China Merchants Bank Unifies AI Workloads on Kubernetes, Doubles GPU Utilization

Tue, 08 Sep 2026 14:50:52 +0000

Platform/SRE — Learn: Interesting case study on consolidating AI training and inference onto shared Kubernetes infrastructure, improving accelerator utilization from 35% to 60%+ — useful reference architecture if evaluating GPU workload scheduling.
CI/CD — Skip
Leader — Learn: Notable FinOps result: 60%+ reduction in inference cost per million tokens via platform consolidation — worth reviewing as a benchmark when evaluating GPU infrastructure efficiency investments.

Gpu-Utilization on CuraDevOps