A company uses an NFS-based data store to store data for ML training.

Option B is correct because Amazon EFS is a managed NFS-based shared file system that supports Linux clients, hybrid access, and file-system semantics such as file locking. AWS documentation states that Amazon EFS provides strong data consistency and file locking, which is an explicit requirement in the question. AWS also states that you can mount EFS file systems on on-premises data center servers when connected to your Amazon VPC through AWS Direct Connect or Site-to-Site VPN, which directly satisfies the hybrid access requirement.

For the SageMaker side, AWS documentation states that SageMaker AI creates a default Amazon EFS volume for Studio domains, and you can also add a custom Amazon EFS file system so domain users can access it in SageMaker Studio. AWS also notes that SageMaker AI Studio notebooks use Amazon EFS as a persistent storage layer. That makes EFS the most natural AWS storage service for a shared file repository that must be accessible from SageMaker notebooks.

The other choices are weaker. Amazon S3 with Mountpoint is object storage access and is not the best answer for a shared NFS-style data store with file locking requirements. FSx for Lustre is high-performance parallel file storage and can support on-premises bursting, but the question explicitly starts from an NFS-based shared data store, and EFS is AWS’s native NFS file service. Amazon EBS cannot be mounted simultaneously across on-premises servers and SageMaker notebooks as a shared, managed hybrid file system. Therefore, the best verified AWS-docs answer is B.