Training and processing jobs previously incurred delays downloading large container images often exceeding 10 GB. SOCI indexing allows SageMaker to load only necessary data to start containers immediately while streaming remaining files in the background, cutting container startup latency by up to 90%.
This feature requires no changes to existing job definitions and is compatible with existing Deep Learning Containers using their soci-tag variants. It is available across all AWS regions supporting SageMaker AI, improving cost efficiency and resource utilization by minimizing GPU idle time before jobs begin running.
Source & further reading
Amazon SageMaker AI now supports SOCI-indexed images to start training jobs in seconds ↗Source published 2026-10-29. This report was prepared with AI assistance from the linked primary-source material. Claims and limitations should be checked against the original announcement or research.