Enabling cross-region inference requires setting the account-level parameter CORTEX_ENABLED_CROSS_REGION, which can only be modified using the ACCOUNTADMIN role via the ALTER ACCOUNT command (ORGADMIN cannot set it). Billing for LLM inference credits is based on the requesting region's service consumption rates — Snowflake documentation states that credits are considered consumed in the requesting region even when the actual processing occurs in a different region. Finally, because the source account is on AWS and the destination LLM region is also on AWS (the same cloud service provider), inference traffic is routed over the AWS private backbone network rather than the public internet, meaning the data stays within the AWS global network during transit, though it is not confined solely to the requesting region.
Community Discussion