Kioxia America has announced the integration of its AiSAQ approximate nearest neighbor search (ANNS) technology into Milvus, starting with version 2.6.4. Milvus, developed by Zilliz, is a widely used open-source vector database supporting artificial intelligence (AI) search, retrieval-augmented generation (RAG), and multimodal workloads. This integration is designed to help data center operators and technology vendors scale AI workloads by alleviating dynamic random-access memory (DRAM) limitations and supporting high-volume vector queries on cost-effective solid-state drives (SSDs).
According to Kioxia, as organizations develop more complex RAG pipelines and AI inference systems, DRAM scalability can become a significant barrier, particularly as the number of embeddings grows into the billions or trillions. AiSAQ addresses this by enabling SSD-optimized vector indexing, which significantly reduces DRAM usage while maintaining high-quality vector search performance. This capability is intended to lower infrastructure costs and simplify scaling large data repositories in AI-centric data center environments.
AiSAQ is open-source software designed to improve vector scalability by storing all elements of RAG databases on SSDs. The technology offers configuration options that allow users to balance between performance and high-volume capacity, supporting ongoing advancements towards trillion-vector scale deployments. Kioxia reports that this approach makes large-scale RAG deployments more accessible for hyperscale operators and technology vendors.
James Luan, VP of Engineering at Zilliz, stated, “Kioxia’s AiSAQ integration expands the range of indexing options available within Milvus and gives Milvus users another powerful way to scale AI retrieval cost-effectively,” adding, “As AI workloads grow to an extremely large number of embeddings, optimizing memory cost becomes essential. AiSAQ further enhances the support of SSD-optimized vector search in the Milvus ecosystem, enabling developers to scale their AI applications and retrieval pipelines.”
Rory Bolt, Senior Fellow, Software at Kioxia America, said, “AI is shifting from building massive foundation models to deploying scalable, cost-effective inference solutions that solve real-world problems,” and continued, “RAG is central to that shift, and AiSAQ was created to help the community take full advantage of SSD-based vector architectures. The integration with Milvus strengthens the open-source ecosystem and supports developers working to build faster, more efficient AI applications.”
More information and access to the open-source AiSAQ software is available on GitHub. Details on Milvus integration can be found in the Milvus documentation.
Source: Kioxia America
