Databricks AI Search Adds High-QPS Scaling, Moving From Prototype to Production
Source: Databricks Blog — 2026-07-28
Summary
Databricks announced general availability of high-QPS (queries-per-second) scaling for its AI Search endpoints, aimed at teams moving retrieval-augmented generation and search workloads from prototype into real production traffic. The headline feature is a target_qps setting that lets teams declare the throughput they need and have infrastructure provisioned automatically to match, rather than manually sizing and scaling search infrastructure themselves.
Key Takeaways
- The new
target_qpssetting lets a team specify a target query throughput and have Databricks provision the underlying infrastructure to support it automatically, instead of manual capacity planning. - This is explicitly framed as a prototype-to-production gap-closer — a signal that plenty of teams have been hitting a wall when their AI Search-based RAG systems needed to scale past initial pilot traffic.
- The change targets the search/retrieval layer specifically, which is often the actual bottleneck in production RAG systems even when the LLM serving side has already been scaled.