A viral claim swept financial markets and social platforms this week. Five U.S. technology giants hide $1.65 trillion in debt off their balance sheets. The number dwarfs their reported borrowings. It grew eightfold since 2022. And it supposedly signals an AI infrastructure binge financed by accounting tricks reminiscent of Enron.
But the math tells a more nuanced story. Those trillions represent future contractual payments. Not all qualify as debt. Few sit concealed. And the disclosures come straight from public SEC filings. The Finterm analysis published July 23, 2026, reconstructs the headline figure with precision. It totals $821.4 billion in uncommenced leases plus $829.2 billion in purchase and construction commitments across Alphabet, Amazon, Meta, Microsoft and Oracle. Simple addition. No mystery.
Yet the framing sparked alarm. X users amplified screenshots. One post declared the obligations “buried in data center leases and GPU supply contracts.” Another warned the commitments bind companies “regardless of what AI revenue does.” The anxiety feels familiar. Investors sense enormous bets on artificial intelligence. They question whether balance sheets reflect the full risk.
The Viral Claim Meets the Filings
Start with the reported debt. The five companies list roughly $430 billion in interest-bearing borrowings. That figure lands directly on their balance sheets and in accompanying notes. No one hides it. Total liabilities reach about $1.35 trillion. This bucket includes accounts payable, deferred revenue, accrued expenses and recognized lease obligations. Treating every liability as debt distorts the picture. It sets undiscounted future cash flows against a mixed stock of current obligations. Apples meet oranges. The comparison collapses.
The uncommenced leases illustrate the point. These contracts cover assets not yet ready for use. A data center under construction. Servers yet to ship. Accounting rules require lessees to record a liability and right-of-use asset when the lease commences. Before that moment, payments appear in footnotes. The nominal sums stretch across years or decades. Discounting to present value shrinks them. An offsetting asset appears. The economics differ sharply from traditional loans where cash changes hands immediately.
Purchase commitments follow similar logic. They bind companies to buy chips, power, equipment or construction services. Many qualify as executory contracts. The supplier must still deliver. The buyer pays upon performance. SEC rules mandate disclosure of material purchase obligations precisely because they can strain future liquidity without sitting on today’s balance sheet. They do not automatically equal debt. Some carry cancellation clauses. Others tie payments to milestones. The $1.65 trillion aggregates nominal amounts over long horizons. It does not mean $1.65 trillion comes due tomorrow.
And yet. The scale commands attention. Combined commitments equal nearly one year of the five companies’ revenue. For Oracle the total exceeds four times annual sales and tops total assets. Meta’s obligations surpass its yearly revenue. Amazon appears lighter at 29 percent of revenue. These ratios do not prove insolvency. They flag where scrutiny belongs. Long-term contracts for AI infrastructure carry real economic weight even if accounting defers recognition.
Recent coverage adds texture. The New York Times reported in November 2025 that Google, Meta, Microsoft and Amazon spent $112 billion on capital expenditures in a single quarter. DealBook noted the turn toward complex financing. Corporate debt. Securitizations. Private placements. Off-balance-sheet vehicles. McKinsey estimates $7 trillion in data-center investment needed by 2030. Hyperscalers seek capital without inflating leverage ratios or denting free-cash-flow metrics that drive valuations.
By May 2026 the Financial Times tallied a $725 billion AI spending spree across the sector. Free cash flow hit a decade low. Oracle joined the list of companies using off-balance-sheet structures. The pattern accelerated. So did questions about transparency.
Analysts differ on severity. Some view the commitments as prudent hedging in a winner-take-most technology race. Others see echoes of past excesses. A Bloomberg report from July 23, 2026, quoted technical accounting consultant Tom Selling. “The accounting treatment itself is in fashion,” he said. “But what if one of these companies was a house of cards and was propping itself up with this accounting treatment? To me, that’s the risk.”
The article highlighted variable interest entities, or VIEs. Meta formed one with Blue Owl Capital for a Louisiana data center. Its maximum exposure reached $46 billion according to SEC filings. Alphabet employs VIEs for leases and power infrastructure guarantees. Microsoft offers scant details yet confirms it avoids consolidating certain entities. These structures can keep associated debt off the parent’s balance sheet if the company does not hold controlling financial interest or primary beneficiary status. Rules evolved after Enron. They still permit judgment calls.
Gil Luria, head of technology research at D.A. Davidson, pushed back against blanket Enron analogies. “Enron’s crime wasn’t having special purpose vehicles,” he said. “Enron’s crime was hiding them.” The current disclosures, however imperfect, appear in 10-K and 10-Q filings. Readers must hunt across notes, liquidity tables and subsequent event disclosures. Different companies use inconsistent labels and time horizons. Friction exists. Concealment does not.
Critics also note that many commitments lock in costs for assets with uncertain returns. Goldman Sachs analysts have flagged limited measurable ROI on AI investments for most organizations. Yet GPU orders and power contracts continue. The bet assumes revenue will eventually catch up. If it does not, those obligations convert from footnotes to cash drains. Free cash flow already suffers. Leverage metrics look cleaner than underlying economics might suggest.
Investors face a classic information gap. Reported debt understates contractual burdens. Headline figures overstate immediate liability. The truth sits in the middle. Sophisticated credit analysts have long adjusted for operating leases and purchase commitments. They build pro-forma balance sheets. They model cash-flow timing. Retail investors and even some portfolio managers rely on simpler screens. Those screens miss the nuance.
Regulators watch. The SEC has increased focus on non-GAAP metrics and alternative performance measures. Lease accounting changed years ago precisely to bring more obligations onto balance sheets. Yet executory contracts and uncommenced leases remain in footnotes. Further tightening could force recognition of more liabilities. That would lift reported debt. It might also depress return-on-capital ratios and complicate executive compensation tied to those metrics.
So what now? Companies will keep building. AI demand shows few signs of abating. Capital markets remain open to tech issuers with strong balance sheets and growth narratives. The $1.65 trillion figure will fade from headlines. Until the next viral thread. Until a major player misses on earnings because capex overruns or utilization lags. Until an analyst downgrades on revised free-cash-flow forecasts that incorporate more of those commitments.
The lesson repeats. Accounting is not reality. It approximates. Disclosures matter more than face amounts. And in an era of trillion-dollar infrastructure bets, footnotes can hide risks only if readers choose not to read them. The debt isn’t hidden. But it does require work to see clearly.
WebProNews is an iEntry Publication