Big Tech Is Hiding $1.65 Trillion in AI Debt

TL;DR

- Alphabet, Microsoft, Amazon, Meta, and Oracle carry roughly $1.65 trillion in off-balance-sheet AI debt from data center leases and GPU contracts, according to a Nikkei Asia investigation. - That hidden total exceeds their $1.35 trillion in officially reported debt, meaning most real obligations sit in financial footnotes. - Meta's off-balance-sheet commitments hit $420 billion, nearly triple its recorded debt. Oracle's jumped roughly 2,900% since 2022. - Tech critic Ed Zitron calls it "a corporate scandal." If you build on any of these companies' cloud or AI platforms, their debt burden is your pricing risk.

Alphabet, Microsoft, Amazon, Meta. And Oracle collectively carry about $1.65 trillion in off-balance-sheet obligations tied to AI infrastructure, according to a Nikkei Asia investigation. That figure exceeds the roughly $1.35 trillion in debt these same five companies officially report on their balance sheets. Combined, their total obligations sit around $3 trillion. And more than half does not appear as debt in any headline financial metric.

If you are building on their cloud platforms or paying for their API tokens, this matters because that hidden debt is the subsidy keeping your prices low.

How Five Companies Stacked $1.65 Trillion in Hidden AI Obligations

The off-balance-sheet AI debt comes from three sources: long-term leases for data centers not yet operating, forward purchase orders for GPUs and servers not yet delivered. And structured joint ventures for AI campuses.

Under U.S. GAAP, these commitments stay in footnote disclosures rather than appearing as current liabilities. They only migrate onto the balance sheet once data centers commence operations or hardware is delivered.

This is legal under current accounting rules.

The obligations are real and legally binding. They simply do not show up in the debt column.

The scale is what makes this different. Nikkei found that off-balance-sheet obligations across these five companies grew roughly eightfold in about four years, tracking the AI infrastructure spending surge. The $1.65 trillion figure appears in quarterly financial statement annotations, not on balance sheets themselves. You have to read the notes to find it.

Here's the thing for small operators: the cheap API pricing you pay today is subsidized by a $1.65 trillion bet. If that bet does not generate returns, pricing changes. It changes fast, and it changes in one direction. Your $5 per million tokens on a frontier model exists given that someone is borrowing billions to build the hardware underneath it.

When the bill comes due, you will feel it in the unit economics of every product you ship on top of these platforms.

Why Meta and Oracle Are the Sharpest Cases

Nikkei highlighted Meta as the most extreme example among the five. Meta's off-balance-sheet commitments total approximately $420 billion, nearly three times its roughly $140 billion of recorded debt. Investors reading Meta's balance sheet see less than one-third of the company's actual financial obligations. That gap is not a rounding error.

It is a structural feature of how Meta is financing its AI infrastructure buildout.

Oracle's trajectory is even steeper in percentage terms.

Its off-balance-sheet commitments expanded about 30-fold since 2022, reaching approximately $273.3 billion. That is a roughly 2,900% jump in four years. Oracle went from a relatively modest infrastructure player to carrying hidden obligations that dwarf its historical balance sheet.

But wait. The problem is not just the size of individual numbers. It is the velocity across all five companies. Off-balance-sheet AI obligations grew eightfold in four years. The spending is accelerating, not stabilizing. Every quarter brings new data center announcements, new GPU orders, new joint ventures.

Each one adds to the footnote total without touching the headline debt figure.

I run a small AI consulting agency.

My clients pay for API calls, cloud compute, and managed models. Every pricing decision I make assumes the underlying infrastructure will cost roughly what it costs today, maybe less over time as efficiency improves. What happens to that assumption when $1.65 trillion in obligations starts rolling onto balance sheets as data centers go live? The answer is not "nothing." The answer is margin compression, and margins compress downward, not upward.

The Accounting Trick That Keeps Trillions Invisible

Under current U.S. accounting rules, gear committed under long-term contracts but undelivered counts as off-balance-sheet. Data centers financed through project arrangements or extended leases receive the same treatment. The commitments exist as legally binding contracts. They just sit in notes and supplemental disclosures rather than headline debt totals that investors and analysts scan first.

Nikkei's analysis found that off-balance-sheet AI-related commitments run about 122% of these companies' recorded debt.

For every dollar of debt an investor can see on a balance sheet, another $1.22 hides in the footnotes. Total combined obligations of approximately $3 trillion mean anyone looking at reported figures alone is seeing less than half the picture.

Tech critic Ed Zitron called the situation "a corporate scandal." You can decide whether that language is proportional. But the structural reality is hard to dismiss: five companies committed $1.65 trillion to AI infrastructure using accounting treatment that keeps most of it invisible in standard financial reporting. Earnings season is here. The footnotes will be in the quarterly filings. The headlines will not mention the $1.65 trillion.

What Small Operators Should Do Right Now

If your business runs on any hyperscaler's infrastructure, you need to price in the possibility that API and compute costs rise materially within the next 12 to 24 months.

Not as the technology gets worse. Since the financial pressure of servicing $3 trillion in total obligations eventually forces pricing changes somewhere in the stack.

Here are concrete steps. First, audit which of your workloads are locked into a single provider. If you are fully on AWS or fully on Azure, a pricing shock hits everything at once. Second, document your current unit economics precisely. Know what you pay per API call, per GPU hour, per gigabyte transferred. And build a sensitivity model for what happens if those rates increase. Third, identify which workloads could move to smaller providers or self-hosted infrastructure if costs jump 20 to 40 percent. You will not move everything. But knowing what is portable gives you room to negotiate when the invoice changes.

The off-balance-sheet AI debt story is not going away. As data centers come online and GPU orders get delivered, those obligations move from footnotes to balance sheets. The $1.65 trillion question is whether AI revenue grows fast enough to cover it.

If it does not, every developer, agency. And small business building on these platforms absorbs the cost through higher API bills, changed terms, or provider consolidation.

Check your provider dependencies this week. Not next quarter. The Nikkei Asia investigation and Tom's Hardware breakdown lay out the full numbers so you can make informed decisions before the pricing picture shifts.