Big Tech has accumulated $300 billion in hidden guarantees due to the AI boom

The largest American tech companies are increasingly using financial guarantees to accelerate the construction of data centers and the purchase of AI chips without directly increasing debt on their own balance sheets. According to Financial Times estimates, over the last 12 months alone the volume of such guarantees reached up to $300 billion.

This primarily refers to so-called residual value guarantees. A data center or computing equipment is financed by a separate structure that raises debt from banks and investors. The tech company guarantees that the asset will be worth at least a certain amount in the future. If after sale or end of lease its value turns out to be lower, the company may be forced to cover the difference.

Such a mechanism allows companies to use their own credit rating to reduce financing costs, but without recording the full project amount as regular corporate debt. At the same time, the risk remains: if demand for computing power turns out to be weaker than expected or equipment quickly depreciates due to the emergence of new chip generations, the guarantees may turn into real financial obligations.

One of the first to apply this model on a large scale was Meta. For a joint data center project with Blue Owl in Louisiana, called Hyperion, the company provided residual value guarantees with an initial threshold of about $28 billion. The data center itself is in a separate joint venture where Meta owns 20%.

In its financial statements, Meta noted that as of the end of 2025, payments under these guarantees were not considered probable, so no separate liability was recognized on the balance sheet. This design allows deploying very capital-intensive AI infrastructure without proportional growth of on-balance-sheet debt.

Nvidia is now developing a similar approach. In August, the chipmaker agreed with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR to create separate AI infrastructure financing platforms that are expected to mobilize over $500 billion of third-party capital over time.

Nvidia explicitly states in its quarterly reports that in certain projects it may provide limited residual value support for equipment. Thus, the company helps its customers attract cheaper financing to purchase accelerators and build "AI factories", without financing the entire infrastructure itself.

Financial Times also cites the example of Broadcom, which this year assumed up to $29 billion of potential liability as part of financing computing equipment for Anthropic. Such deals are becoming increasingly important as the scale of AI investments begins to exceed current cash flows of even the largest tech companies.

The main risk of this model will materialize in the event of a sharp slowdown in the AI boom. Companies are effectively betting that data centers and expensive accelerators will retain sufficient market value and demand will remain high for many years. If these assumptions do not materialize, some of the current off-balance-sheet guarantees may turn directly into financial losses.

Based on materials from: Financial Times, Meta's filing with the SEC, Nvidia