What are the three layers of Snowflake's architecture (Storage, Compute, Cloud Services) and what does each do?

Snowflake is built on three independently operating layers, each with a clearly defined responsibility.

Storage Layer holds all data as columnar, compressed micro-partition files in cloud object storage (S3, Azure Blob, or GCS). Snowflake manages all storage operations automatically — no manual partitioning, indexing, or vacuuming required. The data is always available to any compute cluster that needs it.

Compute Layer consists of Virtual Warehouses — MPP clusters from X-Small to 6X-Large. Each warehouse is a temporary set of cloud compute instances that reads from storage, executes queries, and writes results. Warehouses start, resize, and suspend in seconds, completely independently of the data layer.

Cloud Services Layer is Snowflake's always-on brain. It handles query parsing and optimization, micro-partition metadata management (min/max values per column per partition), authentication and authorization, ACID transaction management, and infrastructure coordination. It runs continuously and is only billed when daily consumption exceeds 10% of compute credits.

-- Create a Virtual Warehouse in the Compute layer
CREATE WAREHOUSE reporting_wh
  WAREHOUSE_SIZE = 'LARGE'
  AUTO_SUSPEND = 300
  AUTO_RESUME = TRUE;

-- COUNT(*) is answered from Cloud Services metadata -- zero compute credits
SELECT COUNT(*) FROM orders;