Electricity scarcity will shape AI's future trajectory: Grid strain, renewables, and carbon impacts

A China Daily opinion piece argues that the real bottleneck for AI is not chips or algorithms but electricity. Training a single large language model can use as much power as hundreds of US homes in a year, and running AI systems day to day adds even more demand. The article highlights how Taiwan, which produces 92% of advanced chips, faces grid strain and is behind on its 20% renewable energy target by 2025, creating a paradox where the chipmaker struggles to power the computers using them. The piece contrasts this with mainland China's strategy of moving energy-intensive data processing to western regions where renewable energy is abundant and cheap, using Guizhou province as an example. It recommends that both Taiwan and Hong Kong acknowledge energy trade-offs, expand renewables, reconsider nuclear timelines, and create dedicated AI energy zones. For Hong Kong, the focus should be on low-energy AI segments like algorithm design and financial AI, while using the Greater Bay Area for computing infrastructure. The article stresses that intra-urban conflicts over grid capacity between AI data centers and hospitals or housing are political issues requiring transparent rules.