AWS launches S3 Vectors with 90% cost savings claim, scales to 2 billion vectors per index

AWS S3 Vectors reaches general availability with dramatic scale increase

AWS has officially launched Amazon S3 Vectors to general availability, transforming its widely-used object storage service into a native vector search capability that could reshape how organizations handle AI applications 1

1

. The service now supports up to 2 billion vectors in a single index, a massive jump from the 50 million vector limit during its July preview phase 1

1

. Organizations can now consolidate up to 20 trillion vectors per S3 storage bucket, eliminating the need to fragment datasets across multiple infrastructure components.

The preview period demonstrated substantial enterprise appetite for integrated vector storage. Customers created more than 250,000 vector indexes and ingested over 40 billion vectors in just four months 1

1

. This adoption pattern suggests organizations are eager to consolidate their AI infrastructure rather than maintain separate specialized systems. Andy Warfield, vice president and distinguished engineer at AWS, noted that S3 Vectors quickly became the most popular S3 preview to date 2

2

.

Query performance improvements target production workloads

AWS refined query latency based on customer feedback during the preview phase. The service now delivers approximately 100 milliseconds or less for frequent queries, with infrequent queries completing in under one second 1

1

. Write performance supports up to 1,000 PUT transactions per second for single-vector updates, while maximum search results per query increased from 30 to 100 results.

These performance enhancements address the core requirements organizations need to query vector embeddings in production environments. Mai-Lan Tomsen Bukovec, VP of technology at AWS, explained that customers specifically requested higher scale and lower latency so they could use Amazon S3 as a primary vector store for their rapidly expanding vector storage needs 1

1

. The improvements enable use cases including hybrid search, semantic search, retrieval-augmented generation (RAG), and AI agent workflows without moving data to specialized infrastructure.

Cost savings claim challenges specialized vector databases

AWS claims organizations can reduce the total cost of storing and querying vectors by up to 90% compared to specialized vector database solutions 1

1

. This aggressive cost positioning directly challenges standalone vector databases like Pinecone and other purpose-built solutions that emerged as essential infrastructure during the early generative AI era. The cost-efficiency of AI workflows becomes particularly compelling as organizations scale to billions of vectors.

Despite the dramatic cost savings claim, AWS positions S3 Vectors as complementary to vector databases rather than a direct replacement. Bukovec described the approach as "performance tiering" based on application requirements 1

1

. Applications requiring ultra-low latency response times still benefit from dedicated vector databases like Amazon OpenSearch, while retrieval-augmented applications with more flexible latency requirements can leverage S3 Vectors' economic advantages.

AI agents drive demand for scalable vector storage

The rise of AI agents is fundamentally reshaping storage requirements for enterprise AI workflows. Warfield emphasized that 2024 marks the first year real productive work has started happening with agents, bringing developers back into hands-on work that fits naturally into busy workflows 2

2

. These agents depend on richer context and scalable data access, creating demand for technologies that bridge model capabilities with organizational data.

"The big bridge that is on us to solve is the bridge between all of that kind of capability and the data that people have," Warfield explained 2

2

. S3 Vectors addresses this by allowing customers to store and query vector embeddings directly within their existing data lake infrastructure. Organizations with mature S3 deployments can now build applications that are "soup-to-nuts S3-based," combining object storage, structured tables, and vector search in a unified platform 2

2

.

Real-world applications span from video intelligence to drug development

Early adopters are applying S3 Vectors across data-intensive industries. March Networks uses the service for large-scale video and photo intelligence, storing billions of vector embeddings economically thanks to the service's storage economics and latency profile 1

1

. In life sciences, organizations are leveraging vector storage for drug development and radiology applications 2

2

.

The variety of use cases suggests vector search is becoming a fundamental data operation rather than a specialized AI technique. Bukovec drew parallels to how enterprises use data lakes today, with transactional databases like Amazon Aurora handling certain workloads while S3 provides application storage and analytics 1

1

. Organizations should watch whether this pattern holds as vector workloads mature, particularly around indexing strategies and the balance between cost savings and latency requirements for different AI applications. The scalability improvements also raise questions about how purpose-built vector databases will differentiate as cloud providers integrate vector capabilities into core storage services.