Beijing issues action plan to grade and boost ‘token factories’

Beijing's move is notable less for the token-economy framing than for treating inference capacity as planned infrastructure, with tiered metrics that let officials rank and steer operators. That shifts the policy lever from model development to throughput, latency, cache efficiency and power — the cost drivers that determine whether inference scales economically. Who feels it most: domestic compute and model-serving providers, plus edge and chip vendors positioned for inference-specific demand. The open question is whether these standards stay evaluative or become a de facto gate for support and deployment.