Powering enterprise AI at scale: The Elastic and NVIDIA cuVS integration

Enterprises are entering a new phase of AI adoption, where the proliferation of unstructured data has made traditional search inadequate for extracting insights. As a result, organizations are turning to high-performance vector databases and semantic search to power modern AI applications, such as generative AI (GenAI), retrieval augmented generation (RAG), and AI agents. 

Meanwhile, organizations are building AI factories to simplify AI deployment, scale performance, and maximize GPU efficiency and capacity. The integrated stack which combines accelerated compute with high-performance vector search is the most economically efficient maxim to push frontier AI to your consumers and the enterprise.

When large enterprises build internal generative AI platforms that must index and retrieve context from petabytes of corporate data efficiently, costs immediately come to the forefront. These AI factories need every advantage possible to reduce expensive workloads.

Similarly, organizations looking to scale their vector search capabilities struggle to do so without linearly increasing their CPU hardware spend. When you set out to build a high-performing vector database, you face the challenge of constructing the vector index (HNSW graph) on the CPU. When comparing every vector, index building can explode into millions, or even billions, of arithmetic operations. And with that complexity comes ingestion bottlenecks. Add in index lifecycle operations like compaction and merges, and your compute overhead can increase significantly.

To help you overcome such challenges, Elastic and NVIDIA together are enabling the Elastic AI Ecosystem, redefining how enterprises build and scale AI factories by delivering GPU-accelerated vector search and powerful AI infrastructure that unlocks next generation, real-time intelligence.

By combining the NVIDIA Enterprise AI Factory validated design with Elasticsearch’s powerful vector database, organizations can deliver deeper insights and real-time, relevant data to AI agents and GenAI applications securely and at enterprise scale. 

With Elastic and NVIDIA, you get the power of a high-performing, feature-rich vector database with GPU acceleration built for modern AI.

Reduce time-to-market by using a full-stack, pre-engineered system for building on-premises AI factories. Elasticsearch is a validated and supported vector database within the NVIDIA Enterprise AI Factory validated design, ensuring a reliable framework for deploying agentic AI applications.

Handle exponentially growing vector embeddings and massive data volumes more efficiently. Integrating NVIDIA cuVS into Elasticsearch delivers a nearly 12x improvement in indexing throughput and a 7x increase in faster force merging. 

Elastic’s NVIDIA-accelerated vector search removes performance bottlenecks, allowing your infrastructure to handle massive query volumes with near-instant response times so that you can scale generative AI and RAG applications efficiently while delivering real-time, relevant insights.

Maintain a high-performance, proven-at-scale vector database without the burden of constant manual optimization. NVIDIA cuVS optimizes and accelerates vector search operations across evolving NVIDIA GPU architectures and CUDA versions, ensuring peak performance and seamless scalability for AI and data-intensive workloads.

Eliminate the complexity of managing external inference providers and infrastructure. Elastic Inference Service (EIS) provides native, high-throughput inference directly within Elasticsearch, using modern NVIDIA AI infrastructure to maintain low-latency performance at scale. Deploy production-ready managed models, including Jina AI’s multilingual embedding and reranker models, natively to ensure precise recall tuning and data sovereignty without extra operational overhead.

Elastic and NVIDIA collaborated to deliver more speed, scale, and value to enterprises that rely on vector search to build and deliver real-time RAG and AI applications. Your team can provide performance at scale with less strain on your infrastructure and your budget. 

With Elasticsearch included in the NVIDIA AI Factory validated design, you get a proven, full-stack, pre-engineered blueprint for accelerating AI applications, meaning  more efficiency and faster time to market. 

The NVIDIA cuVS integration is currently in technical preview for Elastic self-managed enterprise customers (version 9.3). General availability is scheduled for April 2026 with the version 9.4 release, which will provide a production-ready foundation for high-volume vector search and real-time context retrieval.

Learn more about the expanding Elastic AI Ecosystem, download the Elastic AI Ecosystem Developer Guidebook, or connect with Elastic technical experts about high-performance AI at NVIDIA GTC. Visit us at booth #3200, watch live demos, and speak with Elastic experts.