Electricity scarcity will shape AI's future trajectory

The race for supremacy in artificial intelligence is often portrayed as a contest of intellectual prowess: better models, faster chips and more sophisticated algorithms. But this perspective misses a harder, less glamorous truth. The real frontier of AI isn't silicon — it's electricity. Training a single large language model consumes as much electricity as hundreds of United States homes do in a year. The daily operations of AI systems — ChatGPT queries, recommendation engines and autonomous systems — only magnify that demand.

Every breakthrough in AI depends on reliable power supply. Behind every data center lies a grid connection. The growth of AI has intensified pressure on governments and industries to reduce carbon emissions while meeting rising electricity demand.

Nowhere is this energy bottleneck more visible than in China's Taiwan region. The island produces 92 percent of the world's advanced computer chips — the very engines of the AI economy. But its power grid is already under strain. Nvidia's planned "giant AI supercomputer" will initially require 20 megawatts, potentially increasing to 100 megawatts. Yet, Taiwan's goal of sourcing 20 percent of its energy from renewables by 2025 is behind schedule. As one industry analysis states, "This energy bottleneck poses a serious constraint on Taiwan's ability to fully capitalize on its manufacturing strengths in the AI economy."

It is an acute paradox. The region that manufactures chips is struggling to power the computers that use them. AI competition, therefore, is not merely a contest of semiconductor production but also a competition between energy systems — between grids with spare capacity and those without, between cities that can expand capacity quickly and those hindered by slow decision-making and outdated infrastructure.

In the Global East, the Chinese mainland faced this challenge very early. For years, the mainland's digital economy grew unevenly, with computing capacity concentrated in coastal megacities while renewable energy resources — hydro, solar, wind — lay locked in the less developed western regions. The mismatch was costly, leading to expensive transmission and grid strain in Beijing and Shanghai, while excess green energy in the Xinjiang Uygur autonomous region and Sichuan province went unused.

Beijing's solution was to relocate energy-intensive data processing to western regions where electricity is abundant, cheap and increasingly green. The logic is simple but powerful: move the computation to the energy, not the energy to the computation.

Guizhou province exemplifies this approach. With its cool climate that cuts cooling costs, ample hydropower, and available land, Guizhou hosts some of China's largest data center clusters — including those of Tencent, Apple, Tesla and Huawei. Once a poor, mountainous province, Guizhou has now become an integral node in the nation's AI infrastructure.

This model offers three clear advantages. First, it reduces grid pressure on coastal hubs. Second, it utilizes renewable energy that would otherwise be wasted. Third, it creates a form of spatial division of labor: innovation and algorithm design remain in Beijing, Shenzhen and Shanghai, while energy-intensive computation migrates to the west. This aligns AI's electricity hunger with China's geographical and climatic realities.

What lessons can the Hong Kong and Taiwan regions draw from this? Both regions face acute land and energy constraints and aspire to lead in AI. Neither can simply copy Guizhou's model because they lack vast inland territories.

There could be a different path based on a simple causal logic. AI's electricity demand varies across cities, producing differential investment patterns and creating urban disparities. Some cities attract AI investment, while others lose it. In the cities that become AI hubs, rising electricity demand can intensify intra-urban conflicts over power allocation.

For Taiwan, acknowledging the trade-offs is crucial. Its semiconductor dominance cannot be taken for granted if energy remains a constraint. Expanding renewable energy development, rethinking nuclear timelines and creating dedicated "AI energy zones" near existing high-capacity substations are not radical ideas, but survival strategies. The Taiwan region could also designate specific industrial parks where AI computing is prioritized, while outsourcing less time-sensitive data processing.

For Hong Kong, the focus is different but no less urgent. The city has world-class finance, legal institutions and connectivity — but almost no domestic energy generation and very little land for data centers. Hong Kong cannot compete on cheap electricity. Instead, it can specialize in high-value but low-energy segments of AI, such as algorithm design, financial AI, regulatory technology and model alignment. Meanwhile, Hong Kong could also serve as a "gateway coordinator", channeling capital and talent into the Guangdong-Hong Kong-Macao Greater Bay Area computing infrastructure located in less constrained cities such as Huizhou or Jiangmen.

Crucially, both regions must anticipate intra-urban conflict. When AI data centers come to dense cities, they compete with hospitals, schools and housing for grid capacity. Who finances the new substation? Whose power dims during heat waves? These are not merely technical questions or environmental issues, but political ones that demand transparent rules and explicit equity assessments.

The rise of AI will reshape cities. But whether it brings both people and nature together or drives them apart depends on a factor most policymakers ignore: megawatts and timelines, not just models and margins. The Chinese mainland is already crafting its energy-aware AI strategy. For the Hong Kong and Taiwan regions, the question is not whether to follow, but how fast they can adapt and catch up.

Xu Ying is an assistant professor at the School of Government, Central University of Finance and Economics; and Zhang Weishi is an associate professor at the Faculty of Geography, Tianjin Normal University.

The views don't necessarily reflect those of China Daily.

If you have a specific expertise, or would like to share your thought about our stories, then send us your writings at opinion@chinadaily.com.cn, and comment@chinadaily.com.cn.