AI Photo: VCG

China's southern regional electricity market has officially begun full operation, bringing together coal, hydro, nuclear, wind and solar power from South China's Guangdong, Yunnan, Guizhou, Hainan provinces, and Guangxi Zhuang Autonomous Region under a single set of trading rules. More than 290,000 market participants - including generators, retailers, large users, new-energy storage providers and virtual power plants - are registered.

Analysts noted that while residential electricity prices remain regulated and insulated from market swings, industrial users, especially artificial intelligence (AI)-related firms, gain greater flexibility to buy and sell power.

The market aims to improve renewable energy utilization across the region, currently holding above 95 percent, higher than the national average. 

"For data center operators and AI companies, the change is more direct. Electricity is shifting from a fixed input to a manageable variable in the cost of every token," Zhang Xiaorong, director of the Beijing-based Cutting-Edge Technology Research Institute, told the Global Times.

Why does a regional power market matter to the AI industry? Because producing tokens - the basic units of text, code and other output generated by AI models - is a power-intensive business. The significance of the market is that it turns electricity from a fixed local cost into a cross-provincial, price-sensitive input that data centers can shop for and adjust around, said Zhang.

Power now accounts for 50-60 percent of data center operating costs, up from about 40 percent earlier, according to the China Center for Information Industry Development (CCID). 

"That means even a small change in the power prices can move the cost of producing a token at scale," said Zhang.

An anonymous industry insider told the Global Times that one kilowatt-hour can support the generation of roughly 1.67 million output tokens. 

The scale of China's token consumption has continued to grow. According to the China Academy of Information and Communications Technology, prices for Chinese AI large model token services have doubled from the second half of 2025, while usage has tripled. The average daily token use reached about 500 trillion in June 2026, nearly four times the level in December 2025.

"Whoever controls low-priced green electricity controls token competitiveness" has become an industry consensus, said Zhong Xinlong, director of the AI research office at the CCID. He noted that inference loads now dominate, turning data centers into baseload consumers that require both stable round-the-clock supply and flexible peak-valley scheduling.

To deal with the increasing token usage and electricity demand of the AI industry, China has held pilot programs in cross-provincial electricity trading, and early results are already visible. In May, nine data centers in Guangdong completed the first cross-province point-to-point green power deals, purchasing 17.45 million kilowatt-hours of electricity from South China's Guangxi Zhuang Autonomous Region and Southwest China's Yunnan Province at roughly 0.40 yuan ($0.06) per kilowatt-hour - about 0.13 yuan cheaper than comparable in-province transactions. 

Apart from cheaper prices, the new electricity market can also increase the flexibility of power supply to data centers, the insider said.

For example, China Unicom's Shaoguan data center and China Mobile facilities in Guangzhou and Zhanjiang joined Guangdong's spot market as virtual power plants on May 14. A virtual power plant is essentially a software platform that bundles adjustable electricity demand and bids it into the market like a power plant. 

By ramping up computing tasks during low-price periods and curtailing non-urgent loads at peaks, they practiced a "computing power follows power prices" model. The result was that regional average power prices in Guangdong for computing loads fell 32.2 percent year-on-year, and cluster energy use per unit of computing power ran 12 percent below the national average.

Gao Wen, a Chinese Academy of Engineering academician, argued at this year's World Artificial Intelligence Conference that sustainable computing networks require electricity costs ideally no higher than 0.3 yuan per kilowatt-hour when weighing chip, storage, network and power factors together. 

"The current 0.40-yuan cross-province green rate still sits above that threshold, yet the institutional pathway for data centers to pursue cheaper renewable power across provincial borders is now open," said Zhang.

The insider noted that the southern electricity market does not produce tokens itself.

"It is rewriting the electricity price card that increasingly determines who can produce them competitively," said the insider.