Snowflake adds automatic AI model routing to cut enterprise inference costs

Snowflake has added automatic model routing to its Cortex AI Gateway, a move aimed at helping enterprise customers lower the cost of AI queries while keeping traffic inside its security controls. The company said the feature is available through an "auto" setting and can reduce token spending on some workloads by up to three times, based on internal testing.

The system is designed to choose between smaller and larger models depending on the task, rather than sending every request to the most capable option. Snowflake said customers can still lock routing to a single model or a limited set of approved models, and that the routing decision itself carries no extra fee.

Baris Gultekin, Snowflake's vice president of AI, said the company is tying routing to the same governance framework it already uses for data, models and agents. He said open and proprietary inference stays within Snowflake's security boundary, including for open models that can run in a customer's region to meet data residency rules.

The release comes as other vendors, including Databricks, AWS, Google Cloud, Nvidia and OpenRouter, push similar routing tools for enterprise AI. Snowflake's pitch is that its version is built around access control, cost attribution and governed data rather than model choice alone, according to the company and industry analysts.