Big Tech's off-balance-sheet AI obligations nearly double to $3.2 trillion in six months

Off-balance-sheet obligations tied to AI infrastructure at Alphabet, Microsoft, Amazon, Meta and Oracle nearly doubled in roughly six months, climbing from about $1.65 trillion to $3.2 trillion, according to Nikkei Asia figures relayed by AI Weekly on November 3. A second tally published the same day by Kalkine puts the aggregate slightly higher at $3.3 trillion. The two counts agree at the top line but diverge sharply company by company, which makes the headline number the least settled figure in the story.

What the obligations actually are

These are not conventional loans. According to the Nikkei relay, they are future lease payments and multi-year purchase contracts for chips, servers and data-center capacity, disclosed in SEC footnotes rather than on the primary balance sheet. The structures are legal and technically disclosed; the criticism is that the commitments sit outside the statements investors read first. A dev.to analysis that cross-checked both tallies calculates growth on Nikkei's consistent five-company series at about 94% over six months.

Two tallies published the same day disagree

Kalkine's per-company breakdown puts Alphabet at $1.06 trillion, Meta at $866 billion, Microsoft at $593 billion, Amazon at $453 billion and Oracle at $322 billion — amounts that sum to roughly $3.29 trillion.

The aggregates differ by about $100 billion, or roughly 3%, which could be rounding or methodology noise. The company-level numbers diverge far more: the Nikkei relay places Meta at roughly $420 billion, while Kalkine has $866 billion, a gap of more than two times. The most plausible explanation, according to the dev.to analysis, is that Nikkei scopes its count to AI-related obligations while Kalkine counts all off-balance-sheet commitments — but neither tally publishes a methodology precise enough to confirm that, and Nikkei's original article sits behind a paywall. Every per-company figure here depends on what is being counted. The dev.to analysis also flags a fact-check that failed: the AI Weekly relay describes Meta's roughly $420 billion as almost three times its reported long-term debt of about $58.7 billion, but the division works out to about 7.2 times.

The July baseline adds context

A Nikkei tally relayed by TechSpot in July supplies a comparison the November update omits: at that point, the five companies' $1.65 trillion in off-balance-sheet obligations already exceeded their $1.35 trillion of reported debt, making the hidden portion 55% of a $3.0 trillion combined total. The November figures do not restate reported debt, so that ratio cannot be carried forward.

The structures behind the numbers

The Nikkei relay points to Meta's Hyperion data-center project as a template for this kind of financing: Blue Owl Capital owns 80% and Meta 20%, and the deal expanded in June from $27 billion and 2 gigawatts of capacity to $50 billion and 5 gigawatts. Kalkine adds that nearly half of Meta's $866 billion consists of data-center leases that have not yet commenced, with payment periods running as long as 30 years.

Oracle carries the heaviest relative load

Oracle stands out in both tallies when measured against its size. Its $322 billion in Kalkine's count is about 4.8 times Oracle's fiscal 2026 revenue of $67.36 billion, and it comes on top of $125.34 billion in notes payable and other borrowings as of August 31, 2026. Kalkine also reports that financing conditions are tightening at the edges of the AI buildout: CoreWeave's debt yields have climbed above 15%, with credit-default swap spreads at record levels. The five giants' cash flows dwarf CoreWeave's, but the same pressure is building, just more slowly.

Why it matters

The durable story is not the total, which is closer to an order of magnitude than a precise measurement. What holds up is the pace — roughly 94% growth in six months on a consistent series — the structure, meaning multi-decade leases for capacity that has not yet been built, and the asymmetry: the payments fall due regardless of whether the AI revenue expected to cover them materialises. Anyone citing the headline figure should attach the definition behind it; without that, the number says very little.