Kioxia has integrated its AiSAQ approximate nearest neighbour search software into the open-source Milvus vector database, in a move aimed at lowering the cost and complexity of scaling artificial intelligence applications.
The Japanese memory group’s European arm said the technology would be available to Milvus users from version 2.6.4, allowing developers to run large-scale vector searches primarily on solid-state drives rather than relying heavily on DRAM.
Vector databases underpin many modern AI workloads, particularly retrieval augmented generation systems that combine large language models with external data. However, these systems can be expensive to scale because traditional vector search architectures require large amounts of memory.
Axel Störmann, Vice President and Chief Technology Officer at KIOXIA Europe, said the focus of the AI industry was shifting from training ever-larger models to deploying inference systems that are “scalable and cost-effective”.
“RAG is central to this transition,” he said, adding that AiSAQ had been developed to help the developer community make greater use of SSD-based architectures. Integrating the software into Milvus was intended to simplify adoption within the open-source ecosystem, he said.
AiSAQ was first released as open-source software earlier this year. It is designed to store all RAG-related database elements on SSDs, sharply reducing DRAM requirements while maintaining vector search accuracy. Kioxia argues that this approach addresses a growing bottleneck as AI inference workloads increase in size and volume.
Milvus is one of the most widely used open-source vector databases, and the integration is intended to make SSD-optimised vector search more accessible to enterprises building AI applications at scale.
Kioxia said the collaboration with the open-source community would enable a new class of cost-efficient vector search systems to meet rising demand from generative AI and other data-intensive applications.
There’s plenty of other editorial on our sister site, Electronic Specifier! Or you can always join in the conversation by visiting our LinkedIn page.
