The AI buildout is far larger than the capex line admits — about $3 trillion larger, and most of it isn't recorded as debt.

Investors have spent two years tracking the AI boom through one number: capital expenditure. Every quarter, the hyperscalers report how many tens of billions they poured into chips, servers, and data centers, and analysts extrapolate the trajectory. It turns out that number captures only a fraction of the commitment. A Wall Street Journal analysis found that nine major technology companies carry roughly $3 trillion in obligations tied largely to AI infrastructure that do not appear as conventional debt on their balance sheets — against about $600 billion in reported capital spending over the prior year.

That gap is the story. Big Tech AI spending, measured honestly, is roughly five times what the headline capex figure suggests, and the difference is structured to stay out of the debt column. It is not fraud; it is accounting working as designed. But it changes what investors are actually looking at when they look at the AI trade.

Where Big Tech's AI spending actually hides

The commitments break into two buckets, according to the Journal's review. About $1.2 trillion sits in leases for facilities that have not yet begun operating — data center shells and campuses contracted but not switched on. Another $1.9 trillion is in commitments to purchase chips, equipment, energy, and other services. Neither shows up as debt in the usual sense, because under accounting rules a lease obligation for a facility that isn't operational yet, or a multi-year purchase agreement, is disclosed in the footnotes rather than booked as a liability on the face of the balance sheet.

The individual figures are what make the scale legible. Alphabet disclosed $811 billion of purchase and contractual obligations, up from $332 billion just three months earlier — a more than $470 billion jump in a single quarter, as widely reported from the Journal's analysis. Meta reported $347 billion of future lease commitments. The companies in the review span the AI supply chain: Alphabet, Amazon, Microsoft, Oracle, Nvidia, Broadcom, SpaceX, AMD, and Meta.

None of this is hidden in the sense of being secret — it is all disclosed somewhere in the filings. It is "hidden" only in the sense that it does not appear where investors habitually look. The obligations live in contractual-commitment tables and lease-maturity schedules, not in the debt line that drives most valuation models. An investor scanning balance-sheet leverage would see companies that look conservatively financed. An investor reading the footnotes would see a very different, far more committed posture.

Why the structure exists — and why it matters now

There are ordinary reasons a company would prefer a long-term lease or a purchase commitment over issuing debt to build a facility itself. Leasing lets you secure scarce capacity — data center shells, power, GPUs — without tying up capital or expanding reported borrowings. In a market where compute is the binding constraint, locking in future supply is rational, even necessary. The hyperscalers are using every available instrument — leases, special-purpose financing, and long-dated purchase contracts — to grab capacity before rivals do.

The reason this matters more in 2026 than it would have in a calmer market is utilization. Every one of these trillions in commitments is a bet that demand for AI — training, inference, agents, enterprise applications — will keep rising fast enough to justify the capacity being locked in. If it does, the obligations convert into productive, revenue-generating infrastructure and the off-balance-sheet structure looks like prudent forward-buying. If demand disappoints, those same commitments become fixed obligations to pay for capacity that isn't earning its keep — and fixed obligations are exactly what turns a slowdown into a squeeze.

This is the quiet through-line connecting the week's other AI news. When Nvidia guarantees $105 billion of a data center's lease payments, or when a single company's contractual obligations jump $470 billion in a quarter, the same dynamic is at work: the industry is converting confidence about future demand into binding present commitments, using structures that keep the leverage out of plain sight. The confidence may well be justified. The point is that the leverage is real whether or not it sits in the debt column.

Stakeholder analysis: who carries this risk

For equity investors, the finding is a call to read past the headline balance sheet. A company can look under-leveraged on conventional metrics while carrying enormous fixed commitments in its footnotes. Valuation models that ignore those obligations understate both the scale of the AI bet and the downside if utilization slips. The prudent response is not panic but literacy: the contractual-commitment tables are now as important as the capex line.

For the companies themselves, the structure is a double-edged tool. It lets them move fast and secure scarce capacity without bloating reported debt or spooking investors with a leverage spike. But it concentrates operational risk on one variable — demand — and removes the flexibility that owning less and committing less would preserve. A company that has pre-committed trillions has, in effect, made a large, largely irreversible bet on its own demand forecast.

For lenders and credit markets, the off-balance-sheet layer complicates risk assessment across the sector. When leases, guarantees, and purchase commitments interlock across chipmakers, cloud providers, and AI labs, the true distribution of exposure gets harder to trace. That opacity is precisely what makes analysts nervous about systemic fragility: it is difficult to price a risk you cannot fully see.

For the broader economy, the numbers reframe how big the AI cycle really is. If nine companies carry $3 trillion in AI-linked commitments, the buildout is a macro event — one whose fate is tethered to whether AI demand keeps compounding. That is a lot of the economy's forward momentum riding on a single, still-unproven assumption about adoption.

The takeaway for operators and investors

Two durable lessons come out of the Journal's tally.

First, in any capital-intensive boom, the reported spending number is the floor, not the ceiling, of the real commitment. Leases, purchase agreements, and financing structures can multiply a company's true exposure several times over while leaving the balance sheet looking calm. Anyone trying to gauge the size of the AI bet — as an investor, a competitor, or a supplier — has to read the footnotes, because that is where the actual scale lives.

Second, off-balance-sheet does not mean off-the-hook. A contractual obligation to pay for capacity is an obligation whether or not accounting rules require it in the debt line. The structure changes where the number appears; it does not change who owes it. The $3 trillion is real money that real companies have promised to pay, and the entire edifice rests on demand rising to meet it.

The AI infrastructure race has been narrated as a spending story — who is writing the biggest checks. The Journal's analysis reframes it as a commitment story, and commitments are stickier and more consequential than spending. Capex, at least, reflects money already deployed against assets you own. A trillion dollars in future lease obligations is a promise about a future that has to arrive. The bill for the AI boom is already written. What's still unsettled is whether the demand shows up to pay it.

Frequently Asked Questions

What did the Wall Street Journal find about Big Tech AI spending?

The Journal's analysis found that nine major technology companies carry roughly $3 trillion in commitments tied largely to AI infrastructure that do not appear as conventional debt on their balance sheets — compared with about $600 billion in reported capital expenditure over the prior year.

Why isn't the $3 trillion counted as debt?

The obligations are mostly leases for facilities not yet operating (about $1.2 trillion) and commitments to buy chips, equipment, energy, and services (about $1.9 trillion). Under accounting rules, these are disclosed in filing footnotes rather than recorded as liabilities on the face of the balance sheet, so they don't appear where investors typically look.

Which companies were included in the analysis?

The review covered Alphabet, Amazon, Microsoft, Oracle, Nvidia, Broadcom, SpaceX, AMD, and Meta. Alphabet alone disclosed $811 billion of purchase and contractual obligations, up from $332 billion three months earlier, and Meta reported $347 billion of future lease commitments.

Why does off-balance-sheet AI spending matter?

Because it means the AI buildout is far larger and more leveraged than headline capex suggests, and the entire structure depends on AI demand continuing to rise. If utilization disappoints, these commitments become fixed obligations to pay for capacity that isn't generating enough revenue — a risk that isn't visible in standard balance-sheet metrics.

Editor's note — sources: The Wall Street Journal analysis, as reported via MSN, Yahoo Finance, and Seeking Alpha.

Share this article: