S3 Files: File Access on S3

For years, Amazon Simple Storage Service was the cloud’s giant warehouse shelf: cheap, durable, and great for storing objects, but awkward if your software expected folders, file locks, and ordinary read-write file calls. On April 7, 2026, Amazon Web Services added Amazon S3 Files so the same bucket can now be mounted and used like a shared file system. (aws.amazon.com) That changes a boring but expensive habit inside cloud systems: copying data out of a bucket into a separate file store just so an application can open it with normal file commands. Amazon says S3 Files lets compute services work with data “without your data ever leaving S3.” (docs.aws.amazon.com) A file system is the thing behind everyday actions like opening a folder, renaming a file, or locking a document so two people do not overwrite each other at once. S3 Files adds those file system semantics on top of bucket data, including reading, writing, directories, and locking files. (docs.aws.amazon.com 1) (docs.aws.amazon.com 2) Amazon built this using Amazon Elastic File System, which is its managed network file system, and then linked that layer to bucket data underneath. The result is a shared file system with mount targets inside a virtual private cloud, so applications reach it over the network like other managed file systems on Amazon Web Services. (docs.aws.amazon.com 1) (docs.aws.amazon.com 2) Amazon is pitching speed hard because the old object-storage path often forced developers to redesign software around storage application programming interfaces instead of file calls. The launch post says S3 Files gives compute resources access with about 1 millisecond latency and keeps changes synchronized back to the bucket. (aws.amazon.com) The first audience is not every database on earth. Amazon’s own examples center on file-based applications, artificial intelligence agents, machine learning training jobs, and teams that already keep large data sets in buckets but still need ordinary file tools. (aws.amazon.com) (geekwire.com) That is why this arrived next to Amazon’s recent push to make Simple Storage Service more than a dump truck for raw files. Amazon has also been adding S3 Tables for Apache Iceberg analytics data and S3 Vectors for storing and querying vector embeddings used in search and artificial intelligence systems. (aws.amazon.com 1) (aws.amazon.com 2) The appeal is easy to see in container systems. Amazon Elastic Container Service already documents S3 Files as a volume type, which means a container can mount bucket data the same way it mounts other shared storage instead of running a custom copy step before startup. (docs.aws.amazon.com) The risk is that a cleaner interface can hide the old physics. A bucket-backed shared file system may feel like a local disk, but it still sits on networked infrastructure with synchronization rules, mount targets, and a design optimized for shared access to data already living in Simple Storage Service, not for every latency-sensitive or coordination-heavy workload. That tradeoff is visible even in Amazon’s docs, which describe caching only a fraction of actively used data as files and directories for low-latency access. (docs.aws.amazon.com) So the real shift is not that Amazon turned buckets into magic disks. It turned one of cloud computing’s most common storage copies into an optional step, which will save time and complexity for many pipelines, while tempting some teams to use one general-purpose layer where a purpose-built file system or database still fits better. (aws.amazon.com) (techzine.eu)