Wall Street Banks Move into GPU-Backed Lending Across Asia
TREE NEWS reports: International banks are quietly entering the GPU financing market in Asia, a space previously dominated by private credit funds willing to take on higher risk. In recent months, major global lenders have played key roles in approximately $3.8 billion in GPU loans for three AI infrastructure providers: GMI Cloud, Zankore, and PaleBlueDot AI. Citigroup, JPMorgan, Barclays, Deutsche Bank, Santander, and Japan’s Sumitomo Mitsui Banking Corporation are all currently evaluating GPU-linked loans.
The capital is critical. PwC estimates that Asia’s data center spending could reach $8.2 trillion by 2050, with the vast majority flowing into GPUs and servers. Hundreds of data centers are under construction across the region, and developers are simultaneously raising funds to purchase chips.
Global Banks Lead, Asian Lenders Follow
Citigroup served as the sole debt advisor for Zankore’s $3.1 billion borrowing in Indonesia, while JPMorgan acted as placement agent for PaleBlueDot AI’s $255 million credit facility. About six bankers and financial advisors in Asia say they are in talks or aware of more GPU-related financings. GMI Cloud is negotiating a new $3 billion loan with banks and private lenders to procure chips for a Thailand data center.
Among traditional lenders already involved, global investment banks are currently the main force, leveraging their experience in handling complex structures. Some U.S. banks have deployed headquarters experts to assess chip value. Citigroup, Barclays, and Deutsche Bank declined to comment, while JPMorgan, Santander, and SMBC did not immediately respond to requests for comment.
Asian local banks are also becoming more active. United Overseas Bank of Singapore, which co-underwrote Zankore’s $3.1 billion loan with four other banks, is leading negotiations for its next round of financing. Zankore Chairman Vikram Sinha said at a conference on September 22 that the company aims to expand its AI data center capacity tenfold to 1 gigawatt, requiring continuous financing. “We are very clear about the scale we want. We want to take the difficult path, working with banks and syndicates,” Sinha said.
Chip Valuation Remains a Challenge
Eric Tan, a partner in the banking and finance practice at Hogan Lovells Cadwalader, said: “As deal sizes grow and borrowers demand more competitive pricing, banks will play an increasingly important role.” However, rapid technological iteration exposes lenders to risks of “rapid depreciation, technological obsolescence, and rental volatility.”
As banks become a more important source of funding, borrowers will need to more fully demonstrate that project revenues are sufficient to repay loans. Traditional lenders may require more conservative underwriting standards and higher debt service reserves.
Private credit is not lowering its bar either. Mike Arougheti, head of Ares Management, one of Asia’s largest private credit institutions, said that although GPU financing is the largest funding gap in the AI boom, Ares maintains high standards. “You have to prioritize risk appetite, not the willingness to deploy capital,” he said at the Barclays Global Financial Services Conference in September. “At least no one can explain to me clearly what the depreciation curve for this technology looks like.” He also noted that returns on such loans are limited, typically only about 100 to 200 basis points higher than other AI infrastructure loans.
Repayment Tied to Customers
Most Asian GPU loans follow the structure pioneered by CoreWeave, one of the earliest and largest users of such financing. In many deals, loans are repaid through revenue from selling data center computing power, with customer contracts and the chips themselves typically serving as collateral. For lenders, the most critical questions are therefore whether the customer is reliable and how long the contract term is.
Deals backed by Nvidia are the easiest to approve. In the GMI Cloud Taiwan project loan completed last month (provided by 14 banks) and the Zankore loan, Nvidia agreed to purchase all unsold computing power, providing a backstop if customer contracts fall through. In exchange, these two Nvidia Cloud Partner program members will charge buyers higher prices than Nvidia’s committed rates and share revenue with Nvidia.
The debate over AI’s potential dangers and whether guardrails are needed will also continue to influence risk assessments. “It may take some time for the market to find equilibrium. Deal structures may tighten, sovereign support may intervene, and transaction models will evolve accordingly,” Tan said.
Key Takeaways for Investors
- GPU financing is becoming mainstream: The entry of global banks signals growing institutional confidence in AI infrastructure as an asset class, potentially lowering borrowing costs and accelerating data center buildout across Asia.
- Depreciation risk is real: Rapid GPU obsolescence remains a key concern. Investors should monitor how lenders structure reserves and covenants to mitigate this risk.
- Nvidia’s backstop matters: Deals with Nvidia’s guarantee are easier to finance, suggesting the chipmaker’s role extends beyond hardware supply into financial risk absorption.
- Watch for sovereign involvement: As deal sizes grow, sovereign support or government-backed guarantees could become more common, altering the risk-return profile for lenders and investors.
