Case Study: Subaru accelerates AI development and cuts ML container image pulls from hours to minutes with CNCF projects

Subaru cuts ML image pull time from 3 hours to 3 minutes with Cloud Native Computing Foundation

Subaru Corporation, an automotive manufacturer, faced significant challenges in accelerating its AI development for its next-generation EyeSight safety system. Their on-premises GPU environment suffered from extremely slow pull times for large machine learning container images, taking over three hours, which severely hampered development cycles. Additional hurdles included manual, error-prone application deployments and a lack of automation for complex multi-step machine learning workflows. To overcome these issues, Subaru partnered with the Cloud Native Computing Foundation (CNCF).

By adopting CNCF technologies including Kubernetes, Harbor, Argo CD, and Argo Workflows, Subaru built a highly efficient AI model development platform. The solution optimized image delivery, standardized deployments via GitOps, and automated end-to-end machine learning pipelines. This resulted in a dramatic 60x improvement, reducing container image pull times from three hours to just three minutes. The Cloud Native Computing Foundation's tools also enabled Subaru to manage 25 application definitions declaratively, significantly improving operational consistency, reproducibility, and overall development speed.

Subaru