Five US Tech Giants Sit on $1.65T in Hidden AI Debt

Five of America's largest technology companies, including Alphabet, Microsoft, Amazon, Meta, and Oracle, are sitting on $1.65 trillion in financial obligations that do not appear on their official balance sheets, according to a new study by Nikkei Asia published on July 20. That figure has grown roughly eightfold over the past four years, and it now exceeds the $1.35 trillion in formal debt these companies do report, raising questions about the true scale of risk embedded in the global AI buildout.

Four of the five companies are set to report second-quarter earnings this week. The $1.65 trillion sitting in the financial footnotes almost certainly will not make the lead.

What the "Hidden Debt" Actually Is

The obligations in question are disclosed, but in the fine print of quarterly filings rather than on the balance sheet itself. The bulk of the figure comes from two sources:

• Long-term data center lease agreements that have not yet commenced
• Multi-year GPU and server supply contracts signed to secure AI computing capacity

Under current US accounting rules, these commitments only move onto the balance sheet once a facility becomes operational or a contract begins. Until then, they live in the footnotes, visible to anyone who reads them but easy to miss in standard credit analysis.

The Bank for International Settlements and Moody's had each independently flagged this structural gap before the Nikkei study attached specific dollar figures to named companies. In March 2026, a BIS research paper described the mechanism as "shadow borrowing." Moody's earlier warned that uncommenced lease commitments alone were growing at a pace that standard leverage ratios were not capturing. S&P has already acted on the concern in at least one case, cutting Oracle's credit rating over what it described as stretched leverage.

• Meta stands out as the most exposed among the five. The company's off-balance-sheet obligations are estimated at approximately $420 billion, nearly triple the debt it formally reports.
• Oracle's hidden liabilities have grown the fastest, reaching $273 billion by the end of May, a figure more than 30 times higher than it was four years ago.

The combined visible and hidden obligations of all five companies total roughly $3 trillion, according to estimates cited by AI Weekly, nearly double what appears in standard credit models used by most institutional investors.

Why the Risk Is Harder to Price Than It Looks

Server hardware, the core asset inside these data centers, typically needs to be replaced every 18 to 36 months as chip generations advance. The bonds being issued to finance the buildout often carry maturities of five to twenty years. That gap creates what analysts at AI Weekly describe as a "refinancing cliff" – the asset depreciates far faster than the debt used to fund it.

"What if one of these companies was a house of cards, and was propping itself up with this accounting treatment?" – Tom Selling, accounting consultant, quoted by The Next Web

If AI demand grows more slowly than projected, data center utilization rates fall. Underutilized facilities then require write-downs, and those losses fall not on the tech giants, who have structured much of this debt off-balance-sheet through special purpose vehicles and private credit arrangements, but on the lenders, bond investors, and insurers who funded the structures.

• Morgan Stanley has flagged the rising leverage in a client note
• Moody's has warned about the scale of uncommenced lease commitments
• S&P cut Oracle's credit rating over stretched leverage
• BIS called the mechanism "shadow borrowing" in a March 2026 research paper
• Michael Burry, the investor known for calling the 2008 subprime crisis, highlighted the Nikkei findings on X on July 21

The five companies are projected to spend $725 billion in capital expenditure in 2026 alone, nearly five times what they spent in 2023. And because AI investment spending has already outpaced cash generation, the group issued roughly $121 billion in new corporate bonds in 2025, up from approximately $40 billion in 2020, a tripling of bond issuance in five years.

The Crypto Angle

One of the clearest pressures on Bitcoin and broader crypto prices since May 2026 has been institutional rotation into AI-related equities. Spot Bitcoin ETFs recorded over $8 billion in outflows between May and June as capital chased AI stocks higher.

If the AI trade begins to reprice, and this week's earnings season, with four of the five companies reporting, is the next stress test, some of that capital could reverse direction. The same institutions pulling money from crypto to fund AI equity exposure may find themselves reassessing that rotation if AI balance sheet risk starts to be priced more seriously.

Bitcoin miners pivoting to AI hosting, a trend accelerating sharply in 2026, are directly dependent on Big Tech's willingness to sign long-term compute contracts. Riot Platforms, IREN, Core Scientific, and TeraWulf have all announced AI and HPC deals premised on that demand holding. If hyperscaler capex moderates in response to balance sheet scrutiny, the pipeline of new mining-to-AI conversion contracts could slow.

Sources

• Nikkei Asia – Five US tech giants' hidden debts soar to $1.65tn on opaque AI funding: https://asia.nikkei.com/business/technology/five-us-tech-giants-hidden-debts-soar-to-1.65tn-on-opaque-ai-funding
• The Next Web – Big Tech is hiding $1.65tn in off-balance-sheet AI debt: https://thenextweb.com/news/tech-giants-hidden-off-balance-sheet-debt-ai
• Yahoo Finance – Five Tech Giants Are Using Enron's Accounting Strategy to Conceal $1.65 Trillion in AI Debt: https://finance.yahoo.com/technology/ai/articles/five-tech-giants-using-enron-184404206.html
• AI Weekly – Five US tech giants' hidden AI debt hits $1.65T, Nikkei says: https://aiweekly.co/alerts/five-us-tech-giants-hidden-ai-debt-hits-165t-nikkei-says