A team loads 50 TB of historical transaction data into a large table that continues...

Snowflake automatically divides table data into immutable, columnar micro-partitions. Users do not manually define these partitions.

Snowflake stores metadata for each micro-partition, including information such as the range of values for individual columns. When a query filters transaction_date to January 2025, the optimizer can compare the predicate with the minimum and maximum date values recorded for each micro-partition.

Micro-partitions whose date ranges cannot contain qualifying rows are excluded from the scan. This operation is called micro-partition pruning and can significantly reduce the amount of data read.

Micro-partitions normally contain approximately 50 MB to 500 MB of uncompressed data; they are not fixed at 500 MB. Snowflake uses columnar—not row-based—storage for table data.