BigQuery

Use it when

• You want a warehouse with effectively no operations: no clusters, no sizing, no maintenance windows.
• The stack is Google Cloud, or the analytics should sit beside GCS, Dataflow, and Vertex AI.
• Workloads are irregular; on-demand pricing by bytes scanned means idle costs nothing, and reservations cap spend when usage stabilizes.
• Streaming ingest and in-warehouse ML (BQML) shorten the path from event to model.

Think twice when

• Heavy, steady scan volume on demand gets expensive; the move to slot reservations changes the cost conversation entirely, so model both early.
• Multi-cloud portability matters; storage is proprietary Capacitor and the dialect is GoogleSQL, though Omni reaches S3 and Azure Blob and managed Iceberg is available.
• Latency targets are interactive-application grade; BI Engine helps dashboards, but serving engines like ClickHouse hold sub-second under concurrency.

How it runs

Fully managed and serverless on Google Cloud: queries run on Dremel's shared execution layer over Colossus storage, with compute and storage billed separately. Pick on-demand (bytes scanned) or slot reservations with autoscaling. The Storage Write API handles streaming; federated queries and BigLake/Iceberg tables reach external data.