How Caching Docker Layers on Google Cloud Build Can Cut CI Costs by 94%

A development team discovered their CI pipeline was spending roughly $340 per month rebuilding identical Docker layers from scratch on Google Cloud Build. The root cause is that Cloud Build uses ephemeral workers that discard all local BuildKit cache data after each job completes. By using Docker Buildx's registry cache exporter to store OCI cache blobs in Artifact Registry, teams can persist layer cache across builds at a fraction of normal storage costs. Properly configured with memory limits and registry-backed cache references, this approach achieved a 94% cache hit rate across six production services after just three deployment cycles. The key implementation details include setting a hard 6 GB memory ceiling for BuildKit and using the ignore-error flag to handle missing cache references on first-time builds.