
How to pick AWS storage (without reading every doc)
Five questions to pick the right AWS storage for your app, including AI and agent workloads.
How to make AWS storage decisions (without reading every docs page)
If you've ever needed to save data on AWS and had no idea which storage to pick, I feel you. There are a lot of choices, and it's completely okay not to know what they all do. Luckily for you, you don't need to. You just need to answer five questions, in order.
Step 1: How does your app get to the data?

Every AWS storage service fits into one of three groups, and each group matches a way your app reaches its data:
- Object storage (Amazon S3): your app talks to it over an API.
- File storage (Amazon EFS, Amazon FSx, S3 Files): your app opens files by path, over a file protocol like NFS or SMB.
- Block storage (Amazon EBS): it attaches to a server like a disk.
If you already know how your app reads and writes data, you've pretty much cut your options down to one group.
Step 2: What shape is your data?
The shape of your data narrows things down before you ever look at speed or price.
Your data | Where it fits |
|---|---|
Images, video, logs, backups | S3 general purpose buckets |
Tables you query with SQL | S3 Tables |
Vector embeddings for similarity search | S3 Vectors |
Files in folders that apps open by path | EFS, S3 Files or FSx |
Raw disk for a database or boot drive | EBS |
Step 3: Is this a new app or one you already have?
If it's a new app, go with what you found in steps 1 and 2. Services where you don't have to set up storage space ahead of time, like S3, EFS and S3 Files, mean less work for you later.
If you're moving an app you already have, the system it runs on today usually tells you where to go:
Your app runs on | Move it to |
|---|---|
NetApp ONTAP | FSx for NetApp ONTAP |
ZFS | FSx for OpenZFS |
Windows File Server | FSx for Windows File Server |
Lustre | FSx for Lustre |
Apache Hive tables on S3 | S3 Tables (it can convert them to Apache Iceberg for you) |
To move the data itself, use AWS DataSync for online transfers, or AWS Data Transfer Terminal for physical transfers.
Step 4: Do two kinds of apps need the same data?
This is the part I didn't know before. Sometimes your data lives in one kind of storage, but an app needs to reach it a different way. Say your data is in an S3 bucket, but one of your tools can only open files from a folder. You don't have to copy the data over. A couple of features give that tool its own way to reach the same data:
Your data lives in | Your app also needs | Use |
|---|---|---|
An S3 bucket | File access over NFS | S3 Files |
FSx for NetApp ONTAP or OpenZFS | Access through the S3 API, without copying data | S3 Access Points for FSx |
So when do you pick S3 Files over EFS? Choose S3 Files when your data already lives in S3 and also needs to be opened by path. Choose EFS when the file system is the main home for your data.
Step 5: Building with AI or agents?
Match storage to each step of the work:

- Training data: S3, S3 Files, EFS, or FSx for Lustre when lots of servers read at once.
- Fast training and running models: S3 Express One Zone, for single-digit millisecond reads on data you use a lot.
- RAG and semantic search: S3 Vectors. You can also pick it as the vector store for Amazon Bedrock Knowledge Bases.
- Agent memory: S3 Vectors, so an agent can save embeddings and pull back related context.
- Helping agents find data: S3 Metadata keeps an always-updated table of everything in a bucket for you, so you don't have to build one. S3 annotations add business context so agents can find the right files.
- Agents that expect a file system: EFS or S3 Files. Amazon EC2, AWS Lambda, Amazon ECS and Amazon EKS can all mount S3 Files.
And always, how much is this going to cost?
Once you're down to one or two options, compare price. Each service has its own pricing page: S3 , EFS , EBS and FSx . To guess your monthly bill, plug your numbers into the AWS Pricing Calculator .
If you don't know how often your data gets read, S3 Intelligent-Tiering moves it between access tiers for you, and S3 Tables supports those tiers for table data too.
The quick version
If your app... | Look at |
|---|---|
Talks to data over an API | S3 |
Opens files by folder path | EFS, S3 Files or FSx |
Needs a disk attached to a server | EBS |
Stores tables you query with SQL | S3 Tables |
Stores embeddings for similarity search | S3 Vectors |
Needs single-digit millisecond reads | S3 Express One Zone |
Has data in S3 but needs to open it as files | S3 Files |
Runs on NetApp ONTAP, ZFS, Windows File Server or Lustre today | The matching FSx |
Where to go next
Did any of this not make sense at first? It didn't for me either, until I sat crying underneath a pile of documentation. The good news is you only need step 1 to get started, and the rest follows.
If you build with a coding agent, the aws-storage skill in the Agent Toolkit for AWS gives your agent these same rules, so it picks the right storage for you too.
Enjoyed reading this content? Let the author know!
Your likes, comments, shares, and saves help creators reach more builders.
Loading recommendations
Loading article