Maximize AI Impact: More Model Choice, Smarter Routing

Details

Snowflake CEO Sridhar Ramaswamy is reframing enterprise AI success around 'intelligence efficiency' — measuring business output per compute dollar rather than raw token consumption. The blog announces expanded model choice in Cortex AI Gateway, adding GLM-5.3 and DeepSeek-V4-Flash 0731 to a roster that already includes Anthropic, Google, Mistral AI, OpenAI, and SpaceXAI. The core argument: model rankings shift constantly as new releases drop and costs fall, making vendor lock-in to a single frontier model economically risky. Snowflake's answer is neutral multi-model routing that picks the cheapest model meeting quality and latency requirements for each task type. The economic logic holds: as open-model inference costs fall, workloads previously too expensive to automate become viable. Enterprises that route dynamically across a portfolio capture that savings automatically instead of renegotiating vendor contracts. The post is primarily thought leadership — specific routing algorithms, SLA guarantees, and pricing details for Cortex AI Gateway are not disclosed. The announcement matters mainly as a signal that major cloud data platforms are positioning themselves as model-agnostic middleware layers rather than single-model storefronts.