300 Billion Dollars.. What Are the Tech Giants Hiding in the AI Race?

Listen to the article, the audio text is automatically generated by a system.

Major technology companies have used a new financing mechanism that has enabled them to support their massive expansion in artificial intelligence infrastructure by providing guarantees worth up to $300 billion in less than a year, without loading most of these commitments onto their balance sheets.

Companies like Meta, Nvidia, and Broadcom have adopted what is known as "residual value guarantees," which ensure a minimum value for data centers or chips in the future, allowing investors and lenders to finance projects on better terms and at a lower cost.

These guarantees represent the latest financing tools that technology companies have resorted to in order to accelerate the race to build artificial intelligence infrastructure. According to an analysis by the Financial Times, major companies have made commitments of this kind worth up to $300 billion over the past twelve months.

Bankers have described these guarantees as "effective from a balance sheet perspective," as they are typically provided in favor of debts issued by special purpose companies that own the assets, rather than the technology companies themselves. This way, projects benefit from the credit strength of the sponsoring companies without fully recording the obligations on their financial books.

The spread of these financing structures has added a new layer of risk in the event of a massive bet on artificial intelligence failing due to weak demand, excess computing power, or stumbling in developing sustainable business models.

Doug Colandria, a director at the credit rating agency KBRA, stated that last year witnessed a "notable expansion in off-balance-sheet commitments," which complicates the credit risk profiles of those companies.

This trend came as technology companies accelerated their spending on artificial intelligence at a pace that began to exceed the cash flows generated from their operations, prompting them to seek ways to maintain their credit ratings and continue accessing low-cost debt markets.

Morgan Stanley analysts estimated the total off-balance-sheet commitments and credit supports of seven major companies in the data center operators and chip manufacturers sector to be over $3.1 trillion.

Tim Musial, head of fixed income at CIBC Private Wealth US, noted that the market is increasingly concerned about these future commitments amid intensifying competition for the computing capabilities necessary for artificial intelligence.

The guarantees of the remaining value have helped reassure lenders who are facing concerns about the rapid decline in technology value with the evolution of new generations of chips and servers.

Under these arrangements, the sponsoring companies commit to covering any gap between the actual price of the asset when sold in the future and the previously agreed minimum guaranteed value.

This mechanism has led to a reduction in financing costs, as projects supported by remaining value guarantees are priced at a premium ranging from 100 to 150 basis points above the borrowing cost of the guaranteeing company itself.

Broadcom adopted this model during the current year when it took on a $29 billion exposure in a deal to sell 1 gigawatt chips to a special purpose company that will lease them to Anthropic.

The company's quarterly disclosures showed that this credit support has not significantly impacted its balance sheet. It is also expected to use the same mechanism to finance additional shipments of chips over the next year, including 5 gigawatts for Anthropic and 1.3 gigawatts of chips designated for OpenAI.

The company confirmed to investors that the profitability strength of major artificial intelligence laboratories and the sustainable value of core assets make the likelihood of activating these guarantees low.

In the same context, Nvidia revealed last month a similar customer financing program, indicating that it could provide support of up to 25% of the value of some deals that Goldman Sachs and Wall Street investors are working to gather funding for, which could reach $500 billion for infrastructure projects relying on Nvidia technologies.

The company also provided guarantees worth $105 billion in favor of SB Energy, a subsidiary of SoftBank, which is developing a massive data center complex in Ohio to serve OpenAI.

Nvidia will not record any obligation related to these guarantees in its balance sheet before the lease contracts begin in 2028, while the deal will secure the company the operation of the site using its hardware exclusively for 20 years.

Meta preceded its competitors in using this mechanism when it provided a residual value guarantee of $28 billion in support of its joint project with Blue Owl to develop the 'Hypron' data center with a capacity of 2 gigawatts in Louisiana.

The project spans an area of 4 million square feet and is expected to be operational by 2030, with electricity consumption sufficient to power approximately 1.5 million homes.

These guarantees helped the project raise $27 billion in debt at a cost close to Meta's own borrowing costs, with extremely limited exposure recorded on the company's books. The group also used a similar structure in July to support another data center with a capacity of 1 gigawatt in El Paso.

Although these guarantees typically remain off-balance-sheet in most cases, credit rating agencies do not completely ignore them.

S&P analysts added that their ratings take into account the difference between the guaranteed asset value and their estimates of its value if sold under market pressures.

Pierre George, Head of Infrastructure Research at S&P Global, explained that the largest gap observed between the expected asset value and the associated liabilities was about 25%, which could add billions of dollars to the guarantor's debt when calculating adjusted leverage.

He pointed out that S&P's assessment of the "Hyperon" project belonging to Meta included a sharp reduction in the asset's value, yet the final result showed that the project's value is still higher than the amount of supported debt, which spared Meta any increase in leverage indicators.

He stated that the gap between the expected actual value of assets and the guaranteed values "is approaching zero in most current deals," which explains the continued interest of technology companies and lenders in this financing model.