Opinion | For Hong Kong to succeed in AI, energy cannot be an afterthought

The artificial intelligence competition is by nature an energy competition. The conventional narrative focuses on faster chips, yet training and inference consume vast amounts of power. The harsh geographic paradox is that the regions most advanced in artificial intelligence (AI) often face the most acute power constraints.
Hong Kong has emerged as a premier global AI hub. According to the Global AI Competitiveness Index, the city ranks third globally as an AI financial powerhouse. The chief executive’s 2025 policy address further underscored this momentum, highlighting Hong Kong’s world-class research, talent and capital.
But a critical blind spot remains: the energy-compute nexus. Hong Kong’s electricity balance has been in deficit since 1994 with annual shortfalls in recent years of over 1,500 kilowatt-hours per capita, necessitating imports including from the Daya Bay nuclear power station. While the government promotes AI industrialisation, it has yet to fully integrate energy-aware planning into its strategy.
Initiatives such as the AI Supercomputing Centre in Cyberport and the 10-hectare Sandy Ridge data facility cluster under construction reflect a massive commitment to compute capacity. Yet these projects prioritise technical throughput without corresponding long-term energy infrastructure planning, highlighting a mismatch between digital ambition and power supply resilience.
Hong Kong can leverage the Greater Bay Area as its hinterland. Instead of attempting to house all its data centre demand locally – which places unsustainable pressure on its ageing power grid – Hong Kong should act as the high-value brains of the Greater Bay Area. By serving as a gateway coordinator, the city can channel capital and talent into compute infrastructure located in cities like Huizhou or Jiangmen, where energy capacity is more abundant.
By specialising in high-value, low-energy segments – such as financial AI and regulatory technology – Hong Kong can solidify its position in the global AI hierarchy. Its legal prowess offers a unique opportunity to shape AI governance, bridging global regulatory divides.

This regional strategy – of leveraging the rest of the Greater Bay Area as Hong Kong’s energy hinterland – addresses the power bottleneck at the macro level. A complementary local strategy is also needed to tackle the intracity competition for grid capacity. Here, the mainland provides a vital model.
The “Eastern Data, Western Computing” initiative was launched to resolve the uneven distribution of compute and energy. By relocating energy-intensive processing to western provinces like Guizhou – where hydropower is abundant – China has decoupled high-value innovation in coastal hubs from energy constraints.
The logic is simple but powerful, aiming to move the computation to the energy, not the energy to the computation. Guizhou now hosts massive data centre clusters for tech giants such as Tencent Holdings, Apple and Huawei Technologies. Innovation, finance and algorithm design remain in high-density hubs such as Beijing, Shenzhen and Hangzhou. It is a deliberate exercise in statecraft, designed to align AI’s electricity demands with geographical realities.
This model of regional integration is increasingly critical. For instance, Taiwan’s manufacturing dominance is currently threatened by power bottlenecks. Nvidia CEO Jensen Huang has highlighted this crisis – as grid capacity remains a scarce resource, manufacturing and computing strength are increasingly tethered to local grid resilience.
Hong Kong, though not a manufacturing hub, faces a similar vulnerability on its data centre-intensive trajectory. The global AI race is no longer merely a contest of semiconductor fabrication; it is a competition between energy systems. To thrive, regions must shift from isolated city planning to integrated, intercity collaborations.

At the city level, AI infrastructure intensifies the competition for grid capacity, pitting data centres against essential public services. In a dense urban environment like Hong Kong, a data centre is not a neutral neighbour; it competes with hospitals, schools and housing for electricity.
Whose lights will dim as El Nino brings heatwaves? This is not just a technical or environmental issue; fundamentally, it is a political and equity issue. As we face ever more volatile climate patterns, grid allocation becomes a matter of public equity.
Cities are moving beyond reactive, firefighting approaches to proactive and anticipatory governance. With smart platforms like the “city brain” operating in Beijing’s Haidian district, the authorities can leverage real-time data to anticipate energy strain, dynamically balancing the needs of the digital economy with essential public services. The winners of the global AI race will be those that can navigate these spatial and political realities.
For Hong Kong, the path is clear: embrace Greater Bay Area integration and refine internal governance to ensure AI-driven prosperity does not compromise social stability. By treating energy as a strategic constraint rather than an afterthought, Hong Kong can ensure its digital infrastructure supports, rather than strains, the city’s future.
The rise of AI will reshape the regional and urban landscape of Hong Kong. AI development is a necessary, but not sufficient, condition for prosperity. On the other hand, having cheap, resilient and green energy provides a competitive edge, but it does not automatically generate innovation.
China has begun building its energy-aware AI strategy. For Hong Kong and the rest of the Greater Bay Area, the question is not whether to follow, but how fast they can shift the paradigm and catch up. The path is twofold: regionally, integrate into the Greater Bay Area’s energy-compute geographical redistribution; locally, adopt smart, equitable governance to prioritise grid allocation. Only by treating energy as a strategic constraint can Hong Kong turn its AI ambition into sustainable reality.