Wall Street's stance on AI financing is undergoing a subtle shift: banks remain willing to participate in deals and collect fees, but they are becoming more cautious about taking on risk. JPMorgan and Goldman Sachs jointly led a $5 billion AI infrastructure loan but refused to provide underwriting guarantees for the $3.7 billion GPU chip portion.

According to Bloomberg, the financing for emerging cloud computing company Volta Infrastructure Holdings is divided into two parts: a $1.3 billion term loan underwritten by the banks, and a $3.7 billion portion structured on a "best efforts" basis, meaning the banks do not commit to purchasing any unsold portion.

The financing cost of this deal is equally noteworthy. The two loans carry yields to maturity of 11.21% to 11.75%, placing them at the high end of the current AI financing market. By comparison, a CoreWeave GPU loan in July of this year had a yield to maturity of approximately 10.44%.

Banks have not stopped providing financing for AI infrastructure, but they are no longer willing to unconditionally absorb the risk of chip loans. When high-cost financing meets a more cautious underwriting approach, the market faces a question: if banks are unwilling to backstop these loans, who will take them on? Whether private credit and other sources of capital can fill the gap will also test the true demand and risk tolerance for AI infrastructure financing.

Deal Structure: Banks Only Guarantee the $1.3 Billion Loan

According to PitchBook LCD data, Volta's total financing amounts to $5.012 billion with a six-year term. Of that, a $1.3 billion senior secured term loan C is secured by cash collateral providing a letter of credit for a data center lease; the $3.712 billion GPU term loan is used to purchase approximately 36,000 Nvidia GPUs.

Both loans carry a spread of SOFR plus 625 to 650 basis points, were issued at 97% to 98% of face value, and neither received a credit rating. The project will build a data center complex in Norway, operated by Bitcoin mining operator Bitdeer. According to earlier Bloomberg reports, Anthropic has signed a six-year computing power agreement with the site.

Volta was founded this year, with investors including a16z, Altimeter, Michael Dell, and Nvidia. Nvidia is both a shareholder and the GPU supplier.

The choice made by the two banks essentially separates the financing arrangement from the ultimate assumption of risk. For the $1.3 billion lease loan, the banks provide underwriting guarantees; for the $3.712 billion GPU loan, the banks are only responsible for best-efforts sales and do not commit to taking on any unsold portion. This means whether the chip loan can be successfully financed will depend more on the willingness of external investors to subscribe.

11% Financing Cost: Investors Demand Higher Risk Compensation

Volta's loan carries a yield to maturity of up to 11.75%, higher than some previous comparable GPU financings.

In March of this year, CoreWeave completed an $8.5 billion DDTL 4.0 financing, becoming the first GPU-backed financing to receive a Moody's A3 investment-grade rating, with the floating-rate portion carrying a spread of SOFR plus 225 basis points. By July, another $2.6 billion loan from the company saw its spread raised from SOFR plus 425 to 450 basis points to 550 basis points during roadshow, with a yield to maturity of approximately 10.44%. Volta's loan yield to maturity has climbed even higher, reflecting investors' more cautious pricing of financing risk.

The broader credit market is also under pressure. Goldman Sachs credit strategists including Amanda Lynam estimated in an October 9 analysis that the U.S. leveraged loan default rate will remain elevated at around 5% in 2027, with a recovery rate of approximately 56 cents per dollar, down from 64 cents in 2022.

For Volta, the higher financing cost not only means a heavier interest burden but also reflects investors' concerns about the borrower's creditworthiness and the value of the GPU collateral. While GPUs can serve as loan collateral, their value still depends on market demand, technological iteration, and secondary market liquidity. Combined with the fact that the loans are unrated, the pool of potential investors may also be constrained.

Risk Transfer: Can Private Credit Step In?

Banks tightening their underwriting commitments does not mean AI financing demand has disappeared. Private credit institutions and insurance companies are still participating in related projects, but whether they can absorb the risk that banks are unwilling to hold at a cost borrowers can afford remains to be seen.

According to IFR, a $42 billion senior secured loan backed by Broadcom and aimed at Anthropic launched its syndication sale last Monday. Institutions including Apollo, Blackstone, and Japanese life insurers are also providing funding for AI infrastructure projects.

However, the shift of risk from banks to private credit does not mean the risk has disappeared. The Bank of England has warned that banks may still indirectly bear related risks by lending to private credit funds. JPMorgan Chief Financial Officer Jeremy Barnum also stated that the bank's exposure to the AI theme has spread across multiple areas including construction loans, equity markets, and working capital revolving credit facilities.

Market views on whether high-cost financing is sustainable are not unanimous. Goldman Sachs' Rich Privorotsky, citing Broadcom CEO Hock Tan's view, said that computing power demand still exceeds supply, and frontier AI labs capable of generating substantial revenue have the ability to bear high-cost financing. But if computing power demand or revenue growth falls short of expectations, high interest will increase debt servicing pressure, and GPU collateral may not fully cover loan losses.

Banks Continue to Participate but Are Unwilling to Bear All the Risk

The approach taken by JPMorgan and Goldman Sachs is not an isolated case. According to earlier Bloomberg reports, Morgan Stanley transferred some data center risk exposure through significant risk transfer (SRT) instruments; Blue Owl previously failed to raise third-party funding for a $4 billion CoreWeave data center project.

These cases point to the same issue: the bottleneck in AI infrastructure financing is shifting from capital demand itself to who bears the risk and how much return they should receive. Banks can continue to organize deals, and private credit can provide funding, but financing can only truly materialize if investors are willing to accept the corresponding risk and pricing.

The commitment deadline for Volta's GPU loan is 5:00 p.m. Eastern Time on October 14. The final subscription outcome will serve as a window into market risk appetite: if the loan needs to raise its spread or reduce its size, it would indicate limited willingness among investors to take on such high-risk financing; if it is successfully completed, it would mean the market still has capital willing to bear the related risk, but the price may be higher financing costs.

For AI infrastructure companies, strong computing power demand does not equate to controllable financing risk. When banks are no longer willing to backstop chip loans, the question ultimately comes back to one thing: can the cash flow generated by the project in the future support such high financing costs?