Elastic and NVIDIA together unlock next generation enterprise AI search

Elastic and NVIDIA have integrated NVIDIA cuVS GPU acceleration into Elasticsearch's vector indexing pipeline, delivering up to 12x faster indexing throughput and 7x faster force merging compared to CPU-based approaches. On a cost-adjusted basis, GPU acceleration provides approximately 5x higher indexing throughput. Elasticsearch is now a validated vector database within the NVIDIA Enterprise AI Factory validated design, offering a pre-engineered blueprint for on-premises AI deployments. The integration also includes native GPU-accelerated inference via Elastic Inference Service (EIS). The feature is currently in technical preview for self-managed enterprise customers on version 9.3, with general availability planned for April 2026 in version 9.4.