Hong Kong's AI Push Collides With a Structural Energy Deficit

A city ranked third globally as an AI financial powerhouse, according to South China Morning Post opinion writers Ying Xu and Weishi Zhang, has been running an electricity deficit since 1994. That gap, now exceeding 1,500 kilowatt-hours per capita annually and plugged partly by imports from the Daya Bay nuclear facility across the border, is the uncomfortable baseline against which Hong Kong is approving large new AI infrastructure: an AI Supercomputing Centre at Cyberport and a 10-hectare Sandy Ridge data facility cluster under construction in the Northern Metropolis, both approved without adequate long-term energy planning, the authors contend.

The core argument Xu and Zhang make is direct: "the artificial intelligence competition is by nature an energy competition." Compute-heavy AI workloads draw real watts, and a city that has needed to import electricity for three decades is poorly positioned to absorb a wave of GPU clusters without consequences that ripple into public energy supply. The risk is not theoretical. Approved projects are already in motion, and the energy math behind them has not been publicly reconciled.

The proposed fix is a division of labor. Hong Kong should lean into what its regulatory environment and financial ecosystem already support: high-value, lower-energy AI applications in financial technology and regulatory technology. The computationally intensive work should be routed to Greater Bay Area cities like Huizhou and Jiangmen, where land and power capacity are more readily available. The region's economic integration makes this feasible in a way it would not be for most other financial centres managing the same tension.

The honest caveat is that the piece does not address the regulatory and data residency complexities that arise when Hong Kong financial institutions run workloads across the border. Nor does it quantify how much compute demand financial AI and regtech actually generate relative to frontier training runs. What the reporting does not give you is a worked-out cross-border governance framework that would make GBA compute routing practical for institutions subject to Hong Kong law.

For anyone building or buying AI infrastructure in the region, the direction is worth watching regardless. If Hong Kong policymakers take the specialization argument seriously, it could reshape where regulated financial AI gets built and run across the entire bay area.