As Beijing pushes to develop its AI industry, a business report released by the listed Chinese company Chinese Online (中文在线) has drawn attention to how the country’s AI computing resources are being used. According to the report, 55 percent of AI tokens consumed in China go to short dramas and video generation, 24 percent to e-commerce advertising, and only 15 percent to research and development. The figures have fueled public debate over whether more than half of China’s computing power is being used to churn out “feel-good” dramas.
The data comes from the company’s investor relations activity record. It shows that token consumption for AI short dramas and video generation is more than double that of software development. Chinese Online said the heavy usage stems from the fact that short dramas demand extremely fast video generation and token turnover.
Commentators have questioned whether directing so much computing power toward a single content format is an efficient path for deploying the technology or a collective drain caused by overcapacity.
A report on AI-driven global micro-drama development, jointly released by a research center under China’s National Radio and Television Administration and other bodies, found that the global micro-drama market entered a boom phase in 2025. Its total size approached US$18 billion, with China accounting for more than US$14 billion, or 77.78 percent. Separate data shows 430,000 micro-dramas went online in China from January to August this year, 13 times the total for all of 2025.
Reports note that AI-generated live-action dramas typically consume two to three times as many tokens as AI animated series because of stricter visual requirements. Production teams also repeatedly generate and screen footage to meet directors’ demands for narrative and camera work, which drives consumption higher.
Not everyone agrees with the characterization. Liu Xing, public relations head for ByteDance’s AI platform Doubao, said only about 10 percent of the company’s total computing power goes to training and running video generation models.
Source: Central News Agency (Taiwan), October 20, 2026
https://www.cna.com.tw/news/acn/202610200105.aspx