S3 Files as a scale case study

For 20 years, Amazon Simple Storage Service was built like a giant warehouse for blobs of data: you stored an object with a name, but you did not open it like a normal file on a shared drive. On April 7, 2026, Amazon Web Services launched Amazon S3 Files to let the same bucket show up as a file system without moving the data somewhere else. (aws.amazon.com) That fixes a boring but expensive problem. Teams that wanted file-style access before usually had to keep one copy in object storage and another copy in a separate file system, then build software to keep the two copies synchronized. (aws.amazon.com) Object storage is like a shipping yard where every package is durable and cheap to keep. A file system is like a shared office cabinet where many workers expect to open, edit, rename, and lock documents in place. (aws.amazon.com) Amazon S3 Files tries to make those two worlds act like one system. Amazon says the bucket stays directly accessible through Amazon Simple Storage Service application programming interfaces while the same data is mounted as a network file system for compute resources. (aws.amazon.com) The scale claim that caught attention came from Werner Vogels, Amazon’s chief technology officer. In an April 2026 post, he wrote that Amazon Simple Storage Service sends more than 300 billion event notifications per day to serverless listeners that process new objects. (allthingsdistributed.com) Those notifications are the plumbing that tells downstream software something changed. Amazon’s own documentation says Amazon S3 Files watches bucket changes through Amazon Simple Storage Service Event Notifications and reflects adds, edits, and deletes back into the mounted file system. (docs.aws.amazon.com) Amazon also says versioning must be turned on for the linked bucket. When a file is edited through the file system, Amazon S3 Files writes the change back as a new object version, and when there is a conflict, the bucket is treated as the source of truth. (docs.aws.amazon.com) That design is why people are using this launch as a scale case study instead of just a product announcement. If a storage system can keep file views lined up with object views while handling hundreds of billions of daily notifications, it is aiming at workloads where data arrives constantly and many machines need to touch the same dataset. (allthingsdistributed.com, aws.amazon.com) Mortgage pipelines are a clean example because they mix huge document stores with file-hungry tools. One loan file can involve scanned forms, PDFs, images, extracted text, underwriting models, and audit logs, and older architectures often copied those artifacts between object buckets and file shares just to satisfy different software. (aws.amazon.com) Amazon is pitching a simpler path: keep the data in Amazon Simple Storage Service once, mount it where file-oriented software expects folders, and let object-oriented services keep using the same bucket. The launch page says thousands of compute resources can connect to the same Amazon S3 file system at the same time without duplicating data. (aws.amazon.com) There are still boundaries. Amazon’s monitoring page says Amazon S3 Files logs management events such as creating file systems and mount targets, but it does not log file read and write data events, which means teams in regulated industries still need to think carefully about audit design. (docs.aws.amazon.com) So the story here is not that Amazon added another storage product. It is that Amazon is trying to turn its biggest object store into a place where file-based software, event-driven software, and large shared datasets can all meet without the usual copy-and-sync machinery in the middle. (aws.amazon.com, aws.amazon.com)