Improved Models Alone Won't Suffice: China's AI Competition Shifts Focus to Costs, Products, and Monetizing Users.

      Editor's note: This article was composed by Lucia, a TechNode reporter. During an online media briefing on September 1 for the 23rd UBS Securities A-Share Seminar, the focus of the discussion regarding China's AI sector was on a key question: can reduced costs and broader AI implementation lead to tangible business value? UBS Securities' China internet analyst Xiong Wei identified three important themes in China's large-model sector: model capability, "token ROI," and monetization. As Chinese developers continue to refine their models—especially in coding and agentic capacities—companies are becoming more discerning about the level of intelligence required for specific tasks.

      This indicates a transition from "token-maxxing," which encourages increased AI usage, to "token optimization," according to Xiong. Rising AI expenses have made it more challenging for businesses to assess the economic value derived from their token usage, prompting buyers to focus more on the interplay between performance and cost. Xiong noted that this trend is advantageous for Chinese open-source models, which, with their enhanced capabilities and lower costs, are becoming more suitable for repetitive or lower-risk tasks.

      Xiong estimated that some leading Chinese models could be developed for less than one-tenth the cost of their foreign counterparts, while average API pricing may range from 10% to 20% of that of international competitors. However, reduced AI costs do not automatically equate to increased revenue. Kenneth Fong, the head of China internet research at UBS, pointed out that major Chinese internet platforms are confronted with a more fundamental issue: user traffic and time spent on mobile devices are no longer experiencing significant growth. While AI can reduce content production costs and improve the efficiency of advertising or recommendation systems, consumer attention remains limited. He cited AI-generated short dramas as an example, noting that producing more shows at a lower cost does not guarantee they will engage viewers more effectively.

      A new initiative from Mango TV demonstrates both aspects of this dynamic. The Later Journey to the West, a fantasy series produced by AIGC, premiered on Mango TV and during Hunan Satellite TV’s prime-time slot on August 31, marking the first AIGC long-form series to achieve this prime-time placement. Adapted from an anonymously authored fantasy novel from the late Ming or early Qing dynasties, the narrative follows a contemporary group of characters who embark on another quest to retrieve Buddhist scriptures after the original characters are misinterpreted.

      The first season is set to consist of 30 episodes, each approximately 40 minutes long. The series is being produced in collaboration with Mango Lingchuang, Mango TV’s internal AIGC production team, which had supported over 40,000 professional users and more than 3,900 projects by mid-2026. The platform generated 109 character assets and 143 scene assets for the series while also experimenting with a “produce, review, and broadcast in parallel” model, which allows subsequent episodes to be produced while earlier ones are aired. Positive initial audience engagement was evident, with real-time ratings from the August 31 premiere ranking first among provincial satellite channels in that time slot, according to ITHome, while ChinaTimes reported that it had accumulated 27.57 million views on Mango TV by September 2.

      However, it remains uncertain whether these efficiencies and early audience successes can evolve into a sustainable business model. Prior to the series launch, Hunan’s broadcasting regulator urged the project team to investigate a viable technical approach for AI-driven long-form storytelling as well as a commercialization strategy for AIGC seasonal dramas. This challenge resonates with points raised by both Xiong and Fong: as AI becomes less expensive and production costs decline, the emphasis is shifting from the volume of AI production to whether its application can yield adequate economic value.