Better Models Aren’t Enough: China’s AI Race Turns to Cost, Products and Paying Users

Token Optimization

The AI race in China is undergoing a significant transformation, moving beyond the pursuit of ever-more-capable models to a more pragmatic focus on cost-effectiveness and demonstrable business value. UBS Securities analyst Xiong Wei points to a critical shift from 'token-maxxing,' where the goal was to encourage maximum AI usage, to 'token optimization.' This new paradigm emphasizes efficiency, with companies becoming more discerning about the precise level of AI intelligence required for specific tasks. The rising costs associated with extensive AI consumption have prompted buyers to closely evaluate the balance between AI performance and the associated price tag. This trend is particularly beneficial for Chinese open-source AI models, which are increasingly competitive due to their improving capabilities and significantly lower development and API costs compared to their international counterparts. Xiong estimates that some leading Chinese models can be developed for less than a tenth of the cost of overseas alternatives, with API pricing around 10% to 20% of global competitors. This cost advantage positions these models favorably for handling repetitive or lower-risk workloads, where the economic benefits of AI can be more readily quantified.

User Attention Constraints

Despite the advancements in AI model capabilities and the potential for cost savings, a fundamental challenge persists: the finite nature of user attention and time, particularly on mobile devices. Kenneth Fong, UBS's head of China internet research, highlights that growth in user traffic and engagement on major internet platforms has plateaued. While AI can streamline content creation and enhance recommendation systems, its ability to capture a larger share of users' limited attention is not guaranteed. The example of AI-generated short dramas illustrates this point: producing more content at a lower cost does not automatically translate into increased viewership or engagement. The success of such content hinges on its ability to resonate with audiences and hold their interest, a factor that AI alone cannot guarantee. This underscores the need for AI strategies to align with user behavior and market demand, rather than solely focusing on technological output.

Commercialization Pathways

The debut of 'The Later Journey to the West,' an AI-generated fantasy series on Mango TV, exemplifies the dual nature of AI's impact on the media industry. The series, produced using Mango TV's in-house AIGC platform, Mango Lingchuang, showcases significant production efficiencies, generating numerous character and scene assets and testing a 'produce, review and broadcast in parallel' model. Early audience traction, with strong ratings and millions of plays, suggests potential for AI-driven content to capture viewer interest. However, the underlying challenge remains the commercialization of these AI-driven productions. Regulators have urged exploration of both a workable technical path for AI storytelling and a viable commercialization strategy for AIGC dramas. This mirrors the broader industry sentiment that while AI can reduce production costs and increase output, its ultimate success will be measured by its ability to generate sustainable economic value and achieve profitability in a competitive market.