Off-Balance-Sheet Financing: The Accounting Gap Behind the AI Data Center Buildout

An accounting technique that keeps certain lease obligations out of a company's reported liabilities — and, per Moody's, now hides an estimated $662 billion to $1.2 trillion of the hyperscalers' AI data center commitments from their balance sheets.

Created 2026-08-21 Last reviewed 2026-08-21

What it is

Off-balance-sheet financing is a set of accounting arrangements that let a company acquire the use of an asset — a building, equipment, a data center — without recording the corresponding debt as a liability on its balance sheet. The classic vehicle is the operating lease: rather than buying a facility outright (which shows up as an asset offset by a loan), a company leases it and expenses the rent payments as they come due. For decades this let firms report lower leverage than their real financial commitments implied, since long-term lease obligations sat in the footnotes of financial statements rather than on the balance sheet itself.

Regulators moved to close much of this gap. In February 2016 the Financial Accounting Standards Board issued Accounting Standards Update 2016-02, “Leases (Topic 842),” which the FASB and the SEC described as ending one of the largest forms of off-balance-sheet accounting: it required companies to recognize most lease assets and lease liabilities directly on the balance sheet. That standard closed the old loophole for leases already in effect.

It did not close every loophole. Two mechanisms still let obligations stay off the books: leases that have been signed but not yet “commenced” (the clock for balance-sheet recognition starts when the lessee takes control of the asset, not when the contract is signed), and shorter-duration leases — increasingly common in data center deals — that are backstopped by guarantees to cover payments if the lease isn’t renewed, guarantees that current disclosure rules don’t clearly require companies to report as liabilities.

Why it matters for AI governance and narratives

The AI buildout has turned this accounting mechanism into a live financial-stability question. Amazon, Microsoft, Google, Meta and Oracle are financing an unprecedented share of their data center capacity through leases with the specific structural features — uncommenced status, short terms with backstop guarantees — that current GAAP treats as exempt from balance-sheet recognition. Moody’s put the scale of these hidden commitments at $662 billion as of year-end 2025, equal to 113% of the same five companies’ combined on-balance-sheet debt, and warned the true figure could be materially higher once broader lease and purchase-commitment categories are counted. A Nikkei study cited by Fortune in July 2026 estimated total hidden debt at the five firms at $1.65 trillion, an eightfold increase over four years.

For an observatory tracking how AI is framed and understood, this is a case where the contest over meaning runs through accounting technicalities rather than public rhetoric. Builder-ecosystem narratives about AI capex tend to describe spending in terms of committed capacity and compute — a story of scale and inevitability. Off-balance-sheet reporting is part of what makes that framing possible: it lets the infrastructure buildout look smaller, and the underlying leverage look lower, than the cash commitments actually are. Capital-market skeptics invoking terms like “data center bust” are making the inverse argument — that risk is being systematically underpriced because a large share of it is structurally invisible to standard balance-sheet analysis. The dispute is not just about whether AI infrastructure spending is prudent; it is about whether the accounting categories used to describe that spending let investors and the public see the commitment at all.

Key facts and dates

FASB’s Topic 842 leases standard, issued February 25, 2016, required companies to bring most lease liabilities onto the balance sheet, closing what regulators called one of the largest historical forms of off-balance-sheet accounting. Moody’s February 2026 report on hyperscaler data center leases found $662 billion in commitments still excluded from the big five tech firms’ balance sheets under the uncommenced-lease and backstop-guarantee exceptions that Topic 842 left open, and quoted Moody’s VP David Gonzales warning that “under the current rules it is possible to defer the reporting of this liability until the last minute, so the fear is that this indicator of health will be lagging.” By July 2026, estimates of the scale had grown further: Moody’s put off-balance-sheet deals at roughly $1.2 trillion, with more than $820 billion tied to data centers still under construction, while a Nikkei analysis cited by Fortune put total hidden debt at the five firms at $1.65 trillion — exceeding their combined official balance-sheet debt of $1.35 trillion. Moody’s has also noted that despite these exposures, the hyperscalers’ investment-grade credit ratings are not facing immediate pressure, since the companies retain strong operating balance sheets overall.

Where to learn more

Sources

Primary regulatory source: the official standard that closed most off-balance-sheet lease accounting, and the basis for understanding which exceptions remain
Reports the February 2026 Moody's analysis directly, including the $662B figure, the accounting mechanism, and an on-record Moody's analyst quote
More recent (July 2026) figures from Moody's and a Nikkei study showing the scale of hidden hyperscaler debt continuing to grow
Accessible secondary explainer of the underlying accounting concept and how ASC 842 changed lease reporting
Referenced in: Editorial No. 272