China pivots to cheap tokens over bigger models at AI conference

The 2026 World Artificial Intelligence Conference in Shanghai, held July 17–20, brought together more than 1,100 exhibitors across three districts and four venues. Throughout the event, a single concept dominated the conversation at vendor booths and in keynote presentations: the "token factory." This repeated emphasis signals a notable reorientation in how China's AI sector is positioning itself. Rather than racing to build the largest or most capable models—a central narrative in global AI competition—Chinese vendors and industry speakers were collectively highlighting their focus on token production: the ability to generate outputs at volume while maintaining cost efficiency. The prominence of this message across such a large and representative gathering suggests this is not an isolated vendor preference but a coordinated strategic direction within the broader Chinese AI industry, reflecting either emerging technical insights about value creation or a recognition of resource and market constraints that make token efficiency more valuable than raw model scale.

The recurrence of the "token factory" phrase across multiple vendor booths and keynotes at China's largest AI showcase suggests a deliberate industry-wide messaging shift. Rather than focusing on building ever-larger foundation models—the dominant narrative in Western AI development over the past two years—Chinese vendors and thought leaders are publicly emphasizing token production and efficiency. This pivot appears to reflect a pragmatic reassessment of where competitive advantage and scalable value lie in the current market: not necessarily in parameter count or model size, but in the ability to generate tokens (text or other outputs) at scale and cost-effectively. The scale of the event itself—more than 1,100 exhibitors across multiple districts—underscores how broad and institutionalized this shift has become within China's AI ecosystem.