Databricks launches Lakehouse//RT real-time warehouse beta

Databricks frames Lakehouse//RT as a response to a shift in who generates queries. Human analysts ask questions at human pace, but AI agents can issue hundreds of queries per second around the clock, and customer-facing dashboards and embedded analytics add millisecond latency demands. Traditional warehouses were built for human-scale volume, so the usual answer is to add capacity in large blocks, which means paying for idle compute between spikes.

The company's pitch is that cost should scale with efficiency rather than demand. It lists four mechanisms: higher concurrency so more queries share resources, faster execution so each query holds compute for less time, autoscaling in small increments instead of whole-cluster steps, and a single system rather than a separate serving layer such as Apache Druid, Apache Pinot, or a cache. The last point targets the duplicate infrastructure, second data copy, and sync pipelines that come with bolting real-time serving onto a warehouse.

Databricks positions the launch against a future of query volumes growing by orders of magnitude, arguing that organizations will not be able to outspend the demand curve. The beta and the discount through January 2027 give customers a window to test whether the efficiency claims hold for their own workloads.