豆包 public relations chief 刘星 said Monday that 字节 devotes about 10% of its computing capacity to video-model training and inference.
The June 3 investor record from 中文在线集团股份有限公司 placed 55% of China’s AI token use in the short-drama and video-generation category. It assigned 24% to e-commerce and marketing and 15% to software development.
The 55% figure counts token use across applications, while 刘星’s 10% figure concerns 字节’s computing capacity. The record does not quantify video generation’s share of China’s overall computing capacity.
中文在线 said producers generate and select video shots repeatedly, driving up token use. It said the expense of high-quality production makes it harder for small studios to compete with producers that have greater resources.
