Vikram Sinha, Zankore’s chairman, told a conference on 22 September that the company would have to keep raising money as it expands its artificial intelligence data-centre capacity by 10 times to 1 gigawatt.
Zankore has already closed a $3.1 billion senior term loan from a five-bank consortium in Indonesia, for Nvidia graphics processing units for a new AI factory. Citigroup Inc. acted as the sole debt adviser on the borrowing.
United Overseas Bank Ltd. jointly underwrote the loan with four other banks and is leading discussions as Zankore seeks a fresh $6 billion from banks. The outcome of that fundraising remains unclear.
The deal is part of a broader move by banks into lending backed by graphics processing units, or GPUs, the specialised chips used in AI infrastructure. Banks have played key roles in loans totalling roughly $3.8 billion to GMI Cloud, Zankore and PaleBlueDot AI in recent months.
GMI Cloud is seeking a $635 million multi-tranche loan based on expected cash flow from customers that have contracts to rent chip time. It is also in talks with banks and private lenders over a new $300 million loan to buy chips for its Thai data centre.
PaleBlueDot AI was reported to be seeking roughly $300 million to help a Chinese client buy Nvidia chips for use in Japanese data centres. JPMorgan Chase & Co. was the placement agent for its credit facility.
Private credit funds initially led GPU lending in Asia while banks were still becoming comfortable with the asset class. Global investment banks have since moved in, drawing on their experience with complex deal structures. Citigroup, JPMorgan Chase, Barclays Plc, Deutsche Bank AG, Banco Santander SA and Japan’s Sumitomo Mitsui Banking Corp. are evaluating GPU-linked loans.
The potential prize is large. PricewaterhouseCoopers LLP estimates that spending on Asian data centres could reach $8.2 trillion by 2050.
For lenders, however, the collateral presents a difficult valuation problem. These loans are generally being written with three-to-five-year payback windows. Eric Tan, a banking and finance partner at Hogan Lovells Cadwalader, said GPU financing leaves lenders exposed to rapid depreciation, technological obsolescence and volatile rental rates.
The longer-term value of the chips, geopolitical tensions and risks tied to AI itself are all concerns for lenders. US export restrictions and regulatory compliance requirements add further complexity to the deals.
Michael Burry has argued that hyperscalers are assigning chips useful lives of five to six years even though they will be economically obsolete in two to three years. He estimated industry-wide depreciation understatement at $176 billion from 2026 through 2028.
Banks are responding with conservative underwriting, stronger debt-service reserves and more rigorous checks on borrowers’ revenue. Warnings have mounted that the region’s data-centre borrowing is already testing banks’ exposure limits.
Recent deal jitters have also shown how rapidly lenders’ appetite can fade when risk cannot be syndicated smoothly. The central issue is whether banks can distribute these loans widely enough to avoid concentrating the risk on their own balance sheets.
Comments (0)