Big Tech's real AI spending is $3 trillion higher than the headline capex numbers show

[SilverRider]

A Wall Street Journal analysis of footnotes buried in securities filings found nine major tech companies are carrying roughly $3 trillion in off-balance-sheet commitments mostly tied to AI. Spending that doesn't show up in the capital expenditure figures companies report each quarter and which is growing faster than that traditional capex, which totaled about $600 billion over the past year across the same companies.

Alphabet's own purchase commitments and contractual obligations illustrate the scale. Standing at $811 billion as of June 30, more than double the $332 billion reported just three months earlier, with the company citing technical infrastructure, inventory, and agreements to secure energy for data centre usage, some of which run as far out as 2054, without detailing exactly why the number jumped so dramatically in a single quarter.

These off-balance-sheet exposures include things like agreements to buy other companies' stock in the future, backstopping leases for other tenants, and equity commitments, Nvidia alone committed to $27 billion in future equity investments between April 2026 and the end of its fiscal year in January 2027. The obligations are roughly triple what these companies owe under their existing leases and long-term borrowings combined.

The genuinely concerning part isn't the accounting technicality itself, it's that purchase commitments and signed leases mostly can't be cancelled, meaning this spending exists whether or not the anticipated AI revenue actually materialises on the timeline these companies are betting on. Alphabet and Amazon have both recently posted negative free cash flow, meaning their capital spending already exceeds what their operations bring in, before even factoring in these trillions of additional off-balance-sheet obligations.

So the honest framing is this isn't necessarily reckless. Cheap long duration debt makes sense if the AI buildout pays off the way these companies are betting, but it does mean the actual financial exposure to an AI slowdown is meaningfully larger and less visible in standard reporting than most investors and casual observers currently understand it to be

Why Big Techs Ai Spending Is 3 Trillion Higher Than It Seems E1067Bb2www.wsj.com

[Mark94]

Feels like the footnote disclosure point deserves more scrutiny from regulators if $3 trillion in real financial commitment is only visible to someone actually reading securities filing footnotes carefully. That's a genuine transparency gap for regular investors trying to assess actual risk

[GradientPanther]

Negative free cash flow at Alphabet and Amazon before even counting the off-balance-sheet trillions is a quite sobering data point. That's already spending beyond what the core business generates, and the actual total exposure is meaningfully larger than that headline number suggests

[Dave_37]

Interested to see how this actually gets stress tested if AI revenue growth really does slow down meaningfully.

Feels like the practical mechanics of what happens when a company is locked into decades of non cancellable spending against disappointing returns deserves its own dedicated follow up

[Tel75]

Would like to know how this compares proportionally to previous major infrastructure buildouts. Telecoms during the dot com era or utilities building out the original power grid, feels like there might be a useful historical parallel for how this kind of long duration commitment usually plays out. Worth remembering

[Wandering Matt]

From where I sit, this isn't inherently reckless the way some coverage frames it. Cheap long term financing for infrastructure that could pay off for decades is a completely reasonable strategy if the underlying demand actually shows up on schedule