AI workloads are resource-intensive, driving up costs. This talk explores model pruning techniques and Kubernetes-native strategies for scalable AI deployments, focusing on resource scheduling, autoscaling, and efficient inference serving in cloud.