Why is published content gaining new value?

  • Large models need high-quality data for training. EpochAI forecasts that publicly available high-quality human text data could be completely exhausted between 2026 and 2032; several people close to traditional content companies say that AI companies in China have already begun purchasing compliant datasets from content organizations.
  • Industrial Securities believes traditional publishing copyrights are high-quality, have ascertainable ownership rights and can be updated on an ongoing basis. They could therefore become training-data assets needed for AI production, in addition to being sold to readers.

How do data value and IP value differ?

  • Training-data value comes from using content to train models; IP value comes from adapting and developing stories, fictional worlds and characters. The two correspond to different uses of published content.
  • Industrial Securities believes that as AI lowers the production barriers for short dramas, motion comics and animation, the scarcity of strong story concepts and character assets becomes more pronounced. IP from both traditional publishing and online literature could be revalued.

How are the two publishing companies positioning themselves?

  • China Science Publishing & Media has been developing knowledge-service products such as professional knowledge databases since 2019. In 2025, it launched Yuan Zhi, a medical large model, and has been seeking to expand ways to monetize AI knowledge services. Its academic content resources provide the foundation for such services.
  • Duzhe Publishing & Media is advancing reading and writing platforms, integrated publishing platforms and the digitization of the Siku Quanshu, while exploring the use of AI to empower its digital publishing, marketing and cultural and creative businesses.