Firebolt

Use it when

• The warehouse serves an application, not a dashboard: high concurrency with sub-second answers, where indexes (primary, aggregating, join) beat brute-force scans.
• Ingest, transformation, and serving need isolation without copies; multiple read-write engines share the same data.
• You want to try the real engine free; Firebolt Core self-hosts the same distributed engine with no data or query limits.
• PostgreSQL-compliant SQL keeps the dialect familiar.

Think twice when

• The managed service is AWS-only (six regions); multi-cloud plans need Core or a different warehouse.
• Core is source-available, not open source: Elastic License 2.0, unpublished source, and no offering it as a hosted service.
• The workload is classic broad-scan BI over huge history; scan-optimized warehouses and lake engines fit that economics better, and the real comparison set here is ClickHouse, StarRocks, and Pinot.

How it runs

Managed on AWS: stateless engines scale from one to 128 nodes and up to ten clusters, autoscaling on CPU, RAM, and queue time, over S3 with local SSD cache. Iceberg reads in place; READ_STREAM ingests from Kafka exactly-once. Firebolt Core ships as a container, deployable from a laptop to a multi-node cluster.