Every generation gets one great bet. Your great-great-grandparents watched America lay track across a continent. Your grandparents drove the shiny new interstates.
Your parents waited for a dial-up modem to screech its way online. Each bet changed how you live.
Each one also left a trail of burned investors. Railroad overbuilding helped trigger the Panic of 1873, and the fiber-optic glut of the late 1990s turned plenty of paper millionaires back into regular people.
Most of us think about AI as something that lives on a screen. It’s the chatbot that drafts your emails, or the feature on your phone you never asked for.
That’s the part you see. The part you don’t see is made of concrete, copper, transmission lines and chips, and it’s going up at a pace no living American has witnessed.
Now an economist has put a price tag on that hidden half, and it tops every investment boom in U.S. history relative to the size of the economy.
J Studios / Getty Images
Brookings puts a $10.3 trillion price tag on the AI buildout
Investment in data center buildings, power systems, networking gear and specialized chips will reach $10.3 trillion from 2025 to 2032, an average of 3.63% of U.S. GDP each year, according to Brookings.
The paper comes from Stijn Van Nieuwerburgh, a finance and real estate professor at Columbia Business School. He presented it at the Brookings Papers on Economic Activity conference on Sept. 25.
The buildout is shaping up as the biggest economic bet in U.S. history, reported The Wall Street Journal.
Here’s how the great American booms compare, measured as average yearly spending as a share of GDP per Van Nieuwerburgh:
- Electrification: about 0.5%
- Canals: about 0.66%
- Telecom and fiber (the internet boom): about 1.1%
- Interstate highways (1950s): about 1.1%
- Railroads (1870 to 1890): about 2.2%
AI’s 3.63% runs roughly 50% above the railroad peak and more than triple the internet boom, Van Nieuwerburgh said on the Brookings podcast released Oct. 8.
I ran the $10.3 trillion against the Census Bureau’s count of 134.8 million U.S. households. That works out to roughly $76,400 per household, or about $9,550 a year for eight straight years.
Nobody is mailing you that invoice directly. But a bet that size touches your retirement account, your loan rates, and your power bill, whether you signed up or not.
Why Big Tech’s data center debt is getting harder to see
What struck me most in Van Nieuwerburgh’s work wasn’t the size of the spending. It was how the money was raised.
Tech giants such as Meta Platforms (META) and Microsoft (MSFT) are now spending more on AI than they bring in from operating cash flow, Brookings co-editor Jón Steinsson said on the same podcast. To close the gap, they’re turning to leases, joint ventures and private credit.
More Artificial Intelligence:
- Bill Gates says some jobs should be reserved for humans
- Microsoft’s CEO says AI industry is skipping a step that could matter a lot
- Palantir hits record as its next AI market gets awkward
Van Nieuwerburgh walked through Meta’s Hyperion data center in Richland Parish, La., as his case study. Meta kept a 20% stake, and funds run by Blue Owl Capital (OWL) own the rest.
The joint venture issued about $27 billion of debt against a campus worth roughly $30 billion, he explained. That’s about 90% debt, and none of it sits on Meta’s own balance sheet.
The math gets uncomfortable fast. With that much debt, a 10% drop in cash flow wipes out the equity, and anything worse starts hitting bondholders, according to Van Nieuwerburgh.
Related: OpenAI’s $20 billion revenue gap rattles the AI trade
Meta signed five back-to-back four-year leases rather than one long lease. If it walks away early, it owes a payment that covers the debt, but accounting rules don’t require either obligation on its books today.
The paper’s warning is that these off-balance-sheet structures can “make correlated exposures hard to observe before a downturn,” Brookings noted. Van Nieuwerburgh also estimated that roughly $1.4 trillion of debt has been issued through such structures, with about half held by banks.
Your electric bill is already paying for the construction spree
You don’t need to own a single tech stock to feel this one. Electricity and water costs are each up around 5% over the past year, partly because of AI demand, Federal Reserve Governor Lisa Cook said in a Sept. 28 speech.
In the short term, “AI appears to be adding inflationary pressures to the economy,” Cook said, according to the Federal Reserve.
In my analysis, that turns an abstract number into a household line item. If you pay $150 a month for power, a 5% jump adds about $90 a year, before any future rate increases from your utility.
The Fed also voted in September to raise interest rates by a quarter point, with inflation running 3.8% over the 12 months through August, Cook noted. That keeps pressure on credit card balances, auto loans and adjustable-rate mortgages.
How the spending race could hit your 401(k) and bond funds
If you own an S&P 500 index fund through your 401(k), you already hold a big slice of this bet. The same handful of companies writing the checks dominate those funds.
Newer pockets of risk sit in private credit funds and bond portfolios that buy data center debt. Check what your target-date or income fund holds before you assume you’re out of it.
Big money managers are already adjusting, as Ray Dalio’s Bridgewater reshaped its portfolio amid bubble warnings.
The real question is how a reckoning would land. Van Nieuwerburgh framed it as a choice between an internet-style stock drawdown that stings investors and a 2008-style hit that spreads through banks into every corner of the economy.
He isn’t calling it a bubble. He’s calling for better disclosure while the deals are still being written, and he made the case for optimism, too.
“We should probably be rooting for the success of this technology,” he said.
The railroads went bust more than once and still built modern America. Your job is to make sure your own finances can ride out the bust and still be standing for the payoff.
Related: Trump’s ‘Super Intelligence’ order rolls through the tech world. But will it last?
