Snowflake

Use it when

• Operational simplicity is the priority: no infrastructure to run, warehouses that pause and resume, and scaling that is a size setting.
• Workload isolation matters; separate virtual warehouses for ETL, BI, and data science bill independently against one copy of the data.
• Data sharing across accounts and clouds is a first-class need; it runs on AWS, Azure, and Google Cloud.
• Snowpipe Streaming, semi-structured data, and managed Iceberg cover your ingestion and format needs without extra systems.

Think twice when

• Cost discipline is weak; per-second credits with a 60-second minimum reward tuning and punish always-on warehouses, which is where most Snowflake bill stories start.
• Openness is the strategy; storage is proprietary micro-partitions (Iceberg support notwithstanding), where Databricks SQL keeps tables engine-neutral.
• The workload is high-QPS application serving; ClickHouse-shaped engines hold sub-second latency at concurrency more cheaply.

How it runs

Fully managed across the three clouds: independent virtual warehouses (MPP compute) scale and pause per workload over shared storage; multi-cluster warehouses absorb concurrency automatically. Governance runs on a role hierarchy with masking and row-access policies under the Horizon catalog. Time travel is a day by default, up to 90 on Enterprise.