Big Tech keeps $300 billion in AI infrastructure guarantees off the books
The world's largest technology companies have extended up to $300 billion in guarantees to finance artificial intelligence infrastructure over the past year, with most of this exposure kept off their balance sheets. According to the Financial Times, these commitments back data centres and semiconductor chips financed through independent vehicles, allowing the tech groups to put their financial muscle behind the projects without fully recognising the liabilities in their accounts.
The structure matters for investors and creditors because it obscures the true scale of risk. These so-called residual value guarantees establish a minimum future value for computing equipment or data centres held by special-purpose entities. If the asset sells or leases for less than the agreed amount, the technology company commits to covering the shortfall.
How the guarantees work in practice
Meta pioneered this approach with a data centre in Louisiana, providing a $28 billion guarantee. Broadcom assumed an exposure of $29 billion to finance chips leased to Anthropic, an AI research company. Nvidia went further, offering $105 billion in guarantees to SB Energy for a data centre serving OpenAI in Ohio.
The model delivers financing at rates close to what the tech companies themselves pay, since lenders treat the guarantees as near-equivalent to direct corporate debt. In exchange, the guarantors absorb the risk that assets lose value — whether through chip obsolescence, excess installed capacity, or weaker-than-expected demand for AI services.
Morgan Stanley calculates that seven large technology and semiconductor companies now hold more than $3.1 trillion in off-balance-sheet commitments and credit support. Credit rating agencies have begun adjusting debt levels to reflect part of these obligations.
Why the risk is hidden
Under current accounting rules, companies record only a fraction of their actual exposure from these guarantees. They treat the commitments as contingent liabilities, booking the modelled probability of loss rather than the full amount guaranteed. The remainder stays off the books until a payment becomes probable.
This matters because the sums involved are vast. Global spending on AI is expected to reach about $2.7 trillion in 2026, a 49.5% increase from the previous year. Of that, roughly $1.48 trillion will go to AI infrastructure. The four largest US hyperscalers alone plan to invest around $700 billion in 2026, mainly in data centres, chips and AI networks.
Rating agencies warn that the race to build AI infrastructure is consuming free cash flow and threatening credit quality. Five major US tech companies held approximately $2.8 trillion in AI-related off-balance-sheet obligations as of September 2026. As commitments such as data centre leases gradually appear on balance sheets, adjusted debt levels will rise significantly.
The Portuguese and European context
For companies operating in Portugal or the European Union, accounting rules derive from the International Financial Reporting Standards, adopted locally through the Sistema de Normalização Contabilística. The relevant standard for guarantees is IAS 37, which governs provisions, contingent liabilities and contingent assets.
A provision is recognised when a company has a present obligation — legal or constructive — from a past event, it is probable that an outflow of resources will be needed to settle it, and a reliable estimate can be made. Contingent liabilities, where the outflow is only possible rather than probable, are not recognised in the financial statements but must be disclosed in the notes.
This means that technology groups such as Meta and Nvidia, when preparing accounts for EU subsidiaries, must apply these standards. The guarantees would typically be disclosed as contingent liabilities unless a payment becomes likely enough to warrant recognising a provision.
What changes for investors and creditors
The issue is not that these structures are illegal or hidden — they appear in footnotes to financial statements. The concern is that rating agencies and investors may be underestimating the true exposure. Fitch Ratings has identified a potential correction in the AI market as one of the biggest credit risks to the global economy in 2026, noting that stock valuations are near levels seen during the dot-com boom.
There is also a circularity to the ecosystem. Tech companies often invest in AI laboratories that then spend heavily on their cloud services. If demand fails to meet expectations, that "circular AI ecosystem" could amplify systemic risks.
For now, the largest players maintain robust balance sheets. But as the commitments grow, the gap between what appears on the books and what the companies ultimately guarantee widens. Investors watching the AI boom would be wise to read the footnotes.
