How to reduce Snowflake costs: 15 tactics with example SQL

Understand where credits go

Your bill has three main parts: compute (virtual warehouses), serverless features (Snowpipe, automatic clustering, serverless tasks and others) and storage. Start with credits by service type, then by warehouse, then by query. The views in SNOWFLAKE.ACCOUNT_USAGE lag by up to a few hours, so look at the last 7 to 30 days.

-- Credits billed by service type, last 30 days
SELECT service_type,
       ROUND(SUM(credits_billed), 1) AS credits_billed
FROM snowflake.account_usage.metering_daily_history
WHERE usage_date >= DATEADD(day, -30, CURRENT_DATE())
GROUP BY service_type
ORDER BY credits_billed DESC;

-- Most expensive repeated queries, last 7 days
SELECT query_parameterized_hash,
       ANY_VALUE(warehouse_name) AS warehouse,
       COUNT(*) AS runs,
       ROUND(SUM(credits_attributed_compute), 2) AS credits
FROM snowflake.account_usage.query_attribution_history
WHERE start_time >= DATEADD(day, -7, CURRENT_TIMESTAMP())
GROUP BY query_parameterized_hash
ORDER BY credits DESC
LIMIT 20;

QUERY_ATTRIBUTION_HISTORY excludes warehouse idle time and very short queries, so its total will be lower than WAREHOUSE_METERING_HISTORY. The gap between the two is roughly your idle cost, which tactic 3 attacks.

Check Cost Insights in Snowsight

Snowflake's Cost Insights appear in Snowsight under Admin » Cost management » Account Overview, in the Optimization insights tile. Snowflake recalculates them weekly. They flag items such as rarely used tables with automatic clustering, rarely used materialized views and search optimization paths, large tables that are never queried, short-lived permanent tables, active warehouses with large gaps between queries, and inefficient multi-cluster usage. Each is a ready-made shortlist for tactics 9 to 11.

Fix queries that scan too much data

Snowflake stores tables in micro-partitions and skips the ones a filter rules out (pruning). Queries that scan most partitions of a large table are usually the most expensive ones you can fix.

SELECT query_id, warehouse_name,
       partitions_scanned, partitions_total,
       ROUND(bytes_scanned / POWER(1024, 3), 1) AS gb_scanned,
       ROUND(total_elapsed_time / 1000) AS seconds
FROM snowflake.account_usage.query_history
WHERE start_time >= DATEADD(day, -7, CURRENT_TIMESTAMP())
  AND partitions_total > 1000
  AND partitions_scanned / partitions_total > 0.8
ORDER BY bytes_scanned DESC
LIMIT 25;

Common fixes: replace SELECT * with the columns you need (Snowflake stores data by column, so fewer columns means less data read); filter on a date or other well-clustered column; avoid wrapping filter columns in functions; filter before joins; and switch full-refresh dbt models to incremental ones. Use Query Profile in Snowsight to see where time and bytes go. Our database optimization guide covers the query side in more depth.

Let the result cache work

Snowflake keeps query results for 24 hours (extended each time they are reused, up to 31 days). If the same query text runs again and the underlying data has not changed, the result returns without running the query on a warehouse. Small differences in the text, such as letter case or table aliases, prevent reuse, and so do functions such as RANDOM() or UUID_STRING(). Make BI tools send stable query text, and avoid disabling USE_CACHED_RESULT.

Review clustering keys and automatic clustering

A clustering key helps large tables that are filtered on the same column most of the time. Automatic clustering then runs as a serverless feature and costs credits every time data changes. It pays off when the table is queried often and changes less often.

SELECT SYSTEM$CLUSTERING_INFORMATION('analytics.sales.orders', '(order_date)');

SELECT table_name, ROUND(SUM(credits_used), 1) AS credits_30d
FROM snowflake.account_usage.automatic_clustering_history
WHERE start_time >= DATEADD(day, -30, CURRENT_TIMESTAMP())
GROUP BY table_name
ORDER BY credits_30d DESC;

ALTER TABLE analytics.sales.orders SUSPEND RECLUSTER;

Suspend or drop clustering on tables that are rarely queried, as Cost Insights may already be telling you.

Weigh materialized views and search optimization

Materialized views and the search optimization service (both Enterprise Edition features) speed up specific query patterns, but Snowflake maintains them in the background with serverless credits plus extra storage. They are worth it for a frequently run query on a large, slowly changing table, and rarely worth it on tables with heavy updates or few readers. Use SYSTEM$ESTIMATE_SEARCH_OPTIMIZATION_COSTS before enabling search optimization, and track spend in MATERIALIZED_VIEW_REFRESH_HISTORY and SEARCH_OPTIMIZATION_HISTORY.

ALTER TABLE analytics.web.events DROP SEARCH OPTIMIZATION;

Set resource monitors on warehouses

A resource monitor tracks warehouse credits against a quota and can notify, suspend after running queries finish, or suspend immediately. By default only ACCOUNTADMIN can create one. Monitors cover warehouses only, not serverless features.

CREATE OR REPLACE RESOURCE MONITOR rm_transform
  WITH CREDIT_QUOTA = 500
  FREQUENCY = MONTHLY
  START_TIMESTAMP = IMMEDIATELY
  TRIGGERS
    ON 75 PERCENT DO NOTIFY
    ON 90 PERCENT DO SUSPEND
    ON 100 PERCENT DO SUSPEND_IMMEDIATE;

ALTER WAREHOUSE transform_wh SET RESOURCE_MONITOR = rm_transform;

Use budgets for account-wide and serverless spend

Budgets cover a calendar month and include serverless features, so they fill the gap that resource monitors leave. They alert; they do not stop work. Activate the account budget in Snowsight, then create custom budgets for teams or projects (up to 100 per account).

CREATE SNOWFLAKE.CORE.BUDGET budgets_db.budgets_schema.finance_budget();
CALL budgets_db.budgets_schema.finance_budget!SET_SPENDING_LIMIT(500);
CALL budgets_db.budgets_schema.finance_budget!ADD_RESOURCE(
  SELECT SYSTEM$REFERENCE('WAREHOUSE', 'finance_wh', 'SESSION', 'applybudget'));

The role needs the SNOWFLAKE.BUDGET_CREATOR database role and APPLYBUDGET on each object it adds. Notification emails must be verified first.

Tag warehouses and queries for chargeback

Costs fall fastest when each team sees its own number. Tag dedicated warehouses with a cost center; for shared warehouses, set a QUERY_TAG per tool or team and report from QUERY_ATTRIBUTION_HISTORY.

CREATE TAG IF NOT EXISTS governance.tags.cost_center;
ALTER WAREHOUSE finance_wh SET TAG governance.tags.cost_center = 'finance';

SELECT tr.tag_value AS cost_center,
       ROUND(SUM(wmh.credits_used), 1) AS credits_30d
FROM snowflake.account_usage.warehouse_metering_history wmh
JOIN snowflake.account_usage.tag_references tr
  ON tr.object_name = wmh.warehouse_name
 AND tr.domain = 'WAREHOUSE'
 AND tr.tag_name = 'COST_CENTER'
WHERE wmh.start_time >= DATEADD(day, -30, CURRENT_TIMESTAMP())
GROUP BY tr.tag_value
ORDER BY credits_30d DESC;

For shared warehouses: ALTER SESSION SET QUERY_TAG = 'team:marketing'; in each tool's connection settings, then group attribution history by query_tag.