Multi-Cluster Warehouse
A multi-cluster warehouse is a Snowflake virtual warehouse that can run more than one compute cluster of the same size at once. It exists to handle concurrency: when many users and queries arrive together, additional clusters let more of them run at the same time instead of waiting in a queue. It is an Enterprise Edition feature.
Product details below were checked against Snowflake’s documentation in September 2026.
Scaling out, not up
A warehouse can grow in two directions. A larger size gives each query more resources — scaling up. More clusters let more queries run in parallel — scaling out. Snowflake’s documentation is explicit that multi-cluster warehouses are not as beneficial for improving slow-running queries or data loading; those are a question of size. The diagnostic is whether queries are slow while running or slow while waiting in a queue. Only the second is what extra clusters fix.
Two modes
A multi-cluster warehouse is configured with a maximum cluster count greater than one and a minimum.
- Maximized: the minimum equals the maximum. When the warehouse starts, all clusters start and run while the warehouse runs. This suits load that is known and steady.
- Auto-scale: the maximum is higher than the minimum. Snowflake starts clusters when queries queue for lack of resources and shuts them down when load falls.
Two scaling policies
In Auto-scale mode, a scaling policy decides how eagerly clusters start and stop.
| Policy | Starts a cluster when | Stops a cluster when | Favors |
|---|---|---|---|
| Standard (default) | A query is queued, or Snowflake estimates the running clusters cannot take on more queries | Load has stayed low for a sustained period | Short waits |
| Economy | Snowflake estimates enough load to keep a new cluster busy for at least six minutes | Estimated remaining work falls below six minutes, after its queries finish | Fewer credits |
The choice follows from what a queue costs. For dashboards and interactive users, waiting is the problem being solved, so Standard fits. For batch work where a short queue is acceptable, Economy avoids starting clusters for bursts that would have cleared on their own.
What it costs
Each running cluster consumes credits at the warehouse size’s rate, so consumption scales with size, the number of clusters actually running, and how long they run. Both scaling directions multiply cost. The maximum cluster count is therefore also a spending ceiling for the warehouse and deserves to be set deliberately rather than generously.
Extra clusters also cannot fix a bottleneck they do not own. If queries are waiting on something shared by every cluster — a source system, a heavily contended table — adding clusters adds cost without removing the wait.
How warehouses, their scaling choices, and credits fit into the rest of Snowflake’s architecture is worked through in Snowflake Architecture: Three Layers and What Each One Bills For.
References: Snowflake Documentation, Multi-cluster warehouses; Snowflake Documentation, Warehouse considerations.
Discover more from Insightful Data Lab
Subscribe to get the latest posts sent to your email.