Research Finder
Find by Keyword
Amazon Introduces S3 Files: Can Object Storage Become the Default Execution Layer for File Workloads?
AWS extends S3 with native file system semantics, introducing a managed working set that enables file-based access without data movement while shifting cost and performance to access behavior.
04/14/2026
Key Highlights
- AWS introduced S3 Files, a new resource that enables S3 buckets to be accessed as POSIX-compliant file systems without data migration
- The service introduces an SSD-backed working set that materializes active data while maintaining S3 as the durable system of record
- Data access splits between direct S3 reads and file layer access, depending on workload characteristics
- Changes are synchronized asynchronously between the file system and S3
- Pricing reflects a behavior-driven consumption model, with costs tied to throughput, working set size, and S3 API activity
The News
AWS announced S3 Files, a new capability that allows customers to access S3 buckets as file systems using NFS. The service provides a fully managed, elastic file system layer built on AWS infrastructure, enabling applications to interact with S3 data using standard file interfaces without requiring application changes or data migration. S3 Files integrates with AWS compute services including EC2, EKS, ECS, Fargate, and Lambda, providing a consistent file-based access model for cloud-native workloads. For more information, read the official company news blog.
Analyst Take
Enterprise data architectures have converged on object storage, and increasingly S3 specifically, as the primary system of record due to its scalability, durability, and cost profile. At the same time, most applications, pipelines, and AI systems continue to operate through file system semantics. This creates a structural mismatch. Execution requires staging data into file systems, introducing duplication, pipeline complexity, and cost.
S3 Files addresses this by inserting a managed file layer directly on top of S3. Instead of moving data into a separate system, the service applies file system semantics to data that remains in object storage. The file system becomes a working surface that materializes data on demand and synchronizes changes back to S3. This reinforces S3 as the authoritative data layer, with all other access models becoming derived views of that system.
The S3 Files model focuses performance and cost on access behavior. Data that is actively used is materialized in the file layer and accessed with low latency, while the full dataset remains in S3. Cost aligns to access patterns, with SSD pricing applied only to the active working set and additional charges driven by throughput and API activity. Importantly, the service includes collision handling during synchronization, with the ability to preserve conflicting changes through versioning or by placing objects into a recovery namespace.
S3 Files does not replace existing file systems. It introduces a model where S3 remains authoritative and file semantics are applied dynamically. EFS continues to serve workloads where file systems define the system of record and strict POSIX behavior is required. FSx for Lustre remains the platform for high-performance training workloads where throughput and parallelism dominate. S3 Files applies to environments where S3 is already the source of truth and file-based access is required without introducing additional systems.
In our view, the broader implication is the evolving role of storage in the era of AI. Storage becomes part of the execution path. In AI environments, this aligns with the emphasis on inference, retrieval, and agent-based workflows that require flexible data access and iteration rather than only bulk throughput.
What Was Announced
S3 Files is introduced as a new AWS resource that enables S3 buckets to be mounted as POSIX-compliant file systems using NFS. Applications can access data using standard file operations without modification, while S3 remains the underlying storage layer.
The service creates a file system projection of an S3 bucket by mapping object keys into directories and files. On first access, S3 Files materializes metadata for objects within the accessed namespace, enabling low-latency directory traversal. Small files, typically under 128 KB, may be loaded into the file layer during directory access to optimize interactive performance, while larger objects can be streamed directly from S3 without full materialization.
Writes occur in the file layer and are synchronized back to S3 asynchronously. Synchronization aggregates updates before persisting them to S3, where object versioning can be used to maintain historical state.
Policy controls define how data is materialized and retained in the working set. These include triggers such as directory read or file read, file size thresholds, and retention periods based on last access. On-demand import through CLI enables explicit prefetching for known workloads. S3 Files integrates with EventBridge to emit events for file and object changes, enabling downstream workflows for monitoring, automation, and data pipeline coordination.
S3 Files introduces access points that segment the file system namespace into isolated entry points mapped to specific directory paths. These access points integrate with IAM and POSIX permissions, with ownership and access controls stored as object metadata. This model supports up to 10,000 independent workspaces, enabling large-scale multi-tenant and agent-based environments.
The pricing model includes SSD-based charges for the materialized working set, throughput-based charges for reads, writes, and synchronization measured per gigabyte, and standard S3 storage and API charges. Automatic optimization includes eviction of data from the working set based on last access and aggregation of updates during synchronization to S3. Data in the file layer expires based on last access, allowing the working set to contract as workloads shift.
S3 Files is now generally available in 34 AWS Regions.
Looking Ahead
S3 Files is likely to gain traction in environments where S3 already serves as the primary data platform and where file-based interaction remains necessary. This includes analytics workflows, data science environments, and AI systems that require iterative access to subsets of large datasets.
Access points enable per-agent workspaces with controlled access to shared data, supporting emerging agent-based architectures. Each agent can operate within an isolated namespace while interacting with a common dataset, allowing execution to scale without duplicating data or building custom isolation mechanisms.
In AI factories, workload segmentation will remain important. Training workloads will continue to rely on FSx for Lustre for throughput and parallelism, while inference and agent workflows benefit from the flexible access model S3 Files provides. This introduces a separation between performance-optimized training environments and access-optimized execution environments.
Enterprise adoption will depend on workload requirements and operational priorities. Environments that require strict file system semantics, deterministic behavior, and native replication may continue to use EFS or ONTAP-based systems. Workloads that benefit from direct access to object storage without staging will align more closely with S3 Files.
Cost management becomes a function of workload behavior. The size of the active working set and the volume of data movement determine total cost. Stable access patterns produce predictable outcomes, while dynamic or exploratory workloads introduce variability in both cost and performance.
S3 Files extends S3 from a system of record into an execution surface, reinforcing its position as the central data layer across analytics, AI, and application workflows.This adds a new option for enterprises aligning storage architecture with AI and data-intensive applications, where data access patterns increasingly define system behavior.
Don Gentile | Analyst-in-Residence -- Storage & Data Resiliency
Don Gentile brings three decades of experience turning complex enterprise technologies into clear, differentiated narratives that drive competitive relevance and market leadership. He has helped shape iconic infrastructure platforms including IBM z16 and z17 mainframes, HPE ProLiant servers, and HPE GreenLake — guiding strategies that connect technology innovation with customer needs and fast-moving market dynamics.
His current focus spans flash storage, storage area networking, hyperconverged infrastructure (HCI), software-defined storage (SDS), hybrid cloud storage, Ceph/open source, cyber resiliency, and emerging models for integrating AI workloads across storage and compute. By applying deep knowledge of infrastructure technologies with proven skills in positioning, content strategy, and thought leadership, Don helps vendors sharpen their story, differentiate their offerings, and achieve stronger competitive standing across business, media, and technical audiences.