How five US tech giants hid a mountain of AI debt from balance sheets
Off-balance-sheet debt from data centre leases, GPU supply contracts, and AI infrastructure SPVs. Data from Nikkei Asia study (Jul 2026).
Estimated total off-balance-sheet debt across the five giants (Nikkei study). The inflection point coincides with the generative AI boom.
Instead of buying a $5B data centre outright, Meta (for example) signs a 10–15 year lease with an SPV — a special purpose vehicle set up specifically for that asset.
The SPV takes out loans from banks or private credit funds to build the data centre. Because the debt is in the SPV's name, it never appears on the tech giant's balance sheet.
The lease payments show up as operating expenses (OPEX), not debt. This keeps headline leverage ratios low and EBITDAR strong. Institutional investors can see through this; retail investors often cannot.
If the AI bubble bursts and data centres sit empty, the SPVs default. The tech giants can walk away (no liability on their balance sheets). The losses land on banks and private credit funds — which is where retail pensions and endowments have been increasingly invested. We all own a piece of this debt.
$1.65T exceeds the combined on-balance-sheet debt of all five companies (~$810B). The market has been pricing these off-book commitments as if they're the companies' problem — but legally, they're not.
Parallels to 2008: off-balance-sheet vehicles (SIVs then, SPVs now), opacity, and systemic interlinkage with the banking sector. The difference: the assets (GPU clusters) are real and could be repurposed — but at a fraction of the build cost.
📉 Stocks tumbling today: Nvidia and Big Tech are down as this study circulates and Chinese AI model fears compound the sentiment.