Amazon S3 Files: Object Storage Now Acts as a Native File System for Cloud Compute

AWS Unveils S3 Files, Bridging Object Storage and File Systems With High-Performance Access

AWS today launched Amazon S3 Files, a new capability that allows any AWS compute resource—from EC2 instances to Lambda functions—to access Amazon S3 buckets as fully functional file systems. The service eliminates the decades-old tradeoff between object storage’s cost and durability and a file system’s interactive, low-latency experience.

“With S3 Files, we are breaking down the last barrier between object storage and file systems,” said Jeff Barr, AWS Chief Evangelist. “Customers no longer have to choose. S3 now becomes a central hub for all data, accessible directly from any compute service with full file system semantics.”

How It Works

S3 Files presents objects as files and directories using the standard NFS v4.1+ protocol. All create, read, update, and delete operations are supported, and changes made through the file system are automatically reflected back into the S3 bucket. The system attaches to multiple compute resources simultaneously, enabling data sharing across clusters without duplication.

“We’ve designed S3 Files to handle both latency-sensitive workloads and high-throughput sequential access,” Barr added. “Frequently used files stay on high-performance local storage, while rarely accessed or large sequential reads are served directly from S3 to optimize cost and throughput.”

Background

For over a decade, AWS trainers and customers grappled with the fundamental differences between object storage and file systems. S3 objects acted like books in a library—you couldn't edit a page, only replace the whole book—while file systems allowed byte-level modifications. This forced workloads into separate storage silos.

S3 Files bridges that gap. It allows any general-purpose bucket to be mounted as a native file system on EC2 instances, containers (ECS/EKS), and AWS Lambda functions. The service uses a high-performance storage layer that intelligently pre-fetches data and offers fine-grained control over what is cached versus served directly from S3.

What This Means

For cloud architects and developers, S3 Files removes a major architectural constraint. Machine learning training pipelines that previously required costly file system replication can now run directly against S3 with interactive file access. Production applications gain the ability to read and write live S3 data without custom synchronization scripts.

“This is a game-changer for AI and data-intensive workloads,” said Barr. “Teams building agentic AI systems or running large-scale analytics can now treat S3 as their single source of truth with file-level flexibility.” AWS claims S3 Files is the first cloud object store to offer this level of file system integration.

By default, the system automatically stores files that benefit from low latency on high-performance storage. For large sequential reads, it streams data directly from S3 to maximize throughput. The service also supports byte-range reads, transferring only the requested bytes to minimize data movement and cost. Intelligent pre-fetching anticipates access patterns, and users can choose to load full file data or just metadata based on their specific needs.

Key Features:

• Native NFS v4.1+ support for all file operations
• Automatic synchronization between file system and S3 bucket
• Attachment to multiple compute resources without data duplication
• Intelligent caching with fine-grained control over local storage
• Streaming for large sequential reads and byte-range requests

S3 Files is available today for all general-purpose S3 buckets in AWS Regions where Amazon EFS is supported. Pricing includes charges for high-performance storage used and standard S3 operations. No additional software or agents are required.