Snowflake’s Plan to Cut AI Costs
TL;DR
Snowflake CEO Sridhar Ramaswamy joins Bloomberg Open Interest to explain why the next phase of enterprise AI is an economics question rather than a race for the biggest model. He describes how model routing sends each request to the model that actually fits, which cuts costs. He also warns that depending on a single model creates concentration risk, and argues AI agents can move workers from repetitive tasks toward higher-value work.
Nauti's Take
The opportunity is concrete: model routing cuts costs immediately because simple requests stop landing on the most expensive model. The risk sits in the added complexity, since every routing layer needs evaluation, monitoring and clear fallbacks.
Small teams should start with two models and one measurable task, while larger setups mainly gain from avoiding vendor lock-in.
Sources