A new study by Nikkei has found that America’s five largest AI infrastructure companies hold roughly $1.65 trillion in off-balance-sheet debt, more than the $1.35 trillion they report on their main balance sheets. Alphabet (GOOGL), Meta (META), Microsoft (MSFT), Amazon (AMZN) and Oracle (ORCL) are carrying more liabilities in financial footnotes than on their primary statements.
The five hyperscalers are spending hundreds of billions of dollars building massive data centers to support AI workloads. Those investments may take years to generate returns, putting pressure on reported financial ratios. To keep balance sheets cleaner, the companies use accounting structures permitted under U.S. GAAP to shift part of that debt out of the main statement.
A common tool is the special purpose vehicle, or SPV. The hyperscalers partner with private equity to create separate entities that borrow money to build data centers. The tech companies typically take minority stakes or sign long-term lease and service agreements. Because they lack controlling interests in the SPVs, they do not have to list the related debt on their primary balance sheets. Instead, the contractual commitments are disclosed in the footnotes of quarterly filings.
The immediate risk is not solvency. Alphabet alone reported net income of $112.1 billion and operating income of $40.7 billion in the second quarter of 2026, and the five companies remain highly profitable. But valuations may be based on a misleading picture of financial health. If off-balance-sheet obligations were included, debt-to-equity ratios would more than double. Moody’s has warned that it considers such debt when assigning credit ratings. A downgrade could blindside investors who never looked into the footnotes.
Bank of America analyst Savita Subramanian estimates the five companies have about $2.3 trillion in footnote debt, even higher than the Nikkei figure. Adding that amount to balance sheets would push the S&P 500’s debt-to-capital ratio from 19% to 21%. She argues that higher leverage is closely linked to equity risk premiums and could compress earnings multiples. If earnings per share stay unchanged, a 30% contraction in valuation multiples would translate into a roughly 30% decline in share prices.
Off-balance-sheet debt is not new. Enron used special purpose entities to hide toxic assets and debt, and banks such as Citigroup and HSBC used structured investment vehicles during the subprime bubble. But there is no evidence the AI hyperscalers are committing fraud or evading regulations. Analysts note the structures are legal and backed by hundreds of billions of dollars in cash flow from core advertising and cloud businesses.
For investors, the bigger concern may be valuation shock and margin compression rather than default. If AI revenue monetization stalls or takes longer than expected, fixed off-balance-sheet obligations could eventually return to income statements as margin drag or asset write-downs. That could trigger a sharp sell-off in technology stocks that spreads across the broader market.
