China Pivots AI Subsidies to Token Consumption as Agent Demand Grows

China's artificial intelligence sector is pivoting from hardware buildouts to commercial token consumption, driven by the rapid adoption of AI agents. According to a report by the China Telecom Research Institute, domestic token usage is projected to surge through 2030, with inference workloads dominating demand. In response, municipal governments in Chengdu, Beijing, and Chongqing are shifting policy support from subsidizing servers to funding token consumption through vouchers, dedicated 'token factories,' and experimental financial instruments, while telecom giants set up token management offices.

The shift in municipal subsidies from capital expenditure on graphics processing units to operational token consumption signals a maturation of China's AI ecosystem toward service monetization and measurable utility. By treating tokens as standardized production units, cloud providers, telecom operators, and local authorities are building an integrated commercial chain that ties raw compute capacity directly to downstream application revenues.

China's artificial intelligence development is entering an execution-focused phase centered on AI agents and measurable software usage, driving domestic telecommunications carriers and local authorities to treat tokens as a core economic unit. According to a report released by the China Telecom Research Institute and covered by state broadcaster CCTV, domestic token demand is set for exponential expansion. Rao Shaoyang, director of the institute's Industry and Corporate Strategy Research Institute, projected that China's annual token consumption would reach 100 quadrillion in 2026 and exceed 35,000 quadrillion by 2030, reflecting a compound annual growth rate near twelvefold. Rao added that domestic computing demand is expected to multiply roughly tenfold annually over the next two to three years, with inference computing accounting for an estimated 80 percent of the total market by 2029. State-owned telecommunications operators are adjusting their internal governance to reflect this shift. Qian Ling, a chief expert at China Mobile, noted that the group established dedicated top-level management offices for computing power and tokens to coordinate infrastructure delivery and usage. China Mobile's internal calculations estimated daily token output at approximately 3.4 trillion last year, with capacity expected to increase further. Simultaneously, municipal policymakers are adjusting their subsidy structures from hardware acquisitions to continuous operating costs. Rather than simply subsidizing GPU servers, local governments are encouraging software utilization through model application programming interface credits and token vouchers, as reported by People's Daily. On September 11, Chengdu unveiled a draft plan proposing up to 100 million yuan in annual token vouchers, offering subsidies of up to 30 percent of purchase costs—or 50 percent for micro and small enterprises—capped at 2 million yuan per applicant. In Beijing, two municipal 'token factories,' including the state-backed Jingsuan facility, have begun operations to package raw computing into standardized model delivery services. Chongqing's Big Data Application and Development Administration also published a draft action plan spanning 2026 to 2028. The framework proposes building a token trading service platform, integrating tokens into manufacturing and municipal management, and encouraging financial institutions to experiment with token-backed credit facilities and real estate investment trusts for data infrastructure. Industry analysts observe that across the supply chain, the focus is coalescing around a single operational loop: converting compute investments into billable token outputs.