Snowflake Cortex AI review
SQL-native AI functions and agents that run inside Snowflake's governance boundary.
TECHi verdict
Cortex is the most disciplined answer yet to a question every data platform now faces: how do you sell AI to a CFO? Flat-priced AI Credits, per-function metering, and prompt-caching discounts make it feel like buying a utility rather than funding a science project — and for regulated enterprises, AI that inherits Snowflake's access controls and never leaves the boundary genuinely beats a better model on the wrong side of the firewall. The catch is captivity in both directions: Snowflake picks the model catalog, and heavy inference is cheaper on raw provider APIs than routed through a warehouse. Adopt Cortex for governed, data-adjacent AI — extraction, classification, analytics agents. Route high-volume, latency-sensitive inference elsewhere, and don't let the SQL convenience hide the meter.
Pros
- + AI functions inherit Snowflake RBAC, masking, and audit — data never crosses the governance boundary to reach a model
- + Flat-priced AI Credits ($2.00 global, every edition) make spend forecastable — still rare in enterprise AI
- + AISQL means an analyst can ship a working extraction or classification pipeline in an afternoon, no Python service required
Watchouts
- - Only rational while Snowflake is your data center of gravity; the model menu is Snowflake's choice, not yours
- - Per-token pricing is predictable but batch jobs aren't — one frontier-model AISQL query over a million rows can cost five figures
Frequently asked
What are the pros of Snowflake Cortex AI?
AI functions inherit Snowflake RBAC, masking, and audit — data never crosses the governance boundary to reach a model. Flat-priced AI Credits ($2.00 global, every edition) make spend forecastable — still rare in enterprise AI. AISQL means an analyst can ship a working extraction or classification pipeline in an afternoon, no Python service required
What are the watchouts / cons of Snowflake Cortex AI?
Only rational while Snowflake is your data center of gravity; the model menu is Snowflake's choice, not yours. Per-token pricing is predictable but batch jobs aren't — one frontier-model AISQL query over a million rows can cost five figures
What is the TECHi verdict on Snowflake Cortex AI?
Cortex is the most disciplined answer yet to a question every data platform now faces: how do you sell AI to a CFO? Flat-priced AI Credits, per-function metering, and prompt-caching discounts make it feel like buying a utility rather than funding a science project — and for regulated enterprises, AI that inherits Snowflake's access controls and never leaves the boundary genuinely beats a better model on the wrong side of the firewall. The catch is captivity in both directions: Snowflake picks the model catalog, and heavy inference is cheaper on raw provider APIs than routed through a warehouse. Adopt Cortex for governed, data-adjacent AI — extraction, classification, analytics agents. Route high-volume, latency-sensitive inference elsewhere, and don't let the SQL convenience hide the meter.