‘Highly selective’: Singapore banks see strong AI financing demand but stay alert to debt risks

[SINGAPORE] Singapore banks remain keen to finance South-east Asia’s booming data centre sector, but are taking a selective approach amid growing concerns over debt-funded artificial intelligence infrastructure spending.

Industry watchers noted that while established operators with strong financial backing and long-term customer contracts continue to attract funding, smaller developers face tightening liquidity conditions.

The cautious lending stance comes against the backdrop of a sharp sell-off on Wall Street on Thursday (Nov 5) led by AI-linked stocks, as midterm election results led to fears that the data centre boom could stall.

According to Singapore’s three banks, AI-related financing accounted for less than 1 per cent of OCBC’s loan book, compared with less than 5 per cent at UOB. DBS, meanwhile, said its AI financing was primarily concentrated in data centres, which accounted for about 2 per cent of its loan book.

At their latest Q3 results briefings, all three emphasised prudent lending standards, focusing on financially strong borrowers and reliable cash flows.

“Lenders are increasingly focused on bankability and downside protection,” observed Emmanuel Hadjidakis, principal, Baker McKenzie Wong & Leow, and Asia-Pacific chair of Banking & Finance, Baker McKenzie.

He noted that lenders in South-east Asia had already begun tightening loan terms and conditions before the recent market volatility.

John Hall, founder and managing partner of Brookland Group & Partners, cautioned that AI infrastructure remains a relatively unproven asset class despite its rapid growth.

“Many projects depend on future revenue that has not yet materialised, and investors are now pricing that risk more carefully,” he noted.

Hall pointed to signs of greater investor caution, including several stalled regional listings – an Australian data centre operator pulling its initial public offering and a planned Singapore data centre Reit listing pushed to next year.

As equity capital becomes harder to secure, debt funding could also become more challenging, although the impact varies across borrowers, he said.

Banks are still backing established operators with major cloud computing companies as customers, while smaller and newer players will find lenders asking for more equity, stronger customer commitments and proof of secured power before they commit, he added.

Banks remain positive on AI financing

Singapore’s major banks continue to see AI infrastructure as a growth opportunity even as they keep an eye out for risks.

At OCBC’s Q3 earnings briefing, group CEO Tan Teck Long said he expects the order book to be “pretty strong” over the next two years, supported by resilient demand for data centres and AI infrastructure amid bottlenecks in the supply chain.

Elaine Lam, head of global corporate banking at OCBC, noted that there may be periodic market adjustments and cyclical swings given the pace of growth and heightened liquidity demand, but long-term fundamentals and demand outlook remain constructive.

OCBC does not view data centres as a single asset class, instead assessing projects individually based on factors such as demand visibility, location, access to power and connectivity, regulatory approvals and sponsor quality, she said.

“When we finance a data centre project…we require a data centre project to secure the hyperscaler to anchor a large part of the capacity,” said OCBC’s Tan.

He added that the bank has a “prudent credit underwriting policy” where it finances only deals backed by cash flows rather than speculative builds.

Similarly, DBS chief executive Tan Su Shan said that the bank has been “highly conservative and structured” in its approach to data centres financing despite strong growth.

It does not finance speculative developments, focusing instead on projects backed by investment-grade hyperscalers.

She added that DBS leads and arranges financing for many data centre projects, but does not necessarily retain a large portion of these loans on its balance sheet.

For some US projects, DBS also has arrangements under which hyperscalers bear increases in operating costs, helping to protect project cash flows. The bank also said that it was “quite comfortable” with its data centre financing exposures, which it described as “structured very well”.

DBS’ Amit Sinha, global head of telecommunications, media and technology, said that for larger projects, syndicated lending remains important, as it allows banks to manage concentration risk and balance-sheet capacity.

He added that the financing ecosystem is broadening beyond banks, which will remain central to greenfield and brownfield projects, while mature, income-generating assets can tap capital markets, so sponsors can recycle capital.

Private equity and private credit can complement these structures where different risk-return profiles are needed, said Sinha.

Meanwhile, Leong Yung Chee, group chief financial officer, UOB, noted that there have been signs of stress in parts of the broader AI financing market, including movements in credit default swaps and bond yields for lower-rated borrowers.

As for the bank, it applies rigorous credit processes and prudent underwriting standards, focusing on borrowers, counterparties and projects with strong fundamentals and sustainable cash flows, he said. He added that there are currently no non-performing loans in UOB’s AI-related financing portfolio.

“The sector is among the fastest-growing areas across industries and presents significant opportunities,” Leong said. “At the same time, we are mindful of developments across the broader AI financing market and remain highly selective in the opportunities we pursue.”

GPU financing faces tougher scrutiny

In particular, loans backed by graphics processing units (GPUs) – the high-performance chips powering AI workloads – are drawing greater attention due to rapid technology obsolescence and steep depreciation, said industry watchers.

Baker McKenzie’s Hadjidakis said that for GPUs, lenders want much tighter control over the collateral, including where it is located, if it can be moved or replaced, and what happens as its value declines.

“Access and redeployment rights are particularly important because security is only useful if the lender can actually realise the asset,” he said, adding that the challenge for lenders is to balance lender protection with the operational flexibility these businesses need.

“Lenders look at model, age, depreciation and secondary-market value, but also whether the equipment can generate cash flow or be redeployed,” noted Hadjidakis.

Borrowers are responding with structures that allow substitution, upgrades and relocation, provided agreed collateral-value and eligibility tests continue to be met, he added.

Looking ahead, Brookland’s Hall said the widening gap between big and small players creates room for more flexible capital structures, which could help developers secure funding where conventional lenders require more safeguards.

“Structured and hybrid financing, combining debt with equity-linked features, can bridge projects through this period while traditional lenders remain selective,” said Hall.