They have not yet received the services, but they will.
Five of the world's most powerful technology companies have quietly committed to nearly a trillion dollars in future data center obligations, yet two-thirds of that sum remains invisible to investors under accounting rules drafted before the digital age existed. Moody's has surfaced this hidden architecture of debt — $662 billion in unrecorded liabilities — not as evidence of wrongdoing, but as a warning that the financial statements guiding markets may be telling only part of the story. As artificial intelligence reshapes the scale and urgency of infrastructure spending, the gap between what is owed and what is disclosed has become a question the entire financial system may soon be forced to answer.
- Meta, Amazon, Alphabet, Microsoft, and Oracle have quietly accumulated $662 billion in data center lease obligations that legally vanish from their balance sheets — an amount exceeding the combined adjusted debt of all five companies.
- The mechanism is a loophole rooted in century-old accounting rules: leases not yet commenced need not be recorded, and short AI-era lease terms allow companies to argue they are not 'reasonably certain' to renew, keeping renewal costs off the books entirely.
- Residual value guarantees — promises to cover a landlord's losses if a facility loses value — add another layer of concealment, with Meta alone shielding $28 billion in obligations by deeming a payout 'not probable.'
- Moody's analysts warn that when these leases do commence, the cash outflows will be enormous and inescapable, leaving companies dangerously exposed if AI demand falters or technology shifts faster than expected.
- The rating agency is now weighing 'nonstandard adjustments' that would treat hidden obligations as present debt — a move that could trigger credit downgrades, raise borrowing costs, and force a reckoning across the industry.
Five of the world's largest technology companies — Meta, Amazon, Alphabet, Microsoft, and Oracle — have committed to $969 billion in future data center lease payments, yet $662 billion of that total sits legally invisible on their financial statements. A new Moody's analysis reveals how this is possible: under GAAP accounting, companies only record lease obligations once a lease has actually begun. Since most of these AI infrastructure deals won't commence until sometime between 2025 and 2031, the liabilities can be lawfully excluded for now. Alphabet alone disclosed $42.6 billion in unrecorded future payments in a single quarter — nearly double what it had reported just three months earlier.
The structure of modern AI deals makes the concealment especially effective. Because specialized AI chips become obsolete in four to six years, tech companies negotiate shorter initial lease terms with renewal options, allowing them to argue they aren't 'reasonably certain' to renew — the threshold GAAP requires before renewal costs must be recorded. To satisfy landlords unwilling to build billion-dollar facilities without guarantees, the companies offer residual value guarantees instead. Meta, for instance, attached a $28 billion guarantee to leases beginning in 2029, then kept the entire sum off its books by judging a payout 'not probable.'
Moody's analysts are careful to note that no rules have been broken — the liabilities are real, merely deferred. The deeper concern is what happens when they arrive. If AI demand softens or technology pivots unexpectedly, these companies will still face locked-in, multibillion-dollar commitments with little room to maneuver. To underscore the scale, total data center capital spending now runs at roughly $646 billion annually — about 2% of U.S. GDP, comparable to the combined output of Singapore, Sweden, and Argentina.
Moody's has signaled it may begin treating these hidden obligations as present debt when calculating creditworthiness, a move that could trigger downgrades and raise borrowing costs across the sector. The broader question is whether other rating agencies and regulators will follow — and whether accounting standards built for a different era can keep pace with the infrastructure ambitions of the AI age.
Five of the world's largest technology companies have committed to spending hundreds of billions of dollars on data centers that don't yet appear on their financial statements. According to a new analysis from Moody's, Meta, Amazon, Alphabet, Microsoft, and Oracle collectively owe $969 billion in future lease payments for AI data centers as of the end of 2025. But here's the accounting sleight of hand: $662 billion of that total—nearly two-thirds—sits in a legal gray zone where these companies don't have to report it to investors or regulators. That hidden obligation alone equals 113% of these five companies' combined adjusted debt.
The reason these liabilities vanish from balance sheets comes down to a technicality in accounting rules written nearly a century ago. Under generally accepted accounting principles, or GAAP, companies only have to record lease obligations once the lease has actually begun. Since most of these data center leases won't start until sometime between 2025 and 2031, the companies can legally exclude them from their current financial statements. Alphabet alone disclosed in its third quarter 2025 filing that it had signed leases for $42.6 billion in future data center payments that won't appear on its balance sheet until the facilities are actually ready to operate. Just one quarter earlier, that same figure had been $23.9 billion—a jump of nearly $19 billion in unrecorded obligations in three months.
What makes this possible is the unique structure of modern AI infrastructure deals. Historically, data center leases ran for 10 to 15 years. But the semiconductor chips and specialized equipment required for artificial intelligence become obsolete in just four to six years. So tech companies are negotiating much shorter initial lease terms with renewal options, which allows them to argue they're not "reasonably certain" to renew—a threshold GAAP sets at greater than 70% confidence. If a company can't clear that bar, it doesn't have to record the renewal costs. The problem is that landlords won't build billion-dollar facilities without financial guarantees. To bridge that gap, the tech giants are offering what's called residual value guarantees, or RVGs. These are essentially promises to pay the landlord if the data center's market value drops below an agreed-upon level. Meta, for instance, signed data center leases starting in 2029 with a $28 billion residual value guarantee attached. Because Meta deemed it "not probable" that it would have to pay out on that guarantee, the entire $28 billion obligation stayed off the books.
Moody's analysts David Gonzales and Alastair Drake are flagging this as a systemic risk. They're careful to note that these companies haven't technically violated any rules—the liabilities are real, just deferred. "They have not yet received the services to trigger this liability as of this time, but they will," Gonzales told Fortune. The concern is that these accounting practices obscure the true financial strain on the industry. When these leases do commence, the cash outflows will be enormous and inflexible. If the AI market cools, if technology shifts, if demand doesn't materialize as expected, these companies will still be locked into multibillion-dollar commitments they can't easily escape. The financial flexibility that makes a company resilient in a downturn will be severely constrained.
To illustrate the scale, economist Torsten Slok at Apollo Global Management calculated that total data center capital spending is running at roughly $646 billion annually—about 2% of U.S. GDP, equivalent to the entire economic output of Singapore, Sweden, and Argentina combined. That's nearly 70% of what the U.S. spent on defense in 2025. Moody's has signaled it won't wait passively for these leases to commence. The rating agency said it's considering making "nonstandard adjustments" to how it calculates debt for these companies, potentially treating the hidden obligations as if they were already on the balance sheet. That could trigger credit rating downgrades, which would raise borrowing costs and send a signal to the broader market that the true financial picture is riskier than the published statements suggest. The question now is whether other rating agencies and regulators will follow, and whether the tech industry's accounting practices will finally catch up to the scale of what's being built.
Notable Quotes
The accounting liability is unlikely to reflect certain plausible future scenarios.— Moody's report
They have not yet received the services to trigger this liability as of this time, but they will.— David Gonzales, Moody's analyst
The Hearth Conversation Another angle on the story
Why does it matter that these leases aren't on the balance sheet yet? Aren't they still real obligations?
They're absolutely real, but invisible to anyone reading the financial statements. A banker or investor looking at Meta's debt-to-equity ratio won't see the $28 billion residual guarantee. That changes how risky the company looks on paper.
But the companies disclosed these leases in footnotes, right? Didn't they tell people?
They did, buried in SEC filings. The problem is that footnotes don't flow into the standard metrics—the debt ratios, the credit ratings, the models analysts use. A $662 billion obligation scattered across footnotes doesn't move the needle the way it would if it hit the main balance sheet.
So this is about the rules being outdated?
Partly. GAAP was designed in the 1930s for a different world. It assumes you only owe money when you've received the goods or services. But AI infrastructure is different—these companies are locking in decades of payments for equipment that becomes worthless in years. The structure exploits a gap between when the commitment is made and when the accounting rule kicks in.
What happens when the leases actually start?
The cash starts flowing out, and suddenly these companies have much less financial flexibility. If the AI market cools or technology shifts, they're still locked in. That's what Moody's is worried about—not fraud, but hidden fragility.
Could this trigger downgrades?
That's the real risk. Moody's said it's considering treating these hidden obligations as debt for rating purposes. If they do, and other agencies follow, you could see credit rating cuts that raise borrowing costs across the industry.