A Wall Street Journal analysis of company filings, mostly current through June, puts off-balance-sheet commitments across nine companies at roughly $3 trillion: about $1.9 trillion in purchase commitments and $1.2 trillion in leases that have been signed but have not yet commenced. The nine are Microsoft, Amazon, Alphabet, Meta, Oracle, Nvidia, Broadcom, AMD and SpaceX.

For scale, a Nikkei estimate in July put the figure at $1.65 trillion across five companies. The definitions differ, but the direction does not.

Start with the word "hiding"

Some of the coverage uses it. It is the wrong word, and getting this right matters more than the headline.

These obligations appear in the notes to audited financial statements, which is where accounting standards require them. Alphabet disclosed $811 billion of purchase commitments and contractual obligations as of June 30, up from $322 billion three months earlier, a 152% increase in one quarter. Nvidia quantified $30 billion of cloud-service commitments. Oracle disclosed $10 billion. Meta reported $238 billion of commitments while declining to break out the cloud-capacity portion.

Nobody is concealing anything. The companies are telling you, in the place the rules say to tell you, and almost nobody was reading it until the numbers got large enough to notice.

What off-balance-sheet actually means

It is a technical description, not an accusation.

A balance sheet records assets you control and liabilities you owe now. A commitment to buy something in future, or a lease that has been signed but whose term has not begun, is neither: you do not yet control the asset and the obligation has not yet crystallised. Accounting standards therefore keep it out of the headline totals and require it in the notes, where the amount and timing are set out.

For most companies in most years this is a minor disclosure. It becomes material when the commitments are large relative to the balance sheet, long-dated, and concentrated in a single bet, which is the situation here.

Why it changes how the numbers read

Two ways, both worth holding.

The first is leverage. Analysts compare debt to earnings to judge whether a company can service what it owes. If a substantial part of what a company is obliged to pay sits outside the debt figure, conventional leverage ratios understate the commitment. Moody's made this point in July, putting direct debt across six major technology firms at $460 billion against $1.2 trillion of off-balance-sheet lease and purchase commitments. The rating agencies are already adjusting for it; the headline metrics most investors see are not.

The second is flexibility, and it is the more important one. Capital expenditure can be slowed if demand disappoints: a company delays the next tranche and preserves cash. A signed purchase commitment cannot, or not without a negotiation and a penalty. Converting discretionary spending into contractual obligation is precisely what removes the option to stop, and it is what the last year of this build-out has done at scale.

The rate of change is the real signal

Alphabet's quarterly jump from $322 billion to $811 billion is the number to sit with. Whatever one believes about the eventual returns on AI infrastructure, a 152% increase in contractual obligations in three months is a company committing very hard, very fast, to a capacity plan.

Boursel has reported the same pressure from several directions this week: technology now trades about nine basis points wider than the investment-grade bond index, hyperscaler issuance has reached at least $220 billion this year against $12.5 billion in the same period of 2025, and Nvidia has begun investing directly in the electricity supply its customers need. The $3 trillion figure is the same phenomenon measured in the notes rather than in the bond market.

What to watch, and where

The disclosures themselves, in the commitments and contingencies note of each 10-Q. Three things in particular: whether the totals keep compounding at this rate, whether the maturity profile shortens, and whether any company begins to disclose cancellation terms, which would tell you what the obligations are actually worth if demand does not arrive on schedule.

That last one is currently the largest gap in what is known. A $3 trillion commitment with generous exit provisions is a different thing from one without, and almost none of the filings say which it is.