JPMorgan and Goldman Sachs are taking a cautious approach to a $5 billion debt package for AI-focused Volta Infrastructure, according to Bloomberg. The banks are underwriting a $1.3 billion term loan but are only marketing a larger $3.7 billion facility on a best-efforts basis. That larger facility is tied to roughly 36,000 Nvidia GPUs, highlighting a key distinction in how lenders view different types of AI-related risk.
Underwriting means the banks commit to fund the loan even if investors don’t buy it. Best-efforts means they will try to place the debt with investors but won’t guarantee the full amount. The split shows that banks are willing to put their own balance sheets behind cash flows tied to a data center lease in Norway, run with Bitdeer Technologies, but are less eager to warehouse risk linked directly to GPU hardware.
Why the split matters
The $1.3 billion term loan is backed by predictable cash flows from a data center lease. That kind of revenue is easier to model and can be used as collateral, making it more attractive to lenders. The $3.7 billion facility, by contrast, is tied to GPU hardware, which can lose value quickly as newer chips arrive. Resale prices for GPUs in a downturn are hard to predict, so banks are reluctant to take on that risk even though the overall package is said to offer around an 11% yield.
This is a window into how AI-linked leveraged loans may evolve. When banks won’t underwrite the hardware-heavy part of a deal, more of the risk shifts to credit investors upfront. That usually means higher yields, tighter lender protections, or both. Over time, that could steer leveraged-loan funds and collateralized loan obligations (CLOs) toward data-center and contract-backed financings, while “chip loans” either price with a bigger premium or end up in more specialized pockets of the private credit market.
The move comes as banks and investors grapple with how to finance the AI boom without taking on excessive risk. Data centers and the chips that power them require massive upfront investment, and the financing structures are still evolving. JPMorgan and Goldman’s decision to limit their exposure to the GPU-backed portion suggests they see a clear line between infrastructure with stable cash flows and hardware that can become obsolete.
What it means for investors
For everyday investors, this story is a reminder that not all AI-related debt is created equal. If you own leveraged loan funds, CLOs, or other credit products, you may have indirect exposure to AI infrastructure financing. The way banks structure these deals can affect the yields and risks in those funds. A shift toward data-center and contract-backed loans could mean more stable, lower-yielding assets, while GPU-backed loans might offer higher yields but with more uncertainty.
It also signals that banks are becoming more selective about AI risk. That could slow down some projects or push them into private credit markets, where investors are willing to take on more risk for higher returns. For investors in private credit funds, that could mean more opportunities but also more complexity. As always, it’s important to understand what’s backing any debt you own.
The broader market backdrop includes rising interest rates and concerns about an economic slowdown, which make lenders more cautious in general. AI has been a rare bright spot for investment, but even there, banks are drawing lines. This deal shows that the AI financing boom is not a free-for-all; it’s being underwritten with an eye on what can go wrong.
Investors will be watching how this loan performs and whether other banks follow suit. If GPU-backed loans become harder to place, it could reshape how AI companies fund their expansion. For now, the message is clear: cash flows from leases are easier to finance than the chips themselves.
