Semi-structured Data

Snowflake treats semi-structured data as a first-class citizen. You can land JSON/Parquet/Avro/ORC as-is, keep full fidelity in a VARIANT column, and project only the attributes you need—now or later. The engine stores these values columnarly within micro-partitions and uses path metadata to prune aggressively, so “schema-on-read” doesn’t have to mean slow.