AWS launches S3 Files for native mounts

For 20 years, Amazon Simple Storage Service was the cheap, durable warehouse of cloud computing, but it was a warehouse where most software could not just open a file, lock it, rename it, and keep going. On April 7, 2026, Amazon changed that by letting a storage bucket show up like a normal file system instead of a pile of web objects. (aws.amazon.com) Object storage and file storage sound similar, but they behave differently in practice. A bucket stores data as separate objects reached through application programming interfaces, while a file system gives programs folders, paths, permissions, and commands like `open` and `rename` that many tools were built around. (docs.aws.amazon.com) That gap created years of awkward workarounds. Teams either copied data out of Amazon Simple Storage Service into services like Amazon Elastic File System, or they used mount hacks that made buckets look like disks but did not fully behave like one. (aws.amazon.com) Amazon S3 Files is Amazon’s answer to that problem. It creates a shared file system linked to a bucket or even just a prefix inside a bucket, so the data stays in Amazon Simple Storage Service while applications see directories and files. (docs.aws.amazon.com) The speed trick is that it does not copy the whole bucket into a new storage system first. Amazon says S3 Files loads metadata and file contents onto high-performance storage on demand, keeps only a fraction of actively used data there, and routes reads to the layer best suited for them. (docs.aws.amazon.com) That is why Amazon is talking about roughly 1 millisecond latency instead of the slower feel people expect from object storage. The fast path is for active files and metadata, while the underlying source of truth remains the bucket. (aws.amazon.com) The other big change is behavior, not just speed. Amazon says S3 Files supports full file system semantics, including consistency, file locking, and Portable Operating System Interface permissions, which is the Unix-style rulebook that decides who can read, write, or execute a file. (docs.aws.amazon.com) That matters because a lot of software breaks on tiny file-system details. A database checkpoint, a machine learning job, or a build tool may depend on rename operations, advisory locks, and permission bits, and those assumptions are exactly where plain object storage usually falls down. (aws.amazon.com) Amazon also wired this directly into its main compute products. The documentation already shows Amazon Elastic Compute Cloud instances mounting S3 Files, Amazon Elastic Kubernetes Service pods using it through mount targets in a virtual private cloud, and Amazon Lambda functions reading and writing through a local mount path. (docs.aws.amazon.com 1) (docs.aws.amazon.com 2) The practical effect is that companies no longer have to build as many copy pipelines just to satisfy file-hungry tools. A training job, an inference service, and an automation agent can point at the same bucket-backed file system instead of each keeping its own duplicate dataset. (aws.amazon.com) Amazon is also framing this as a multi-compute shared layer, not a one-machine convenience feature. The company says one S3 file system can be attached to multiple compute resources, which means clusters can share the same data without moving it around first. (aws.amazon.com) So the real shift is not that Amazon added one more storage product on April 7, 2026. It is that the cloud’s biggest object store now speaks the language of old file-based software, which makes it easier for analytics jobs, artificial intelligence agents, and legacy tools to work where the data already lives. (aws.amazon.com)