What Is Amazon S3, Really? The 2026 Explanation
S3 launched in March 2006 — before the iPhone, before Airbnb, before "the cloud" was a household term.
One Paragraph
S3 = flat key-value object store over HTTP. You give it data + a key (uploads/photo.jpg), it stores it redundantly. You request the key, it returns the data. Everything else — storage classes, encryption, replication — builds on that simple contract.
S3 vs Everything Else
| vs File Systems | vs Block Storage | vs Databases |
|---|---|---|
| Flat namespace (no real folders) | API, not raw disk | Opaque blobs, not structured records |
| Eventually consistent for overwrites | No seekable random access | No query language |
| Immutable objects | Whole-object read/write | Key-based access only |
Pricing: 4 Components
- Storage: $0.023/GB/mo (Standard)
- Requests: $0.005/1K PUTs, $0.0004/1K GETs
- Egress: $0.09/GB after 100GB free
- Data management: Tags, analytics, replication
A busy 10TB bucket: ~$530/month (storage + requests + egress).
Storage Class Ladder
| Class | Cost | When |
|---|---|---|
| Standard | $0.023/GB | Hot data |
| IA | $0.0125/GB | Monthly access |
| Glacier IR | $0.004/GB | Rare but fast retrieval |
| Deep Archive | $0.00099/GB | "Delete never" |
Self-Hosted Alternatives
MinIO, RustFS, Ceph RGW all implement S3 API. Same s3:// URIs, zero egress/request fees on your own hardware.
FAQ
What does S3 stand for?
Simple Storage Service. Despite the name, there's nothing particularly "simple" about the full feature set anymore — it started simple in 2006 and accumulated capabilities (storage classes, replication, encryption, analytics) over 20 years. But the core contract remains elegantly simple: give me bytes + a key, I'll store them. Give me the key back, I'll return the bytes.
Is S3 a file system?
No. S3 is an object store with a flat key namespace. The / characters in keys look like directory separators but aren't — S3 has no real folders, no hierarchical structure, and no POSIX semantics (no hard links, symlinks, permissions, or in-place modification). Tools like aws s3 sync and s3fs present a filesystem-like interface on top of S3, but underneath it's still a key-value store.
How much does S3 really cost?
It depends on your usage pattern, but a typical production workload has 4 cost components: (1) Storage: $0.023/GB/month for Standard, (2) Requests: $0.005 per 1,000 PUTs, $0.0004 per 1,000 GETs, (3) Egress: $0.09/GB after the first 100GB free tier, (4) Data management: tags, analytics, replication. For a 10TB bucket with moderate activity (1M GETs/day, 2TB egress/month), expect $230 storage + $120 requests + $180 egress = ~$530/month. Self-hosted S3 alternatives eliminate components 2-4 entirely.
Can I run my own S3-compatible server?
Yes. Multiple open-source projects implement the S3 API: MinIO (Go, largest community), RustFS (Rust, focused on performance), Ceph RGW (C++, part of Ceph platform), SeaweedFS (Go, simpler architecture). All support the core S3 operations (PutObject, GetObject, ListObjects, Multipart Upload) and can be dropped in as replacements for AWS S3 in most applications by changing the endpoint URL. See our bare metal deployment guide for details.
What is the maximum file size in S3?
5 TB per object. For files larger than 5 GB, you must use Multipart Upload, which splits the file into parts (5 MB to 5 GB each, up to 10,000 parts). This is handled automatically by AWS SDKs, the AWS CLI, and most S3 clients — your application code typically doesn't need to change. The 5 TB single-object limit has existed since S3's launch and hasn't changed.
RustFS — S3-compatible storage without the AWS bill. Download.
Top comments (0)