Finn Brooks: There's a methodology question buried inside the biggest financial claim you'll hear this year — and getting it wrong changes whether the claim holds up at all.
Finn Brooks: The Wall Street Journal is saying the AI data center buildout is the biggest economic bet in U.S. history. That framing comes from a Brookings Institution working paper — Stijn van Nieuwerburgh, Columbia Business School — puts the total at $10.3 trillion in U.S. data center and AI infrastructure investment, 2025 to 2032.
Finn Brooks: Big number. Sure.
Finn Brooks: The methodology is what actually earns it, though — because the obvious objection is: inflation makes huge present-day dollar figures look bigger than comparable things in the past. You can't just say ten trillion and compare it to the interstate highway system without doing the math honestly.
Finn Brooks: Van Nieuwerburgh uses GDP share. Not raw dollars — what percentage of the whole economy this represents, annually, across the investment window.
Finn Brooks: 3.63% of GDP per year. On average.
Finn Brooks: The railroad era — the thing we hold up as the definitive American infrastructure transformation — peaked at around 2.2% of GDP. That's the previous record. The AI buildout clears it by a lot, and clears every other boom too — electrification, highways, all of it.
Finn Brooks: Once you accept that methodology, the Journal framing isn't a reach. It's just what the data says.
Finn Brooks: And who's driving it — five hyperscalers. Microsoft, Amazon, Meta, Alphabet, Oracle. FactSet projects those five alone spend $4.2 trillion through 2029. The rest of the $10.3 trillion comes from the broader ecosystem, but those five are the center of gravity.
Finn Brooks: You can already see it on the ground. Data center construction is at $85 billion annually — more than doubled in two years. Meanwhile all other U.S. construction value has dropped roughly 8%.
Finn Brooks: That 8% drop is not nothing.
Finn Brooks: That's what crowding-out looks like in the early innings — capital and labor tilting toward one sector so hard that everything else starts to feel it. Whether the thing they're all tilting toward actually pays off… that's the whole question.
Finn Brooks: Here's where I want to put you for a second — Piketon, Ohio, March 2026.
Finn Brooks: SB Energy — that's a SoftBank Group company — breaks ground on a $67.2 billion AI data center program. U.S. officials are calling it potentially the world's largest AI data center. One facility. One site. Sixty-seven billion dollars.
Finn Brooks: That number lives in your body differently when it's attached to a place.
Finn Brooks: And then Michael Cembalest over at JP Morgan Asset Management publishes this analysis — he calls it 'Smothering Heights,' which, respect — and what he shows is that 2025 tech capital spending as a share of GDP roughly equals the combined cost of the Manhattan Project, farm electrification, the Apollo program, the Interstate Highway System, and FDR-era public works. All of them. Together. This year.
Finn Brooks: Just sit with that.
Finn Brooks: Here is the part that keeps me up — and I mean that pretty literally.
Finn Brooks: For this buildout to make financial sense — like, for the numbers to actually close — the AI industry would need to be generating roughly $3.7 trillion in annual revenue by 2032. That's what the investment level demands. That's the math.
Finn Brooks: Current major-player revenues. Right now. About $100 billion.
Finn Brooks: That's a 37x gap. In seven years. And look — I'm genuinely not smart enough to say that's impossible, revenue curves do wild things — but that's not a rounding error, that's a different reality.
Finn Brooks: This is what the revenue justification gap actually means — it's not just that the bet is big, it's that the structural mismatch between what's being built and what must be EARNED is almost incomprehensible at this scale.
Finn Brooks: And here's what makes it different from every other mega-project in that Cembalest list — the Interstate Highway System, the Apollo program, all of it — those were government bets. Public money. A sovereign backstop.
Finn Brooks: This buildout is almost entirely privately funded.
Finn Brooks: No government backstop. No public accountability structure. If the revenue gap doesn't close — if Microsoft, Amazon, Meta, Alphabet, Oracle, SoftBank, all of them have misjudged the curve — the risk doesn't sit with a government that can absorb a loss. It sits with capital markets, and whoever is downstream when the music stops.
Finn Brooks: And that's — I mean, that's the actual stakes of what we're watching get built in Piketon, Ohio right now.
Finn Brooks: What do you actually watch from here — like, what's the number that tells you whether this corrects or cascades?
Finn Brooks: For me it's still that $3.7 trillion. That's the revenue target by 2032 that makes the math close. Against $100 billion today. That gap doesn't have to be a death sentence — revenue curves do strange fast things — but it does have to start MOVING, and soon, and visibly.
Finn Brooks: And then there's inflation sitting on top of all this. Because most of this capex isn't sitting in a vault — it's debt-financed. Microsoft, Amazon, Meta, Alphabet, Oracle — they're borrowing against future returns. Hot inflation raises the cost of that debt in real time. We've already seen it happen: financing costs spike, AI stocks sell off sharp. Not a prediction. That has already occurred.
Finn Brooks: Inflation isn't a tail risk here. It's a live accelerant on a fire that's already lit.
Finn Brooks: I want to be honest with myself for a second, because I've been leaning pretty hard into the van Nieuwerburgh framing and I think that's worth interrogating. The $10.3 trillion is a projection from one Brookings Institution working paper. Rigorous, yes. But other forecasts differ. And the GDP-share methodology — which is why we can call this 'the biggest ever' — that is a methodological choice. A defensible one, but a choice. Someone could contest it.
Finn Brooks: The revenue gap is actually the truer stress test. That one doesn't depend on how you measure history — it just depends on whether the money comes in.
Finn Brooks: Which is what makes the construction data so unsettling. $85 billion a year, doubling in two years — that's happening regardless of whether the revenue gap is closing. The physical commitment keeps compounding while the earning side is still theoretical.
Finn Brooks: That's the signal I'd watch most closely — QTS is pouring concrete in Eagle Mountain, Utah right now. SB Energy broke ground in Piketon. The ground is broken. The question is what's earning.
Finn Brooks: And look — here's the actual tell, the one you can watch in real time. Sometime around 2026, 2027, Microsoft, Amazon, Meta, Alphabet, Oracle are going to issue capex guidance for the following year. If any one of them meaningfully moderates — not pauses, just moderates — that is the first crack in the arms-race logic. That's the moment where one player decided the math stopped working.
Finn Brooks: Because right now none of them can blink first. That's the arms-race trap — you can't stop even if you're uncertain, because stopping signals weakness and the others keep building. So a moderation in guidance isn't a routine update. It's enormously informative.
Finn Brooks: And if nobody blinks — if the revenue gap persists and the building continues — then the question becomes who is holding stranded assets, and which financial institutions absorbed the debt on the way down. That's not catastrophizing. That's just the arithmetic of what happens when a $10.3 trillion bet doesn't close.
Finn Brooks: And the stranded asset question is the one that nobody really wants to sit with — because it's uncomfortable in a specific way. It's not abstract. If the revenue gap doesn't close, you don't end up with nothing. You end up with $10.3 trillion in partially useful infrastructure — servers, cooling systems, transmission lines, buildings — owned by someone, financed by someone, sitting on someone's balance sheet. The question is whose.
Finn Brooks: That's the part the private funding structure makes genuinely strange. Because when the Interstate Highway System turned out to be more expensive or more complicated than projected — the government absorbed it. A sovereign can hold a bad bet. It doesn't mark to market. But Microsoft and Amazon and Meta and Alphabet and Oracle — they borrowed against future returns. The debt is real and the clock on it is real. And the institutions that absorbed that debt on the way up — whoever bought those bonds, extended those credit lines — they're holding it too. If the music stops, it doesn't stay contained to the hyperscalers. It moves.
Finn Brooks: The ground is already broken in Piketon, Ohio. QTS is already pouring concrete in Eagle Mountain, Utah. That part is done. The bet is placed and it cannot be unplaced. So the only live question — the only one that actually matters right now — is whether $100 billion in annual revenue becomes $3.7 trillion before the debt comes due.
