Snowflake is putting cost discipline at the center of its enterprise AI pitch. In a Bloomberg Technology interview, Snowflake CEO Sridhar Ramaswamy said the next phase of enterprise AI is about economics, not simply using the biggest model available. The mechanism is model routing: automatically choosing which AI model should handle a given task. Bloomberg says Ramaswamy described how routing can cut AI costs and why relying on a single model is a problem for enterprise users. VentureBeat reports the same basic thesis in product terms: enterprise teams running AI agents at scale can overpay when one model handles every request. A powerful model may be unnecessary for simple questions, while a cheaper or weaker model may not be capable enough for harder tasks. According to VentureBeat, Snowflake’s gateway now auto-routes queries to address that mismatch. The outlet reports that the approach can cut costs by up to 3x, though that specific figure is not corroborated by the supplied Bloomberg summary. The strategic point is narrower than “AI gets cheaper” across the board. The sources describe a routing layer that attempts to map workload difficulty to model capability, rather than forcing every query through the same model. That matters most when enterprises move from experiments to repeated agentic workflows, where small per-query inefficiencies can compound. The cluster does not provide implementation detail on which models are routed, how Snowflake evaluates query complexity, or how customers are charged. For now, the supported takeaway is that Snowflake is using model routing to sell enterprise AI on operating economics, with VentureBeat reporting a potential cost reduction of up to 3x. Who benefits: Snowflake benefits if customers view its platform as a cost-control layer for AI workloads. Enterprise teams running high-volume AI agents could benefit if routing lowers spend without degrading task quality. Who's exposed: Vendors or teams that default every request to one large model are exposed to cost scrutiny. The supplied reports do not identify specific competitors or customer segments beyond enterprise AI users.