The $2 Trillion Hole in Big Tech's Balance Sheets

The Bank for International Settlements (BIS) coined the term "shadow borrowing" in its March 2026 Quarterly Review to describe these obligations — arrangements that are economically equivalent to debt but largely sit outside corporate balance sheets . The BIS later listed AI-driven asset bubbles alongside sovereign debt as top systemic risks, warning the investment boom could end in a bust if demand fails to materialize .

The Five Financial Techniques — and Which Company Uses Each

1. Special-Purpose Vehicles (SPVs) — Meta's 'Beignet' Structure

Meta employed a classic off-balance-sheet structure for its $27.3 billion Hyperion AI data center project in Louisiana. In October 2025, Meta moved the entire project into a joint venture with private-credit firm Blue Owl Capital, holding only 20% while Blue Owl (via a holding company called Beignet Investor LLC) owns 80% . Seven Delaware-registered companies incorporating the word "Beignet" were created to execute this structure .

Meta's balance sheet records only a $2.37 billion investment related to the project; the remaining ~$25 billion in debt is invisible to investors . Meta's auditor, Ernst & Young, raised a red flag over this accounting treatment . The BIS describes this structure as typical: the hyperscaler holds a minority stake, commits to long-term operating leases or capacity offtake agreements, and may provide guarantees — substituting upfront capex with multi-year operating expenses while keeping the debt off-balance-sheet .

2. Finance Leases — Microsoft's $62.9 Billion

Microsoft reported $62.9 billion in finance lease liabilities as of its Q3 FY2026 (nine months ended March 31, 2026), a figure that appears within its "other liabilities" line rather than as explicit long-term debt . The company's total lease-related obligations (including operating leases and imputed interest) reached $178.7 billion on a gross basis in its FY2025 10-K filing . Additional datacenter leases not yet commenced total $196.6 billion .

3. Credit-Enhancement Derivatives / Guarantees — Google's $43.8 Billion

Google (Alphabet) has provided $43.8 billion in credit guarantees — backstopping lease payments on data centers owned by third parties should the tenant default — but recorded this exposure at only $815 million fair value on its balance sheet . Additionally, Alphabet has $7.6 billion in financial guarantees for energy counterparties to order long-lead power equipment . This guarantee structure expanded nearly sevenfold from $6.5 billion in September 2025 to $44 billion by July 2026 .

4. Residual Value Guarantees

Residual value guarantees function as a form of off-balance-sheet cushioning. Moody's has flagged that such guarantees can expose hyperscalers to unexpected capital calls if data-center asset values decline, especially amid rapid AI chip depreciation cycles .

5. Uncommenced (Future-Dated) Lease Agreements — Amazon and Oracle

Amazon has $106.3 billion in future-dated lease commitments not yet reflected on its balance sheet . Oracle carries approximately $260 billion in similar uncommenced lease obligations, representing the largest single off-balance-sheet exposure of any of the five companies . Moody's reported that total lease commitments across the group ballooned to $1.2 trillion, of which more than $820 billion comes from leases that haven't yet commenced — meaning the data centers are still being built .

The Systemic Risk: BIS Warns of 'Shadow Borrowing'

The BIS's March 2026 Quarterly Review, written by officials Egemen Eren, Ingomar Krohn, and Karamfil Todorov, found that hyperscalers had quietly moved more than $120 billion in AI compute debt off their balance sheets in about 18 months . Key findings from the report:

• Hyperscalers have turned to off-balance-sheet arrangements in partnership with private credit firms to finance infrastructure expansions .
• A typical structure involves a dedicated vehicle (JV or SPE) that acquires or develops data center assets; the hyperscaler provides long-term leases or capacity offtake agreements and guarantees .
• Economically, this substitutes upfront capex with multi-year operating expenses while keeping debt off traditional balance sheets .
• These arrangements deepen connections among Big Tech companies, private-credit funds, and insurers — creating new channels for financial shocks to transmit .

Market Reactions: Credit Stress Signals Are Flashing

S&P's Downgrade of Oracle to BBB−

S&P lowered Oracle's rating to BBB− (one notch above junk) in July 2026, driven by the company's enormous off-balance-sheet AI commitments .

Oracle's CDS Spread Surpasses 2008 Peak

Oracle's credit default swap (CDS) spread has widened beyond its levels during the 2008 financial crisis, reflecting extreme market concern about the company's hidden leverage .

Alphabet's Negative Free Cash Flow

Alphabet posted negative free cash flow in Q1 2026 for the first time since 2004, driven by massive AI capex spending that accrues largely through off-balance-sheet channels . A legal analysis notes Alphabet and Meta are both expected to generate negative free cash flow of up to $28 billion in 2026 .

Moody's 'Circularity' Warning

Moody's warned in July 2026 about "circularity" among interconnected AI investments: hyperscalers are simultaneously the largest customers, financiers, and equity partners in AI data center SPVs, creating interlocking exposures that could amplify losses in a downturn . Moody's specifically said AI spending "threatens credit quality" at Amazon, Meta, and Alphabet .

Caveats and Discrepancies

The $2.13 trillion figure appears in investigative reports, while Nikkei's analysis puts the number at $1.65 trillion; the discrepancy likely reflects different cut-off dates and coverage of additional financial instruments . Some specific data points (Oracle CDS surpassing 2008 peak, S&P downgrade to BBB−, Alphabet's negative FCF) are referenced in secondary reporting rather than confirmed by primary source documents found in this search. The figures are estimates based on financial statement analysis and may not be directly comparable across all five companies due to different disclosure standards.

The bottom line: An estimated $1.65–$2.13 trillion in "shadow borrowing" for AI data center construction sits off the balance sheets of five major US tech companies, using SPVs, credit guarantees, finance leases, residual value guarantees, and uncommenced leases. The BIS has flagged this as a systemic risk, and markets have responded with Oracle's downgrade to near-junk, CDS spreads exceeding 2008 levels, Alphabet's first negative free cash flow in over two decades, and Moody's warning of dangerous circularity in AI financing.