Amazon S3 Vector Search

A 2025 research paper presenting a GPU-driven asynchronous I/O framework for billion-scale approximate nearest neighbor search. The system addresses the fundamental bottleneck of data movement between storage and compute in large-scale vector search.

Strategies for reducing vector database costs including quantization, dimension reduction, efficient indexing, storage tiering, and choosing cost-effective deployment options.

Amazon Aurora Serverless v2 is a cloud-hosted, serverless relational database (Postgres/MySQL compatible) with pgvector support for managed vector workloads, featuring auto-scaling compute/memory, pay-per-use pricing, automated backups, and multi-AZ/multi-region high availability. Suited for enterprise RAG via Amazon Bedrock Knowledge Bases and production AI apps. Provides easier operations than self-hosted Milvus or Postgres, deeply integrated with AWS unlike standalone Zilliz.

Amazon Aurora Machine Learning provides managed vector storage and search capabilities integrated into Aurora PostgreSQL for AI workloads on AWS. Key features include serverless scaling, direct ML model calls via SQL for embeddings, and seamless integrations with Bedrock and SageMaker. Perfect for RAG pipelines and enterprise AI applications, it simplifies vectorization and abstracts infrastructure compared to self-hosted options like Milvus.

Vector search extension for Amazon ElastiCache for Redis, featuring HNSW indexing for k-NN similarity, hybrid lexical+vector search with BM25 fusion capabilities. Used for enterprise semantic caching, real-time recommendations, and RAG applications. Integrated Redis module offers sub-microsecond latency vs standalone like Weaviate, optimized for hot data workloads.

Milvus Distributed is the cluster mode of the scalable open-source vector database for AI embeddings search, supporting HNSW, IVF, and NGT indexes in high-availability distributed setups. It provides GPU support, billion-scale capacity, real-time upsert/query capabilities, and multi-modal vector handling. Suited for RAG, recommendations, and image/video search at enterprise scale. Self-hosted unlike Pinecone's managed offering, and more ANN-centric than Weaviate.