$3.2T hidden AI debt grew 90% in 6 months

Five companies' off-balance-sheet AI obligations nearly doubled in six months, from $1.65 trillion to $3.2 trillion. But two tallies published the same day disagree on Meta by 2x — so the headline total is the least interesting number in the story.

The $3.2 trillion figure comes from Nikkei Asia, relayed November 3 by AI Weekly: Alphabet, Microsoft, Amazon, Meta, and Oracle together carry about $3.2 trillion in off-balance-sheet AI obligations, up from about $1.65 trillion in Nikkei's earlier cut of the same five names. My arithmetic on Nikkei's own numbers: ($3.2T − $1.65T) / $1.65T = 93.9%, which matches both "almost doubled" and the "swells 90%" headline. These are not loans in the conventional sense.

They are future lease obligations and multi-year purchase contracts for chips, servers, and data-center capacity, disclosed in SEC footnotes rather than on the balance sheet an ordinary investor reads first. The structures are legal and technically disclosed — the complaint is that they move the arithmetic off the page people actually look at.

The July baseline adds a comparison Nikkei's new number doesn't repeat. In July, the same five companies' hidden $1.65 trillion already exceeded their $1.35 trillion of reported debt — hidden obligations were 55.0% of the $3.0 trillion combined total. The November update doesn't restate reported debt, so I can't extend that ratio forward from the relays alone. What I can say is that the direction on Nikkei's consistent series is unambiguous: roughly +94% in about six months.

Then a second tally landed the same day and complicated the picture. Kalkine, publishing November 3, puts the aggregate at $3.3 trillion with a per-company breakdown: Alphabet $1.06 trillion, Meta $866 billion, Microsoft $593 billion, Amazon $453 billion, Oracle $322 billion. That breakdown sums cleanly — $1.06T + $0.866T + $0.593T + $0.453T + $0.322T = $3.294 trillion, which rounds to the advertised $3.3 trillion — and gives shares of 32.2% Alphabet, 26.3% Meta, 18.0% Microsoft, 13.8% Amazon, and 9.8% Oracle.

The totals differ by only $100 billion, 3.1% of Nikkei's figure — rounding and methodology noise, plausibly. The per-company figures are where the tallies diverge. Nikkei's relay puts Meta at roughly $420 billion; Kalkine puts Meta at $866 billion.

That's a 2.06x gap, +106.2%. Oracle shows a smaller version of the same problem: Nikkei's July relay had Oracle at $273.3 billion as of May, and Kalkine now has $322 billion, +17.8% — but July-to-November growth and a definitional difference are indistinguishable from the relays alone. The likeliest explanation is that the two tallies count different things — Nikkei's is scoped to AI-related obligations, Kalkine's reads as all off-balance-sheet commitments — but neither relay publishes a methodology precise enough to prove that, and Nikkei's original is paywalled, so I read it only through relays.

Treat any per-company figure here as definition-dependent.

One more check failed, and it's the kind of failure worth naming. AI Weekly's relay says Meta's roughly $420 billion is "nearly triple its reported long-term debt of about $58.7 billion." Divide it out: $420B / $58.7B = 7.2x, not ~3x. Either the comparison base for "nearly triple" is a different debt line than the $58.7 billion quoted in the same paragraph, or the gloss is wrong. I can't resolve that from the relays, and I won't repeat "nearly triple" as if it divided out.

The structural details are more durable than the totals. Nikkei's relay singles out Meta's Hyperion data-center project: Blue Owl Capital owns 80% and Meta 20%, and the deal expanded from $27 billion and 2 gigawatts to $50 billion and 5 gigawatts in June — the archetype of capacity financed off the balance sheet. Kalkine adds that nearly half of Meta's $866 billion is data-center leases not yet commenced, with payment periods extending up to 30 years.

Arithmetical coincidence, flagged as such and nothing more: half of $866 billion is about $433 billion, close to Nikkei's $420 billion Meta figure. I have no evidence the two numbers measure the same thing, and the definitional gap above says they probably don't — but it's the kind of near-match that should make you ask what each tally actually counts before quoting either.

Oracle is the leverage outlier on either tally. Kalkine's $322 billion is 4.78x Oracle's fiscal 2026 revenue of $67.36 billion — the "roughly 4.8 times" in the article checks out — and that sits alongside $125.34 billion in notes payable and other borrowings as of August 31, 2026. Alphabet's "nearly 700% surge in one year" implies a year-ago base of roughly $133 billion if a 700% increase means 8x the original, though "surge of 700%" is ambiguous between 7x and 8x and the relay doesn't pin it down. And the financing context is already tightening at the edges: CoreWeave's debt yields have climbed above 15% with credit-default swap spreads at record levels, per Kalkine. The five giants are not CoreWeave — their cash flows are far larger — but the direction of travel is the same, only slower.

The honest summary is this: $3.2 trillion is an order of magnitude, not a measurement. What survives definitional scrutiny is the run rate on Nikkei's consistent series (+93.9% in six months), the structure (multi-decade leases for capacity that in some cases doesn't exist yet), and the asymmetry (commitments must be paid whether or not the AI revenue arrives to cover them). Quote the total with its definition attached, or don't quote it at all.

Sources: AI Weekly relay of Nikkei Asia, "Nikkei: Big Tech Off-Balance-Sheet AI Debt Doubles to $3.2T" (November 3, 2026) — https://aiweekly.co/alerts/nikkei-big-tech-off-balance-sheet-ai-debt-doubles-to-32t; Kalkine, "Big Tech's AI Tab Hits $3.3 Trillion in Off-Balance-Sheet Commitments" (November 3, 2026) — https://kalkine.ca/news/general-news/big-techs-ai-tab-hits-33-trillion-in-off-balance-sheet-commitments; July Nikkei relay via TechSpot (July 22, 2026) — https://www.techspot.com/news/113206-amazon-meta-microsoft-other-us-tech-giants-hiding.html. All growth rates, shares, and sums are my own calculations from the relayed figures; Nikkei's original was not read directly (paywalled). This is not investment advice.