"AWS has officially launched Amazon S3 Vectors into general availability, unlocking native vector storage and search inside S3 at massive scale and dramatically lower costs."
Amazon Web Services has taken another big step into the world of AI infrastructure with the general availability (GA) of Amazon S3 Vectors, announced during AWS re:Invent 2025. This feature transforms the world’s most popular object storage into a native vector database capable of storing and querying embeddings directly inside S3. For developers building Retrieval-Augmented Generation (RAG), search engines, and AI agents, this is a major shift.
What exactly is Amazon S3 Vectors?
S3 Vectors adds native vector indexing and similarity search to S3. Instead of relying on a separate specialized database, teams can store embeddings alongside their existing datasets and search them directly using S3 APIs.
A single index can handle up to 2 billion vectors, while each bucket can support as many as 20 trillion vectors across 10,000 indexes. Query performance typically averages 100 ms for frequent searches and stays under one second for infrequent queries, returning up to 100 results at a time.
Core capabilities
- Dedicated vector buckets for durable embedding storage.
- Server-side encryption using SSE-S3 or SSE-KMS at the index level.
- Metadata filtering on up to 50 keys (10 non-filterable).
- Object tagging for billing, cost separation, and access control.
- Write throughput of up to 1,000 PUT operations per second.
- Region availability across 14 AWS regions at launch.
How well does it perform?
AWS shared impressive real-world preview results before GA:
- Over 250,000 vector indexes created.
- 40+ billion vectors ingested.
- 1 billion similarity queries processed.
These metrics highlight that S3 Vectors has already been tested at massive scale, making it suitable for large enterprise AI workloads such as recommendation engines, document understanding, and background RAG pipelines.
How does pricing compare to other vector databases?
AWS claims that S3 Vectors delivers up to 90% cost savings compared to most specialized vector databases thanks to S3’s pay-per-use model.
- Data ingestion: charged per logical GB uploaded.
- Storage: priced on total logical vector size.
- Queries: billed per API request plus a tiered cost per TB scanned.
Below are sample monthly cost comparisons published by AWS and industry analysts:
| Vectors Stored | Monthly Queries | Estimated S3 Cost |
|---|---|---|
| 100 million | 1 million | $11.38 |
| 400 million | 10 million | $1,217 |
Competitors such as Pinecone and Weaviate typically charge significantly higher at similar volumes, especially once query traffic increases.
Does S3 Vectors replace vector databases?
Short answer: no, but it changes the economics.
Most analysts describe S3 Vectors as complementary rather than a full replacement for ultra-low-latency vector databases. It excels in:
- Cost-sensitive workloads
- Background processing for AI agents
- Batch or asynchronous RAG pipelines
Workloads that require sub-10 ms latency and extremely high queries-per-second (QPS) will likely continue to depend on purpose-built vector engines. However, S3 Vectors places strong competitive pressure on independent vendors by removing the need for an additional data store for many use cases.
Real-world use cases
Early adopters include companies such as BMW (recommendation systems) and Twilio (semantic search and customer support workflows).
The feature integrates directly with:
- Amazon Bedrock Knowledge Bases for building native RAG systems.
- Amazon OpenSearch Service for hybrid keyword + vector search.
“S3 Vectors makes vector search a built-in capability of object storage rather than another database your team must operate.”
Simple example usage
Developers interact with S3 Vectors using familiar AWS APIs:
aws s3vectors create-index --bucket my-vector-bucket --name product-embeddings
aws s3vectors query --index product-embeddings --vector file://query.json
This keeps AI workloads inside the same infrastructure stack many teams already depend on.
FAQs
Is S3 Vectors suitable for real-time applications?
It works well for many interactive use cases, but applications requiring extremely low latency may still benefit from specialized vector databases.
Do I need to move my existing vectors to S3?
No, but to use S3 Vectors you must ingest embeddings into vector buckets designed for indexing and querying.
Does S3 Vectors support hybrid search?
Yes. When paired with Amazon OpenSearch, teams can combine traditional keyword queries with vector similarity search.
Final thoughts
The general availability of Amazon S3 Vectors means AWS now provides a native, massively scalable, and low-cost vector solution directly inside object storage. While high-performance vector databases will still serve critical real-time scenarios, S3 Vectors lowers the barrier to entry for most AI workloads and could reshape the economics of retrieval-based AI architectures.
