that will require them to generate enormous revenue growth in the coming years to keep up with their debt service. 

“Silicon Valley wants all of us to believe that this is a miracle technology, it’s going to generate trillions of dollars of revenues — and it has to generate trillions of dollars of revenues to be financeable,” Van Nieuwerburgh said. “I'm sure there is a state of the world where that happens. I'm just not sure how likely it is.” 

The sheer scale of the AI build-out is unprecedented in American history. The next-largest capital expenditure boom came from 1870 through 1890, when the U.S. directed an average of 2.2% of gross domestic product toward the build-out of the country’s railway network, the analysis found. 

The projected cost of the AI build-out is more than three times what was spent to build out America’s highway system and six times more than was spent on electrification at the turn of the 20th century.

The paper, presented as part of the Brookings Institution’s semiannual academic conference, estimates the $10.3T price tag will pay for an additional 183 gigawatts worth of computing power by 2032. 

The estimates for project completions are in some ways conservative, with project-level data suggesting the pipeline totals 509 GW, including compute power that will come online after 2032. In the analysis, Van Nieuwerburgh assumed 227 GW of proposed capacity will never be built, while another 117 GW will come online after 2032. The paper assumes it costs roughly $8.2B for every 200 megawatts of compute power built.

The major hyperscalers — Oracle, Amazon, Alphabet, Microsoft and Meta — have grown their capex from $97B in 2020 to more than $400B in 2025. They are projected to clear $800B in 2026, more than the firms’ combined operating cash flow.

Demand is outstripping supply for computing power, but historical precedent makes it likely that dynamic will one day flip, even if that is several years away, Van Nieuwerburgh said. The current landscape is well positioned to support an increased flood of debt, with Van Nieuwerburgh estimating roughly five to eight years of strong growth before oversupply becomes a concern. 

“If history is a guide, credit constraints will loosen, more speculative development will take place, more marginal compute will be built, and sooner or later we're going to have oversupply, just like we do in every real estate cycle, and then the prices will collapse,” he said. “I don't see why this one time is different.”

There are also unique hurdles to the explosion in capex. The large number of massive single-tenant facilities presents a credit risk for lenders if one of the hyperscalers faces its own liquidity crunch. New data center hardware is being developed so quickly that new builds can quickly become obsolete. Long and complex development timelines with bottlenecks around power procurement make delivering projects on time a challenge. 

At least $1.3T in debt has already been committed to underwriting the data center boom, and the sources of capital are increasingly diverse as private credit pushes in along with insurers and pension funds. Banks hitting lending ceilings are using syndication and other financing vehicles to keep deploying capital into the space. 

It is a significant amount of debt, but still less than half the $3T that was tied up in the subprime mortgage crisis that led to the Great Recession, Van Nieuwerburgh said. 

“It's systemic in the sense that every bank is now involved in this and is actually having a lot of concentrated exposure to AI, but in terms of its magnitude, it's still relatively modest,” he said. 

But the deep development pipeline and the $10T price tag that comes with it will balloon the total debt underwriting the AI build-out. How that exposure is distributed through the economy will determine what a downturn looks like, Van Nieuwerburgh said. 

The increasingly complex network of financing structures that developers, tech firms and capital sources are leveraging to finance new construction is making it harder to track exposure in the marketplace, with private credit and off-balance-sheet structures adding a layer of opacity that conceals where risk lies. 

The Brookings research comes as the biggest names in AI call for a slowdown in the tech’s rapid development following several security incidents in which AI agents went rogue and hacked into competitors’ systems. 

That, along with growing pushback from local communities to data centers being built in their neighborhoods, may help the tech sector and broader U.S. economy from pushing too much capital into the development cycle too quickly, Van Nieuwerburgh said.

“We're in the midst of a massive construction boom, and a little bit of gradualism would actually serve us very well,” he said. “It would actually preserve some of the pricing power for the existing data centers.”