About this session

Most analytical databases were designed for a world of dashboards and scheduled reports. Firebolt started from zero with a different set of assumptions: object storage as the foundation, a single binary that scales from laptop to cluster, Postgres wire compatibility, and sub-100ms latency under concurrent load. This session walks through the architecture decisions behind Firebolt's engine, why we chose to build on Iceberg, and how our design allows one system to handle both real-time serving and batch workloads. We'll show benchmarks against ClickHouse, BigQuery, and Trino, and explain why existing solutions forced customers into painful tradeoffs that a ground-up rethink eliminates.