DBS CEO Says AI Costs Are Falling as Usage Grows

There has been a quiet assumption in banking that generative AI would be powerful but expensive — a capability worth having, yet one that would pressure technology budgets as usage scaled. DBS chief executive Tan Su Shan is now pushing back on that assumption with data from inside one of Southeast Asia's largest banks: the more DBS uses AI, the less it costs per unit of work.

Tan told Bloomberg that heavier AI usage has been accompanied by falling costs per token as the bank improves how it deploys the technology. The strategy is essentially routing: simpler requests are handled by smaller language models, DBS tests models from several providers rather than depending heavily on any single one, and caching avoids reprocessing identical queries. Employees, Tan said, are also getting better at optimising how the technology gets used. None of this has blown up the budget — overall technology spending remains around 10 per cent of revenue and broadly stable.

This is the economics of AI that most enterprises are still figuring out. The instinct at many companies has been to standardise on one frontier model and treat token costs as a fixed line item. DBS's approach suggests the opposite: a portfolio of models matched to task complexity, with caching and careful prompt design, can bend the cost curve downward even as adoption spreads. It is a form of discipline that banks, with their compliance-heavy, high-volume workloads, are particularly well placed to practise.

The timing matters. DBS just reported a record second-quarter net profit of S$3.08 billion and raised its full-year outlook, which gives it room to keep experimenting. For the rest of Singapore's financial sector — and the enterprises watching DBS as a bellwether for AI adoption — the message is that scaling AI does not have to mean scaling costs, provided the underlying architecture is deliberate about model choice.

Why it matters for Singapore: DBS is the flagship test case for AI at scale in the city-state, and its cost experience is being watched well beyond banking. As MAS finalises guidelines covering agentic AI and Singapore pushes its AI ambitions further, evidence that AI can be both pervasive and affordable strengthens the case for businesses here to move faster — rather than waiting for costs to fall on their own.