Pick the warehouse that fits the workload, not the one with the loudest sales team.
The data warehouse decision used to be Redshift versus everyone else. Today it is a three-way race between Snowflake, BigQuery, and ClickHouse, with each one optimal for a different shape of workload.
Snowflake
Strengths: separation of compute and storage with elastic scaling, multi-cluster warehouses, mature ecosystem, time travel, secure data sharing. Best for: enterprise BI, multi-tenant analytics, workloads with bursty query patterns.
Weaknesses: cost can spiral without active monitoring. The credit-based pricing rewards careful query optimization and punishes the inverse.
BigQuery
Strengths: fully managed serverless query engine, sub-second performance on petabyte tables, deep integration with the GCP ecosystem (Looker, Vertex AI, Pub/Sub). Best for: GCP-native shops, ad-hoc analytics, ML feature engineering.
Weaknesses: per-query pricing model is hostile to exploratory analysis without flat-rate pricing. Materialized views are less mature than Snowflake's.
ClickHouse
Strengths: blazing fast columnar engine for time-series and event data, dramatically cheaper than Snowflake or BigQuery for the same workload, open-source. Best for: product analytics, observability, ad tech, anything event-shaped.
Weaknesses: less polished tooling, limited ecosystem of dbt adapters and BI integrations, requires more operational effort (managed via ClickHouse Cloud or Tinybird).
The decision matrix
- Enterprise BI with mixed workloads? Snowflake.
- GCP shop, ML-heavy, comfortable with serverless? BigQuery.
- Event analytics, product analytics, real-time dashboards? ClickHouse.
- Less than 1TB and small team? Postgres or DuckDB. The warehouse problem is not yours yet.
The composite pattern
Many mid-size data teams now run two warehouses: Snowflake or BigQuery for canonical analytics, ClickHouse for the high-volume event data that would bankrupt either. The dbt models in the analytical warehouse aggregate and join from both. Cost optimal, query optimal, complexity tax acceptable.
What we ship
For client engagements, we default to ClickHouse Cloud for event-shaped workloads and Snowflake or BigQuery for classical BI. We have stopped recommending Redshift for new projects. The market consensus is mostly settled.