Subaru reduced 30GB container pulls from three hours to three minutes (www.cncf.io)

Subaru has achieved a remarkable transformation in its AI development process by reducing the pull times for its 30+ GB AI container images from three hours to just three minutes—a 60x improvement. This was made possible by optimizing its Kubernetes networking architecture with Envoy Gateway and adopting a cloud-native approach that integrates multiple CNCF technologies. As a result, Subaru has significantly enhanced developer productivity, operational efficiency, and the reproducibility of machine learning (ML) workflows essential for its next-generation EyeSight advanced driver assistance systems (ADAS). By implementing GitOps practices with tools like Argo CD and Helmfile, Subaru standardized application deployments, streamlining its development environments. The automation of end-to-end ML pipelines using Argo Workflows has further accelerated the pace of innovation while reducing operational friction. As AI workloads continue to grow in complexity, Subaru plans to build upon this cloud-native infrastructure to drive further improvements in efficiency and scalability, reinforcing the importance of adopting open-source technologies in driving AI advancements and addressing real-world infrastructure challenges in the automotive industry.