A company is planning to run an AI/ML workload on AWS. The company needs to train a model on a dataset that is in Amazon S3 Standard. A model training application requires multiple compute nodes and single-digit millisecond access to the data. Which solution will meet these requirements in the MOST cost-effective way?