This talk explains how to build a cost optimization oracle for replicated object storage, achieving 1000x cost reduction over prior approaches. The optimal choice of replicated object storage can provide distributed cloud services with high service quality at low cost. However, today's cloud makes for a massive combinatorial space across hundreds of cloud regions/object storage tiers. Further, cloud services store billions of objects with distinct access patterns that demand distinct storage configuration. In this talk, I will explain how to achieve fast and accurate object placement at cloud scale, by precomputing all optimal storage choices for all workloads.