New data from global AI model aggregator OpenRouter shows that, as of mid-March, Chinese large models processed 4.69 trillion tokens in a single week, surpassing the United States for a second consecutive week.

Tokens, the smallest unit of data a model processes, are a key measure of real-world usage and commercial value.

One of the main drivers behind China's rapid adoption is price.

"At comparable performance levels, some overseas models can cost more than ten times as much," Yan Yijun, a representative from the Shanghai-based AI company MiniMax, told China Media Group.

Analyst said that this pricing advantage stems from Improvements in the underlying model architecture

"Faced with complex tasks, domestic models can generate interpretable reasoning chains," said Zhang Libo, a researcher at the Institute of Software under Chinese Academy of Sciences. "If something goes wrong, the entire decision-making process can be traced, which also helps control costs."

Lower energy costs provide another edge.

Energy can account for 70% to 80% of operating costs in large AI clusters, according to industry sources cited by CMG. China's relatively low electricity prices, stable energy supply, and rapidly expanding clean energy capacity have together built a more competitive cost base for its AI industry.

Behind trillions of tokens are increasingly high-frequency, paid use cases, signaling a shift from experimentation to real-world deployment.

At some computing centers, when 5,000 AI chips run simultaneously, they can generate roughly 30,000 yuan (about $4,356) in economic value per minute, according to industry estimates. The surging demand for AI computing power is already reshaping the broader supply chain.