Amazon S3 is AWS's managed object storage service — the de facto substrate for data lakes and lakehouses, with tiered storage classes and fine-grained access control.
Nothing to operate: a fully managed regional service with versioning, lifecycle policies, and IAM- and policy-based access control down to prefix level. Cross-region replication is opt-in, with a 15-minute RTC option. S3 on Outposts extends the API on-premises. It is the reference implementation the entire S3-compatible ecosystem measures itself against.
How Amazon S3 answers the questions Object Storage turns on.
| How it works | |
| S3 API | Native: the reference implementation |
| Durability model | Eleven nines annual across three or more AZs; one AZ on the One Zone classes |
| Event notifications | To SNS, SQS, Lambda or EventBridge, one destination per rule |
| Running it | |
| Minimum duration | None on Standard; 30 days IA, 90 Glacier, 180 Deep Archive |
| Multi-region | Cross-region or same-region replication; RTC puts 99.99% inside 15 minutes |
| Minimum deployment | None: fully managed; S3 on Outposts for on-premises |
| Connections | |
| Interfaces | Object only |
| Access | |
| Immutability | Object Lock: governance or compliance, plus legal hold; SEC 17a-4 |
| Cost | |
| Egress | Charged to internet above 100 GB free a month; free in-region, waived on exit |
| Storage tiers | Standard, Intelligent-Tiering, two IA tiers, Express One Zone, three Glacier tiers |
vs Amazon S3: Open source (permissive) · Self-hosted · Free · Operational complexity: High · Java
vs Amazon S3: Managed
vs Amazon S3: Managed · Free · Subscription