Five US Tech Giants Now Carry 3.2 Trillion Dollars of AI Shadow Debt That Never Hits the Balance Sheet

The five largest American buyers of AI infrastructure now owe about 3.2 trillion dollars in commitments that never appear on their balance sheets, according to a new tally by Nikkei, and that figure has grown by roughly 90 percent in just six months.

The calculation covers Alphabet, Meta, Microsoft, Amazon and Oracle and is drawn from their latest financial statements, with some estimates where disclosure is incomplete. About 60 percent of the total consists of procurement contracts for semiconductors, servers and power that have not yet been fulfilled; the remaining 40 percent is leases on data centers that have not yet commenced. Under US accounting rules, unfulfilled purchase orders and leases that have not started generally stay off the liability side of the ledger and surface only in footnotes.

Nikkei notes that the combined shadow obligations now equal roughly 80 percent of the five companies' total assets. If the current pace of AI-related contracting continues, the figure could soon exceed the value of everything the five companies own on paper.

Google parent Alphabet carries the largest burden: about 1.0549 trillion dollars as of the end of September. The company appears to have partially adjusted its calculation methodology, so the figure cannot be compared directly with earlier periods, but Nikkei estimates it is around 2.6 times Alphabet's level at the end of March. Meta, Microsoft, Amazon and Oracle follow, in that order.

The new numbers extend a picture that has been deteriorating all autumn. Nikkei's July tally put off-balance-sheet AI obligations at the five companies near 1.65 trillion dollars, roughly eight times the level of four years earlier and already above their combined reported debt of about 1.35 trillion dollars. A Wall Street Journal analysis in October widened the aperture to nine companies including Nvidia and Broadcom and put the total near 3 trillion dollars.

None of this is illegal, or even unusual, in isolation. Long-dated equipment orders and pre-commencement leases are standard tools for smoothing capital expenditure. What has changed is the scale: cloud companies are signing multi-year GPU supply deals and gigawatt-scale lease commitments at a pace that locks in cash outflows for years before any corresponding revenue arrives. Rating agencies have started flagging the gap, with S&P downgrading Meta's outlook to negative in late October and spreads on new bond issues from the same five companies widening measurably.

The risk Nikkei points to is asymmetry. If AI demand materializes as forecast, the footnotes describe an infrastructure moat. If it does not, investors will discover that the true leverage of the AI buildout was never visible in the headline debt ratios they were watching. The five companies' reported debt loads looked manageable precisely because the most expensive promises were kept off the books.

For now, the market is still funding the buildout. November's bond pipeline from Microsoft, Google, Oracle and Meta alone could top 60 billion dollars, a record month for the sector. The question the footnotes raise is a simple one: at what point do the shadow numbers get big enough that ignoring them is no longer an option?