<p>AWS launched S3 Files, a new feature that lets you mount any S3 bucket or prefix as a shared NFS v4.1/4.2 filesystem on EC2, ECS, EKS, Lambda, and Fargate — no migration required.</p>
<h2>What it actually is</h2>
<p>Previous attempts to treat S3 like a filesystem (s3fs-fuse, goofys, Mountpoint for S3, S3 File Gateway) were essentially hacks layered on top of the S3 API. S3 Files takes a different approach: it’s built on Amazon EFS infrastructure, with S3 as the durable backing store. Each layer does what it’s good at.</p>
<p>The architecture uses a two-tier caching model. Hot data sits in high-performance EFS storage with roughly 1ms latency. Large sequential reads (128 KB+) bypass the cache entirely and stream directly from S3 at up to 3 GB/s per client. Changes made through the filesystem are aggregated and committed back to S3 as atomic PUTs roughly every 60 seconds, keeping S3 as the source of truth.</p>
<p>This “stage and commit” design is what makes full POSIX semantics possible — in-place edits, file locking, directory renames — things that are genuinely hard to fake over a pure object API.</p>
<h2>Setup</h2>
<p>Create a file system in the S3 console, discover mount targets in your VPC, then mount with a standard mount command. Multiple compute resources can mount the same bucket concurrently without duplicating data. Security is handled through IAM access control, TLS 1.3 in transit, and SSE-S3 or KMS encryption at rest. CloudWatch and CloudTrail monitoring are supported.</p>
<h2>Pricing</h2>
<p>Pricing mirrors EFS Performance-optimized mode: $0.30/GB-month for hot storage, $0.03/GB for reads, $0.06/GB for writes. A few things worth knowing:</p>
<ul>
<li>You only pay the filesystem surcharge on the hot slice you actually touch</li>
<li>Large files (128 KB+) stream directly from S3 with no S3 Files charge</li>
<li>There’s a 32 KB minimum per operation</li>
<li>Sync costs can triple effective read costs for small files</li>
</ul>
<p>Compared to EFS directly, S3 Files wins on large-file reads (free vs. $0.03/GB) and cold storage tiering (free with S3 Intelligent-Tiering vs. $0.01–$0.03/GB per EFS tier transition). EFS wins on small-file first-read cost.</p>
<h2>Who it’s for</h2>
<p>The clearest use cases are ML training pipelines that need shared access to datasets already in S3, agentic AI workloads that require concurrent reads and writes across multiple compute resources, and legacy POSIX applications currently running on EFS that would benefit from S3’s storage economics.</p>
<p>S3 Files is available in all commercial AWS regions now. It joins S3 Tables and S3 Vectors as part of AWS’s broader push to make S3 a multi-paradigm data platform rather than a pure object store.</p>


Collections

AWS launched S3 Files, a feature that mounts any S3 bucket or prefix as a shared NFS v4.1/4.2 filesystem on EC2, ECS, EKS, Lambda, and Fargate. Unlike previous workarounds (s3fs-fuse, goofys, Mountpoint for S3), it is built on Amazon EFS infrastructure with S3 as the durable backing store. A two-tier caching model keeps hot data in EFS at ~1ms latency while large sequential reads stream directly from S3 at up to 3 GB/s. A 'stage and commit' design enables full POSIX semantics including in-place edits, file locking, and directory renames. Pricing mirrors EFS Performance-optimized mode at $0.30/GB-month for hot storage, with large-file reads bypassing S3 Files charges entirely. Key use cases include ML training pipelines, agentic AI workloads needing concurrent access, and legacy POSIX apps on EFS seeking S3 storage economics. Available now in all commercial AWS regions.

How S3 Files adds real file system access to S3 buckets