Wuwen Xinqiong released the Agentic Infra strategy of "Front Shop, Back Factory and One Center" at WAIC, using AI to improve token production efficiency

Zhixi Author | ZeR0 Editor | Mo Ying Zhixi reported on July 21 that on July 20, during the World Artificial Intelligence Conference WAIC 2026, Wuwen Xinqiong co-founder and CEO Xia Lixue proposed that from Pre-training (pre-training) to Post-training (post-training), from Test-time (inference time expansion) to Agentic scaling (agent scaling), in the four Scaling Under the influence of Law, the supermarkets of model training and inference are nurturing each other. The current AI infrastructure (AI Infra) "must serve a larger scale with ultimate efficiency." Based on the two anchors of scale and efficiency, Xia Lixue started from Wuwenxinqiong's AI productivity formula (AI productivity = intelligent resource scale × Token conversion efficiency × AI productivity conversion efficiency), and announced for the first time Wuwenxinqiong Agentic Infra's "front store, back factory and one center" strategic layout, covering three major core technical capabilities and product service systems: "Computing power distribution center": Agentic Infra autonomous infrastructure platform "Token factory": Agentic MaaS large model service platform "AI productivity store": Agentic Infra industry solutions This Agentic Infra product and service system is based on infrastructure, gathering energy and chips downwards, and supporting models and applications upwards. It not only greatly improves the scale and efficiency of the transformation of underlying resources into AI productivity, but also has the native AI capabilities of the Agentic AI era - using infrastructure intelligent swarms to empower resource scale expansion, improve token production, and support the continued evolution of Agentic AI. 1. Agentic Infra autonomous infrastructure platform pursues maximization of the scale of intelligent resources. The core goal of Wuwenxinqiong’s Agentic Infra autonomous infrastructure platform is to maximize the scale of intelligent resources. It is like a "computing power distribution center" that combines scattered and different