Beijing Unveils 3-Year Action Plan to Build 'Token Economy' and AI Infrastructure
The Beijing Municipal Bureau of Economy and Information Technology released the 'Action Plan for Accelerating Token Economy Development (2026–2028).' The policy introduces ten key initiatives, including establishing high-standard 'token factories,' creating token evaluation metrics, building unified distribution platforms, and advancing AI inference technologies. This initiative marks a major strategic shift by treating AI tokens—the fundamental metric of generative AI output—as a standardized industrial commodity and economic driver. It aims to lower inference costs, scale enterprise AI adoption, and position Beijing as a global leader in AI compute infrastructure and agent software ecosystems. The plan establishes technical metrics for token factories, focusing on token throughput, Time to First Token (TTFT), cache hit rates, and Power Usage Effectiveness (PUE). It also prioritizes specialized inference chips, model intelligent routing, credit products for compute infrastructure, and international expansion via multi-lingual token services.
## BACKGROUND
In large language models (LLMs), a 'token' is the fundamental unit of processed text, with compute consumption and API costs typically billed per token. Metrics like Time to First Token (TTFT) evaluate real-time responsiveness, while 'intelligent model routing' automatically directs user prompts to the most cost-effective LLM based on task complexity.