Background & Context§
The relentless race to dominate artificial intelligence has driven the world's largest technology companies into an unprecedented infrastructure spending spree. Data centers, graphics processing units (GPUs), and networking equipment form the backbone of large-scale AI model training and inference. However, much of this spending is structured through operating leases, supply contracts, and special-purpose entities that do not appear on traditional balance sheets. This practice, while common in capital-intensive industries, has reached a scale that now threatens to obscure the true financial health of the biggest names in tech.
The News: What Happened Exactly§
A recent investigation by Nikkei has quantified the scale of these hidden obligations. The study estimates that off-balance-sheet debt at five US technology giants — Meta, Oracle, and three other unnamed major firms — has ballooned to approximately $1.65 trillion as of early 2025. This figure represents an eightfold increase from roughly $200 billion around 2020, soaring in parallel with the explosion of AI investment. Crucially, this hidden debt now exceeds the companies' combined on-balance-sheet debt, making it significantly harder for investors, analysts, and regulators to assess risk exposure.
For Meta, the study found that off-balance-sheet liabilities amount to about $420 billion, nearly three times its transparent debt. These liabilities primarily stem from long-term leases for data center space in states like Georgia, as well as multi-year purchase commitments for NVIDIA GPUs and other AI hardware. Similarly, Oracle has locked in massive data center capacity under operating leases and GPU supply agreements, contributing to its share of the hidden total.
The mechanism behind this off-balance-sheet treatment is straightforward. Under accounting standards such as US GAAP and IFRS, operating leases are not recorded as liabilities on the lessee's balance sheet if they meet certain criteria. Additionally, take-or-pay contracts for GPUs — where a company commits to purchasing a fixed volume of chips over several years — are typically disclosed only in footnotes. As a result, tens of billions in future obligations remain invisible to standard debt-to-equity calculations. The Nikkei analysis aggregated these commitments across the five firms using public filings, regulatory disclosures, and lease data to arrive at the $1.65 trillion estimate.
This revelation has significant implications for financial analysis. Traditional metrics like net debt or enterprise value understate true leverage by a wide margin. For instance, if Meta's $420 billion in hidden debt were recognized, its debt-to-EBITDA ratio would jump from below 1.0x to over 4.0x, crossing the threshold considered risky by many credit rating agencies. Moreover, the AI-driven demand for computing power is unlikely to abate, meaning these commitments will continue to grow. The study underscores a critical blind spot in corporate disclosure: investors may be underestimating the financial vulnerability of tech giants should the AI boom slow or hardware prices decline.
Historical Parallels & Similar Incidents§
The current situation bears striking resemblance to the off-balance-sheet debt crisis in the energy sector during the early 2000s, particularly the Enron scandal. Enron famously used special purpose entities (SPEs) to hide billions in debt, inflating its apparent financial strength while its actual liabilities were mounting. Although the accounting rules have since been tightened, parallels remain. In both cases, companies used complex legal structures to keep large obligations off the books, betting that future cash flows would cover the commitments. The key difference is that Enron's hidden debt was fraudulent and designed to deceive, whereas today's tech giants are operating within legal accounting frameworks — albeit ones that some argue are outdated for the scale of AI spending.
Another apt parallel is the airline industry's use of operating leases for aircraft. Airlines routinely lease planes through operating leases to avoid showing the full purchase cost as debt. For example, Delta Air Lines had over $17 billion in off-balance-sheet aircraft leases in 2019, representing roughly 30% of its total liabilities. However, the airline industry's leases are typically shorter-term (10-15 years) and the assets (aircraft) have high residual value. In contrast, AI data center leases are often 15-20 years, and GPU hardware depreciates rapidly due to technological obsolescence. The hidden debt burden per dollar of revenue is also far larger for tech companies. For instance, Meta's off-balance-sheet liabilities equal about 4x its annual revenue, whereas Delta's were around 0.5x. This comparison highlights the unprecedented magnitude and risk inherent in the tech sector's AI spending.
Lesson: When large, opaque debt liabilities accumulate, a downturn in the underlying business can trigger a wave of covenant violations, asset impairments, and refinancing crises. For AI companies, a sudden drop in demand for compute services or a breakthrough that renders current GPU architectures obsolete could expose the fragility of these hidden debts. Regulators and investors would be wise to push for greater transparency, such as requiring capitalized lease treatment for all long-term data center commitments and mandatory disclosure of GPU take-or-pay contracts as debt equivalents. Otherwise, the $1.65 trillion iceberg could pose a systemic risk to the technology sector.