A financial services company is training a machine learning model on a large dataset stored in Amazon S3, consisting of thousands of CSV files. The training performance is significantly impacted due to high I/O latency when reading the dataset. What immediate solution should be implemented to enhance training performance with minimal operational overhead?