AWS Lets Multiple Models Share One GPU in SageMaker

AWS added NVIDIA Multi-Instance GPU support to SageMaker real-time inference on November 6, 2026, with its benchmarks showing an H100 partition beating a whole A100 by 53–89%.