With a change at the top, OCBC has sent a clear signal to the market: technology is no longer a cost center but the engine of strategy. According to Bloomberg, Singapore’s second-largest bank will raise annual technology spending to more than $771 million, marking the first major move by Tan Teck Long, who became CEO on January 1, 2026, succeeding Helen Wong.

The figure matters not just for its size, but because it comes as Asian banks rethink their AI architectures. The conversation has moved beyond chatbots: large language models (LLMs) are entering critical areas like credit analysis, fraud detection, regulatory compliance, and personalized assistance. These workloads, by nature, require tight control over data and latency.

The return of directly managed hardware

OCBC’s acceleration is part of a broader movement where financial institutions are carefully weighing how much to keep on-premise. For real-time inference workloads – such as an anti-money-laundering system analyzing transactions with LLMs and large context windows – the round-trip to the cloud can become a cost not only in Total Cost of Ownership but also in compliance terms. The banking sector, squeezed between demanding regulators and the need for data sovereignty, is rediscovering the value of self-hosted infrastructure.

This does not mean a data center full of mainframes. Rather, it points to hybrid architectures where models run on bare metal nodes equipped with high-memory-bandwidth GPUs, with fine-tuning and quantization pipelines executed locally to ensure sensitive data never leaves the corporate perimeter. The number announced by OCBC may translate into an upgrade of internal computing capacity, capable of handling hundreds of concurrent inference requests without relying on external providers.

The Tan factor and the trajectory of banking AI

The appointment of Tan Teck Long, with a background in digital business, suggests the push will go beyond infrastructure. Banks that invest in AI inevitably need to redesign model governance: security, audit trails, versioning. In such a scenario, keeping data in-house is not just a technical choice but a prerequisite to obtain clearance from internal steering committees and regulators.

Those watching the on-premise LLM market are paying attention. If a systemic bank like OCBC confirms a commitment on this scale, the ripple effect on vendor choices and reference architectures for the industry could be significant. AI-RADAR has long tracked the balancing act between cloud and on-premise in regulated environments, noting that the real deciding factor is rarely the unit cost of a GPU, but the ability to govern the full model lifecycle without losing control of data.

Looking ahead, OCBC’s move could be read as a declaration of technological independence from Asian credit institutions. While major cloud providers push turnkey AI services, the path of internally managed hardware and open-source frameworks is becoming a concrete alternative for those who cannot afford gray areas in data residency. A path that, in practice, also runs through investments like the one just announced.