Warner Bros. Discovery achieves 60% cost savings and faster ML inference with AWS Graviton

This article describes how Warner Bros. Discovery achieved significant cost and performance improvements by migrating ML inference workloads to AWS Graviton-based instances.

• Achieved 60% average cost savings, up to 88% for catalog ranking models
• Improved latency by 7% to 60% across different ML models
• XGBoost models showed 60% P99 latency reduction
• Used AWS Graviton processors optimized for ML with Neon and SVE instructions
• Leveraged SageMaker Inference Recommender for automated benchmarking
• Employed shadow testing to validate performance before production deployment
• Completed proof-of-concept in one week, full migration in one month
• Serves 125M+ users across 100+ countries with sub-100ms latency requirements
• Planning to migrate remaining models to achieve 100% Graviton adoption

WBD's migration demonstrates how AWS Graviton processors deliver substantial cost savings and performance improvements for large-scale ML recommendation systems while maintaining reliability and user experience.