China Merchants Bank Wins CNCF End User Case Study Contest for Unified AI Training and Inference Platform
China Merchants Bank, a major Chinese commercial bank, won the CNCF End User Case Study Contest at KubeCon + CloudNativeCon China 2026. The award recognizes its unified Kubernetes-based platform that integrates multiple open-source tools to streamline AI training, fine-tuning, and inference across nearly 10,000 heterogeneous accelerators. The bank's new cloud-native infrastructure significantly improved efficiency by raising average accelerator utilization from 35% to over 60%. It also reduced the cost of processing one million tokens (input/output combined) by more than 60%, under similar model and service conditions. The platform uses Kueue, KEDA, Prometheus, HAMi, Fluid, and an in-house Twinkle training framework that allows multiple tenants to share a single base model instance efficiently. This achievement highlights the bank's focus on software innovation to optimize hardware usage through cloud-native technologies. By unifying AI workloads under one control plane, China Merchants Bank is able to better manage resource allocation across different types of compute demands—such as stable training and elastic inference scaling—which supports broader adoption of AI in financial services.
Topics: AI Cloud Native Computing Foundation (CNCF) infrastructure efficiency Kubernetes