DBS chief: market is reassessing AI investment risk and will tighten financing

DBS Group Holdings CEO Tan Su Shan said the market is reassessing the risks of investments tied to the AI infrastructure buildout. In her view, this may lead to stricter discipline in financing. The statement gives no specific amounts, project names or timelines. But it comes from the head of Southeast Asia's largest bank, and that changes how much weight it carries.

Who is speaking and why it matters

Tan has led DBS since 2025, when she succeeded Piyush Gupta. The bank works with corporate borrowers across the region, while Singapore, Malaysia and Indonesia compete for data center sites. When a lender like this talks about repricing risk, it is describing what it sees in loan applications, rates and deal terms. This is a view from inside the credit process, not an outside forecast.

The wording is also telling. Tan does not talk about a bubble and does not predict a crash. She talks about repricing: the market is changing the price of risk, not abandoning the theme. For a banker, the difference is significant. Money for AI projects remains available, but it costs more and comes with tougher requirements.

Where the pressure on financing comes from

Over the past couple of years, the AI buildout has been funded less and less from the free cash flow of tech giants. In 2025, large debt-based financing schemes appeared. Oracle issued bonds worth about $18 billion, and Meta raised about $27 billion for its Hyperion campus through a structure involving Blue Owl. Debt loads are rising, and the payback period of compute capacity remains a matter of debate: chips become obsolete faster than buildings, and demand for AI services does not yet fully cover the investment.

In this setup, banks look at several things:

• who the end tenant of the capacity is and how reliable it is;
• how long the contracts run and whether they cover the term of the loan;
• what share of the project's cost goes to fast-obsolescing hardware;
• how dependent the borrower is on a single customer or a single chip supplier;
• how much headroom the project has on power, grid connection and permits.

These are not new questions. What is new, by Tan's assessment, is that the market has started answering them more strictly than before.

What is known and what is not

Three things follow from the statement:

• the market is revising its assessment of AI investment risks;
• the result may be more disciplined financing;
• the signal comes from a bank that lends against real infrastructure, not one that trades on expectations.

It is not known how large the tightening is, whether it will affect data center construction, chip supply or venture rounds for AI labs, or how much it will show up in rates. This information is absent from the available text, and it is better not to guess at it.

Practical takeaways

Data center operators and site developers would be wise to prepare a more detailed utilization model in advance: with preliminary lease contracts, a power connection schedule and a scenario in which compute prices fall. A bank that demands more data generally treats more favorably those who bring it before being asked.

AI startups whose business model relies on cheap access to compute should check the terms on which they depend on cloud providers, and whether their plans assume a generosity of credit that may come to an end. Investors would do well to distinguish projects with confirmed demand from projects that rest on expected growth.

Watch for specifics: the terms of new syndicated loans, spreads on AI issuers' bonds, and how collateral requirements change. For now, Tan has pointed to a direction. Only real deals in the coming quarters will confirm it or not.