Surging demand for AI agents fuels Hong Kong’s race for computing power

Hong Kong scrambles to add GPU capacity as AI agents proliferate

A new wave of AI agents — autonomous, always-on models that can plan, act and interact for users — is driving a sudden spike in demand for high-performance computing. Hong Kong, long a regional financial and connectivity hub, is racing to expand data-centre capacity and GPU clusters to serve local start-ups, financial institutions and multinational firms that want low-latency access to powerful models. Short bursts of intense compute are turning into steady, hungry workloads. Who will host that power matters.

Local telcos and data-centre operators are the obvious players. It has been reported that firms such as PCCW (電訊盈科) and HKT (香港電訊) are expanding infrastructure, while mainland cloud providers including Alibaba (阿里巴巴) and Tencent (騰訊) are also pitching services to Hong Kong customers who want closer ties to mainland datasets and markets. Building racks of Nvidia GPUs and the associated cooling and power systems is costly. Land, electricity and stringent planning rules are immediate bottlenecks. Investors and operators are rethinking real estate in the city just as demand for rack space intensifies.

Geopolitics complicates the technical push. US export controls on advanced AI chips and broader tech decoupling between Washington and Beijing have reshaped supply chains, and it has been reported that some firms are re-evaluating where to place sensitive workloads to navigate sanctions and data-governance regimes. Can Hong Kong balance its role as an international gateway with regulatory sensitivity to both the US and mainland China? The answer will shape where compute-heavy AI development actually happens in Asia.

What comes next

For startups and banks alike, latency and jurisdiction will drive choices as much as price. If Hong Kong can unlock more power and space quickly, it could become a preferred hub for AI agents serving Greater China and global financial services. If not, compute may shift to the mainland, Singapore or edge sites closer to users. Either way, the race for GPUs highlights a simple reality: AI isn’t just about models and data — it’s about where you put the electricity.