Three Layers
Snowflake splits the warehouse into three independent layers. Storage keeps compressed columnar data in cloud object stores. Compute runs queries on clusters called virtual warehouses. A cloud services layer handles metadata, optimization, security, and transactions.
Separation of Storage and Compute
Because storage lives in object storage, many compute clusters can read the same data at once. Each team can spin up its own virtual warehouse sized to its needs, and they do not fight over CPU. You can resize or pause a warehouse without touching the data.
Why It Scales
- Elasticity: add compute clusters for bursts and remove them when idle.
- Isolation: a heavy report on one warehouse does not slow another.
- Single copy of data: no need to replicate datasets per team.
- Pay for use: compute bills only while a warehouse runs.
Key idea
Snowflake decouples a single shared storage layer from many independent compute clusters, so teams scale and isolate query workloads without copying data.