Inference Costs Could Drop 10x Further as Cross-Cluster Heterogeneous PD Separation Architecture Scales Up: Xia Lixue from Infinite

On July 20, at the 2026 World Artificial Intelligence Conference, Xia Lixue, co-founder and CEO of Infinite, pointed out that the current AI infrastructure must “serve larger scales with extreme efficiency.” As of July, the platform’s daily token call volume has grown 40 times compared to last December. It is reported that Infinite has already driven a 10-fold reduction in inference costs through a series of technological innovations. Xia Lixue predicts that with the large-scale implementation of cross-cluster heterogeneous PD separation architecture, there is still room for another 10-fold decrease in inference costs.