Major artificial intelligence companies have omitted at least $2 trillion in financial liabilities from their balance sheets, according to financial reporting published in August 2026.
The undisclosed liabilities have raised concerns among financial analysts about systemic risks in the technology sector that resemble accounting practices used before the 2008 global financial crisis.
An investigation by the Financial Times on 10 August 2026 revealed that major technology infrastructure providers, known as hyperscalers, use two primary accounting methods to keep massive financing commitments off their official balance sheets.
The first method takes advantage of United States accounting standards that permit lease commitments to be excluded from balance sheet liabilities until the lease term officially starts. Goldman Sachs analysts estimated that hyperscalers have kept approximately $1 trillion in lease liabilities off their balance sheets using this provision.
The Financial Times highlighted the structure of the $27.5 billion Hyperion bond, the largest corporate bond issue in history, as an example of this accounting practice. The Hyperion bond funds a data center designed for Meta, the parent company of Facebook.
Lease rules and corporate bond structures
The data center financed by the Hyperion bond will be owned by a separate financing firm in which Meta holds a 20 percent stake. Meta has agreed to lease the facility for 20 years beginning on 1 June 2029.
Because Meta guaranteed the cash flows for the project, credit rating agency S&P Global Ratings assigned an A+ rating to the Hyperion bond. However, because the lease does not start until June 2029, the $27.5 billion debt obligation does not appear as a liability on Meta balance sheets.
Financial analyst Georgios I. Matsos, writing for Greek financial news website Capital.gr on 20 August 2026, noted that Meta has assumed a binding legal obligation and borrowed $27.5 billion indirectly without reflecting the liability on its balance sheet.
Matsos previously viewed the rise in artificial intelligence stock valuations as justified by strong corporate earnings reported by Google in the first and second quarters of 2026. However, he wrote that recent accounting disclosures indicate broader structural issues in how technology infrastructure is financed.
Off-balance-sheet commitments and insurance funds
The second major method used by technology companies involves future purchase commitments for hardware and services. Morgan Stanley calculated that Alphabet, Microsoft, Amazon, Nvidia, and Oracle have entered into binding future orders totaling $982 billion that are omitted from balance sheet liabilities.
Together, off-balance-sheet leases and unrecorded hardware commitments account for nearly $2 trillion in undisclosed corporate liabilities across the major technology companies.
Analytic reports have also raised questions regarding the origin of capital funding these large scale investments. Suspicions have centered on whether US insurance companies owned by investment funds are directing policyholder money into artificial intelligence projects.
Under US regulatory frameworks, insurance companies are exempt from detailed investment disclosure rules for capital that is reinsured through secondary insurers. Bloomberg reported that insurance firms owned by investment funds have reinsured policyholder pension assets through offshore subsidiaries in Bermuda and the Cayman Islands, where regulatory disclosure requirements are less strict.
While official verification remains unavailable due to offshore opacity, analysts warn that any exposure of policyholder pension funds to technology infrastructure debt would amplify systemic risk across the broader financial system.
Parallels to the 2008 financial crisis
The accounting mechanisms currently used by artificial intelligence firms draw comparisons to the financial engineering that preceded the 2008 global banking collapse. The 2008 crisis began when US financial institutions granted subprime mortgages to high risk borrowers who did not qualify under standard credit assessment rules.
To bypass credit risk regulations, banks packaged subprime mortgages into complex securitised financial products. Many of these structured securities received AAA credit ratings, the highest safety grade available from rating agencies.
Under Financial Accounting Standard 140, a US GAAP rule repealed in June 2009, banks legally transferred securitised subprime loans to controlled offshore entities that were not consolidated into parent financial statements. This allowed $1.3 trillion in non-performing mortgage loans to disappear from corporate financial reporting.
The subprime crisis spread globally on 15 September 2008 when investment bank Lehman Brothers filed for bankruptcy protection, triggering a widespread credit contraction.
Market risks and competitive pressures
Unlike the subprime mortgage instruments of 2008, artificial intelligence represents a functioning technology product with commercial utility. However, market analysts highlight significant commercial risks associated with capital recovery.
A scenario outlined by Forbes noted that American technology companies could face financial strain if enterprise customers choose lower cost artificial intelligence services developed in China rather than higher priced US offerings. If US providers fail to attract sufficient paying customers, revenue shortfalls could mirror the default rates seen among subprime mortgage borrowers.
Foreign Policy described the scale of US investment as a cosmic bet on artificial intelligence, warning that persistent accounting opacity creates vulnerabilities across international financial markets if commercial revenues fail to meet expectations.
