AI’s Hidden Debt: The $1.7 Trillion Financing Risk

Home / Reports / Dark Side of AI / AI’s Hidden Debt: The $1.7 Trillion Financing Risk

Thank you for reading this post, don't forget to subscribe!

In July 2026 a Nikkei study put a startling number on the AI build-out: the five hyperscalers funding it — Alphabet, Microsoft, Amazon, Meta and Oracle — carry roughly $1.65 trillion of “off-balance-sheet future obligations,” a figure that has grown about eightfold since 2022 and now exceeds their combined reported debt of about $1.35 trillion (Nikkei Asia, “Five US tech giants’ hidden debts soar to $1.65tn on opaque AI funding,” Jul 2026). A Prof G Markets chart rounded it to “$1.7 trillion of hidden AI debt.” The number is real and disclosed — but whether “hidden debt” is the right label for it is exactly what is in dispute.

What the $1.7 trillion actually is

The headline comes from a Nikkei study, publicized in mid-July 2026 and amplified by a Prof G Markets chart titled “Off-balance-sheet future obligations (Trillions of dollars).” Nikkei summed disclosed items across Alphabet, Microsoft, Amazon, Meta and Oracle — data-center leases that have not yet started, GPU and cloud purchase commitments, and obligations tied to joint ventures — and arrived at roughly $1.65 trillion, up about eightfold since 2022 and now larger than the group’s ~$1.35 trillion of on-balance-sheet debt (Nikkei Asia, Jul 2026). The concentration is uneven: Nikkei attributes about $420 billion to Meta alone, and roughly $273 billion to Oracle, a figure it says has jumped some 30-fold in four years, much of it tied to the “Stargate” project Oracle is building with OpenAI (Nikkei Asia, Jul 2026; Analysis.org, Jul 2026).

The crucial nuance: these are not secret liabilities. Every component is disclosed in 10-K and 10-Q footnotes under standard accounting (operating-lease commitments, unconditional purchase obligations, unconsolidated entities). “Off-balance-sheet” here is a technical status — leases that have not yet commenced and undelivered purchase commitments do not hit the balance-sheet liability line until the facility switches on or the hardware arrives — not a synonym for concealment. Moody’s, tracking the same universe, estimates lease commitments of about $1.2 trillion across the group, with more than $820 billion from leases that have not yet begun (Moody’s, via CNBC, 24 Jul 2026).

How it is structured off the balance sheet

Three mechanisms do most of the work. First, long-dated data-center leases: rather than buy and finance a building directly, a hyperscaler signs a multi-year lease, which shows up as a future commitment rather than debt until it commences. Second, special-purpose vehicles and joint ventures. The marquee example is Meta’s Hyperion campus in Richland Parish, Louisiana: in October 2025 Meta and Blue Owl Capital closed a roughly $27–30 billion financing — about $27 billion of debt plus equity — through an SPV in which Meta holds a 20% stake, with PIMCO as anchor investor and Morgan Stanley arranging; the bonds mature in 2049 and were rated A+ by S&P. It was the largest private-credit transaction on record (Meta/Blue Owl press release, Oct 2025; Bloomberg; DatacenterDynamics, Oct 2025). Because Meta consolidates only its minority share, most of that debt sits at the fund level, not on Meta’s balance sheet.

Third, private credit is increasingly the lender of record. Total data-center debt issuance nearly doubled to about $182 billion in 2025, and Morgan Stanley projects private credit could supply roughly $800 billion more; Blue Owl alone financed more than $50 billion of data centers in 2025 (Morgan Stanley, 2025-26; Bisnow; PE Insights). Alphabet, for its part, has disclosed about $40.7 billion of future funding commitments to off-balance-sheet vehicles, including a large equity-derivative structure and a Blackstone joint venture (FactSet, 2026) — evidence that the arrangements are documented in filings rather than truly hidden.

The systemic-risk case

Critics argue the structure rhymes with past crises. Newsletter author and Prof G Markets guest Ed Zitron calls the $1.65 trillion “a corporate scandal” that “should be front page news everywhere” and “should be a shareholder riot” (Prof G Markets, via Benzinga, Jul 2026). In his own writing he frames AI data centers as “the subprime” of this cycle, with SPV financing he likens to the CDOs of 2008, and argues that generative AI is “a $30 billion TAM industry pretending to be a trillion-dollar one, propped up by circular financing from NVIDIA” (Ed Zitron, “The Subprime Data Center Crisis,” 2025). The circularity concern is concrete: Nvidia invests in and sells GPUs to firms like OpenAI, which commits enormous sums to Oracle and other clouds, which in turn buy more Nvidia chips — money that can loop back on itself, making organic demand hard to measure (Business Standard, Sep 2025). Michael Burry, of “Big Short” fame, adds an earnings-quality charge: in November 2025 he argued hyperscalers understate depreciation by stretching the useful life of chips on a 2–3 year cycle, calling it “one of the more common frauds of the modern era” and estimating a ~$176 billion understatement across 2026–2028; he disclosed put options against Nvidia and Palantir in his Q3 2025 filing (CNBC, 11 Nov 2025; Sherwood News, 2025). Scott Galloway has warned there could be “nowhere to hide” in markets if the OpenAI story unravels (via AOL, 2026).

The counter-case: disclosed, cash-backed, normal finance

The other side is not merely reassurance. Much of the $1.7 trillion is ordinary lease and purchase-commitment accounting that every large company reports — the same footnotes retailers use for store leases — not covert borrowing. The obligations are backed by businesses that throw off enormous cash: these are among the most profitable firms on earth, with real, growing cloud revenue funding most of the capex. Bank of America argues the hyperscalers still have room for “elevated debt issuance” because gross leverage remains low even after a record bond binge (BofA, via AOL, 2026). And leases are not the same as reckless leverage: a signed lease is an expense against future revenue, not a margin loan that can be called. Analysts who model the arc out to 2029 project operating cash flow rising far faster than capex — one estimate has combined operating cash flow reaching roughly $1.1 trillion — which would let the companies grow into these commitments if AI demand holds (FactSet/Epoch AI, 2026). On this reading, “hidden debt” is a framing choice applied to disclosed, serviceable obligations.

What would have to go wrong

The bear case turns on three hinges. Demand: if enterprise and consumer AI revenue disappoints, the leases and GPU commitments still have to be paid — obligations that behave like debt precisely because they are owed regardless of usage. Refinancing and duration: AI servers may need replacing every 18–36 months while the bonds financing them run 5–20 years (Hyperion’s mature in 2049), a mismatch standard credit models can miss. Ratings and rates: S&P already cut Oracle to BBB-, one notch above junk, in July 2026, citing AI-buildout credit risk, and Moody’s has warned that “unprecedented” spending threatens the credit quality of Amazon, Meta, Alphabet and others as off-balance-sheet leases migrate onto balance sheets when facilities go live (S&P, Jul 2026; Moody’s, via CNBC, 24 Jul 2026). If Burry is right that depreciation is understated, reported earnings would also be flattered today and pressured later. None of these is a prediction; each is a condition that would convert a manageable obligation into a strained one.

How worried to be

Somewhere between panic and complacency. The verified core is solid: five firms have taken on roughly $1.65 trillion of disclosed, off-balance-sheet obligations to build AI infrastructure, private credit is deep in the financing, and rating agencies are flagging the trend. That is a genuine concentration of risk worth watching. But the loudest label — “hidden debt” — overstates the concealment: these numbers come from the companies’ own filings, they sit atop real cash flows, and leases are not the same as fragile leverage. The honest summary is that the risk is real, disclosed and contingent on AI demand living up to the spend — not a smoking gun, and not nothing.

Frequently asked questions

Go deeper

• The Dark Side of AI report series
• AI investment and funding statistics 2026
• AI-native companies 2026
• AI Bubble Tracker — forthcoming

Sources

• Nikkei Asia, “Five US tech giants’ hidden debts soar to $1.65tn on opaque AI funding,” Jul 2026 – https://asia.nikkei.com/business/technology/five-us-tech-giants-hidden-debts-soar-to-1.65tn-on-opaque-ai-funding
• Analysis.org, “Hidden Debt at Five AI Hyperscalers Hits $1.65 Trillion, Nikkei Study Finds,” Jul 2026 – https://analysis.org/hidden-debt-at-five-ai-hyperscalers-hits-1-65-trillion-nikkei-study-finds/
• Ed Elson / Prof G Markets, “AI Has A Hidden Debt Problem,” Jul 2026 – https://www.profgmedia.com/p/ai-has-a-hidden-debt-problem
• Benzinga, “Big Tech’s Off-Balance-Sheet AI Debt Reportedly Tops $1.65 Trillion: Ed Zitron Says It ‘Should Be a Shareholder Riot’,” Jul 2026 – https://www.benzinga.com/markets/prediction-markets/26/07/60654488/big-techs-off-balance-sheet-ai-debt-reportedly-tops-1-65-trillion-ed-zitron-says-it-should-be-a-shareholder-riot
• Ed Zitron, “The Subprime Data Center Crisis,” Where’s Your Ed At, 2025 – https://www.wheresyoured.at/the-subprime-data-center-crisis/
• Meta, “Meta Announces Joint Venture with Funds Managed by Blue Owl Capital to Develop Hyperion Data Center,” Oct 2025 – https://investor.atmeta.com/investor-news/press-release-details/2025/Meta-Announces-Joint-Venture-with-Funds-Managed-by-Blue-Owl-Capital-to-Develop-Hyperion-Data-Center/default.aspx
• DatacenterDynamics, “Meta taps Pimco and Blue Owl for $29bn data center financing,” Oct 2025 – https://www.datacenterdynamics.com/en/news/meta-taps-pimco-and-blue-owl-for-29bn-data-center-financing-report/
• CNBC, “Moody’s says ‘unprecedented’ AI spending threatens credit quality of Amazon, Meta, Alphabet and others,” 24 Jul 2026 – https://www.cnbc.com/2026/07/24/moodys-ai-spending-credit-quality-amazon-meta-alphabet.html
• CNBC, “‘Big Short’ investor Michael Burry accuses AI hyperscalers of artificially boosting earnings,” 11 Nov 2025 – https://www.cnbc.com/2025/11/11/big-short-investor-michael-burry-accuses-ai-hyperscalers-of-artificially-boosting-earnings.html
• Sherwood News, “Michael Burry discloses options bets against Nvidia and Palantir,” 2025 – https://sherwood.news/markets/michael-burry-big-short-discloses-1-1-billion-options-bet-against-nvidia-palantir-puts/
• Business Standard, “Nvidia-OpenAI deal sparks concerns over circular financing in AI boom,” Sep 2025 – https://www.business-standard.com/companies/news/nvidia-openai-deal-sparks-concerns-over-circular-financing-in-ai-boom-125092401589_1.html
• FactSet, “Hyperscalers Tap External Financing as AI Capex Outruns Cash Flow,” 2026 – https://insight.factset.com/hyperscalers-tap-external-financing-as-ai-capex-outruns-cash-flow
• Bisnow, “A quiet corner of private credit has fueled $60B in data center development,” 2025-26 – https://www.bisnow.com/national/news/data-center-development/a-quiet-corner-of-private-credit-has-fueled-60b-in-data-center-development-134982
• Forbes, “Big AI Data Center Owners Are Massively Expanding Their Debt,” 23 Jul 2026 – https://www.forbes.com/sites/eriksherman/2026/07/23/big-ai-data-center-owners-are-massively-expanding-their-debt/