
Queries that access data across regions incur data transfer fees. Design your account architecture to keep data and compute in the same region.

Queries that access data across regions incur data transfer fees. Design your account architecture to keep data and compute in the same region.
Scenario: Query transfers 100GB/day across regions
Same region: $0
Cross-region (AWS): 100GB × $0.02 × 30 = $60/month
Cross-region (different cloud): 100GB × $0.10 × 30 = $300/month
At enterprise scale (10TB/day):
10,000GB × $0.02 × 30 = $6,000/month
10,000GB × $0.10 × 30 = $30,000/month
Problem: Data in US East, European users query from EU West
Solution: Replicate data to EU region, users query local replica
Problem: Source in Asia Pacific, processing in US East, destination in EU West
Solution: Process data in same region as source, transfer only final results
-- Create warehouses in same region as databases
CREATE WAREHOUSE eu_analytics_wh WITH
WAREHOUSE_SIZE = 'MEDIUM'
COMMENT = 'EU region warehouse for EU data';
-- EU users use eu_analytics_wh
-- US users use us_analytics_wh
-- Replicate to target region
CREATE DATABASE analytics_eu
AS REPLICA OF account1.analytics_us;
-- Schedule refresh
ALTER DATABASE analytics_eu REFRESH;
A global company had users in US, EU, and APAC querying central US database. Before: All data in US East, 50% of 10TB daily queries cross-region = 5TB, 5,000GB × $0.02 × 30 = $3,000/month transfer cost.
After: Primary in US East, replicated to EU Central and APAC Sydney, each region queries local replica. Daily replication 30TB/month, 30,000GB × $0.02 = $600/month, query transfers $0 (all same-region). Savings: $2,400/month (80% reduction).
SELECT
SOURCE_CLOUD,
SOURCE_REGION,
TARGET_REGION,
SUM(BYTES_TRANSFERRED)/1024/1024/1024 as GB_TRANSFERRED,
SUM(BYTES_TRANSFERRED)/1024/1024/1024 * 0.02 as APPROX_COST
FROM SNOWFLAKE.ACCOUNT_USAGE.DATA_TRANSFER_HISTORY
WHERE START_TIME >= DATEADD(month, -1, CURRENT_TIMESTAMP())
GROUP BY 1, 2, 3
ORDER BY GB_TRANSFERRED DESC;
Uncover hidden inefficiencies and start reducing Snowflake spend in minutes no disruption, no risk.