Elastic 9.3 Uses GPUs to Index Vectors 12x Faster

March 16, 2026

Elastic 9.3 Uses GPUs to Index Vectors 12x Faster

Published: March 16, 2026 at 12:56 AM

Updated: March 16, 2026 at 12:56 AM

100-word summary

Elastic 9.3 taps NVIDIA's cuVS library to run vector indexing on GPUs, hitting 12x faster throughput and 7x quicker force merges in technical preview. Translation: building those massive vector databases for RAG apps just became dramatically faster. The release also ships an AI assistant that reads your logs, writes database queries from plain English, and suggests fixes when things break. It's available now on self-managed setups, with cloud users able to offload the GPU work to Elastic's managed infrastructure. The catch? GPU indexing is still preview-stage, so hardware limits and stability questions remain for serious workloads.

What happened

Elastic 9.3 taps NVIDIA's cuVS library to run vector indexing on GPUs, hitting 12x faster throughput and 7x quicker force merges in technical preview. Translation: building those massive vector databases for RAG apps just became dramatically faster. The release also ships an AI assistant that reads your logs, writes database queries from plain English, and suggests fixes when things break. It's available now on self-managed setups, with cloud users able to offload the GPU work to Elastic's managed infrastructure. The catch?

Why it matters

GPU indexing is still preview-stage, so hardware limits and stability questions remain for serious workloads.

Sources