Major technology firms have issued guarantees totaling as much as $300 billion over the past year to fund data centers dedicated to artificial intelligence and accelerators. A significant portion of this associated debt is recorded on the balance sheets of special purpose vehicles and does not appear as standard corporate debt for these tech giants, according to the Financial Times (FT).

Companies like Meta, Alphabet, Nvidia, and Broadcom are utilizing such structures to secure financing at the credit quality level of tech behemoths without directly borrowing equivalent amounts.

Typically, a separate entity (SPV) is established for a project, which raises debt and owns the data center or equipment. The tech company then either leases the capacity or backs the deal with a residual value guarantee. Should the asset need to be sold or leased at a lower rate than agreed, the guarantee covers the specified difference as per the contract.

This credit support from tech giants allows borrowers to secure funding at lower costs. FT reports that interest rates for these projects generally exceed the borrowing costs of the guarantor company by only 1 to 1.5 percentage points.

A summary chart of guarantees provided by tech companies for AI infrastructure funding. Source: FT.

One clear example is Alphabet, which saw its maximum obligations related to credit derivatives for data centers surge from $16.9 billion to $43.8 billion over six months, as noted in the company’s quarterly report here.

The fair value of these obligations recorded on its balance sheet was $815 million, alongside $7.6 billion in financial guarantees. Furthermore, Alphabet has entered an agreement to provide an additional $24.1 billion in guarantees for the construction of data centers and energy infrastructure in the future.

In the event of a counterparty default, Alphabet has the option to lease the relevant data centers for its own use, transfer them to third parties, or in certain cases, terminate the obligation through compensation payments.

Nvidia and OpenAI

In August, Nvidia issued guarantees amounting to up to $105 billion for SB Energy related to data centers in Ohio with a capacity of approximately 4.25 GW, which will be leased by OpenAI, phased in as the facilities become operational.

Nvidia may need to fulfill these guarantees in the event of non-payments or bankruptcy by the ChatGPT developer. In such a scenario, the company could take over the lease or require the asset to be re-leased or sold. The potential exposure will gradually decrease as Sam Altman's firm meets its payment obligations.

In July, reports emerged about another discussed credit support scheme by Nvidia for OpenAI's infrastructure.

Meta and Broadcom

Meta has provided residual value guarantees for the Hyperion data center, with a cumulative threshold of around $28 billion that declines over time.

According to a report, if Meta ceases or does not renew the lease and other conditions are met, its maximum payout would be the difference between the fair value of the asset and the set guarantee threshold. By the end of 2025, the organization deemed such payouts unlikely.

FT reports that these guarantees have helped raise about $27 billion in debt for Hyperion at a rate less than 1.5 percentage points above Meta's bond yields.

Broadcom employs a similar model for AI equipment. In June, the company launched the AI XPV platform with an initial funding tranche of $35 billion.

A financial partner acquires racks based on Broadcom's AI accelerators and provides computational power to clients. Broadcom's maximum potential liability under the lease support agreement is approximately $29 billion.

According to FT, this scheme is utilized for supplying equipment to Anthropic.

Off-Balance Sheet Does Not Mean No Risk

In a broader assessment, analysts at Morgan Stanley have identified over $3.1 trillion in off-balance sheet obligations and credit support among seven hyperscalers and AI chip manufacturers. This total includes not only residual value guarantees but also other future commitments.

Off-balance sheet and contingent liabilities of major tech companies, in $ billion. Source: Morgan Stanley Research, Financial Times.

Doug Kolandrea, a senior director at the rating agency KBRA, noted that the off-balance sheet exposure of tech companies has significantly increased over the past year.

"This significantly complicates the credit risk profiles of these companies," he stated.

The economic risks associated with such schemes do not disappear. If the value of data centers or accelerators falls below the contractually stipulated levels or if tenants fail to meet obligations, some losses may shift to the guaranteeing party.

Rating agencies take such structures into account when assessing debt loads. For instance, S&P may add to the calculated debt the difference between the guaranteed level and its valuation of the asset under stress scenarios.

However, in most of the discussed transactions, such adjustments have so far been minimal. In the case of Hyperion, S&P applied a significant discount to the value of the data center but still valued the asset above the secured debt, hence did not increase Meta's calculated leverage.

It is worth noting that in September, reports surfaced regarding OpenAI's acquisition of thousands of Mac mini and Mac Studio computers for specific agency tasks.