JPMorgan Chase is marketing a $5 billion leveraged-loan package for Volta Infrastructure Holdings, highlighting the increasingly expensive financing environment facing smaller companies participating in the artificial intelligence infrastructure boom.

The loan is being offered at a spread of roughly 6.25 to 6.5 percentage points above the benchmark rate, combined with an issue price of 97 to 98 cents on the dollar. At current benchmark levels, that structure equates to an estimated yield of approximately 11%.

The proposed financing is notable because of the scale of the borrowing and the cost attached to it. Earlier AI infrastructure transactions involving other neocloud companies were completed at lower borrowing costs.

AI Infrastructure Is Creating Massive Capital Needs

The cost of purchasing the specialized computing equipment required to run AI models has created an enormous financing requirement across the technology industry.

Investment-grade companies with strong corporate partners can access lower-cost debt markets. The source points to a $35 billion debt package backed by Broadcom that will help Anthropic obtain access to AI chips.

Smaller companies face a different financing environment. Neocloud operators that rent access to GPUs and other AI infrastructure have increasingly turned to leveraged loans to fund equipment purchases and data center expansion.

Deals involving CoreWeave, Lambda and Crusoe have demonstrated lender appetite for this type of exposure, although Volta’s proposed transaction carries a higher all-in yield.

$5 Billion Package Combines Chips and Data Center Financing

The Volta transaction is structured through two debt tranches borrowed by a special corporate entity called Volta Tydal Holdings S.à r.l.

Approximately $3.7 billion will be used to purchase around 36,000 Nvidia GPUs. Another $1.3 billion will cash-collateralize a letter of credit connected to a data center lease.

The facility is located in Norway and will be managed by Volta Infrastructure in partnership with Bitdeer Technologies Group. Anthropic has already signed a six-year agreement for computing capacity at the site, according to people familiar with the transaction.

The structure therefore combines equipment financing with infrastructure financing rather than relying solely on debt to acquire AI chips.

Unrated Debt Creates Another Layer of Complexity

The transaction is reportedly unrated, an unusual characteristic in the leveraged-loan market. The absence of a rating can restrict participation by collateralized loan obligations, which represent some of the largest pools of demand for this type of debt.

The deal is also structured as a “strip,” meaning lenders must purchase portions of both loan tranches to participate. The debt will fully amortize over time, another feature that has become common in AI infrastructure financing but is less typical of corporate debt, where principal is often repaid at maturity.

These characteristics could affect the investor base and the way lenders assess the risk and return profile of the transaction.

AI Debt Markets Continue to Expand

Wall Street is increasingly financing the physical infrastructure required to support the expansion of artificial intelligence. The source estimates that companies have borrowed approximately $625 billion over the past two years for data centers and other AI investments, while the broader AI buildout could require nearly $5 trillion by the end of 2030.

The scale of borrowing illustrates how quickly AI has moved from a technology investment theme into a major capital-intensive infrastructure cycle.

For smaller companies, however, the cost of that capital can become a significant part of the business model. Higher financing costs increase the amount of revenue required from computing capacity simply to service debt obligations.

What Investors Should Watch Next

Volta’s proposed financing provides an important window into the risk and return dynamics developing around AI infrastructure debt. The approximately 11% yield reflects strong investor demand for AI exposure but also the higher borrowing costs facing companies without investment-grade balance sheets or large corporate backers.

The next test will be whether the revenue generated from AI computing capacity can support these financing structures over time. The performance of neocloud companies, GPU utilization, long-term customer contracts and data center economics will be critical variables.

With commitments for the Volta loan due October 14, investor appetite for the transaction could provide another signal of how financial markets are pricing the rapidly expanding AI infrastructure cycle.