Snowflake has added dynamic model routing to its Cortex AI Gateway, letting the platform automatically select the right AI model for each task rather than forcing every request through an expensive frontier model. The update addresses one of the most common complaints from enterprise data teams: runaway AI costs on workloads that did not need the most powerful (and most expensive) model in the first place.

The announcement landed August 18, and it is a meaningful shift in how businesses can manage AI spend within their data infrastructure.

What Dynamic Model Routing Actually Does

The idea is straightforward: not every AI task needs the same model. Summarising a product description, classifying support tickets, extracting fields from a form, running a spell check, these tasks do not require the same horsepower as multi-step financial analysis or complex reasoning over ambiguous regulatory language.

Snowflake’s Cortex AI Gateway now handles that triage automatically. Routine, lower-complexity work gets routed to smaller, more economical models. Deeper reasoning tasks get escalated to frontier models. The system also updates its routing decisions as model performance and pricing change over time, so the routing stays optimal without manual reconfiguration.

Two new open models, DeepSeek-V4-Flash and GLM-5.3, have been added alongside this rollout, expanding the pool of model choices available for cost-effective routing.

The feature spans Snowflake’s product suite, including Snowflake CoCo, Snowflake CoWork, and third-party AI agents integrating via the Cortex AI Gateway API.

Why Enterprise Data Teams Are Paying Attention

Enterprise AI costs have become a serious budget line. When teams first started deploying AI inside their data platforms, most simply pointed every query at whichever frontier model was available. That worked fine in pilot phases with modest query volumes. It stopped working fine when those agents went to production and started processing millions of requests.

Dynamic model routing solves this without requiring data engineers to manually build routing logic into every pipeline. That is significant because building and maintaining custom routing rules adds engineering overhead, introduces bugs, and needs updating every time a new model drops.

Organizations running Snowflake as their core data platform can now let the gateway handle this automatically, spending less on commodity AI tasks while preserving budget for the complex reasoning workloads that actually require frontier model capability.

For regulated industries, there is an additional control layer: administrators can specify which models and providers are available to which users, which matters for global organizations navigating regional data residency requirements or compliance constraints around which AI providers are approved.

What This Means for Business

The cost pressure around enterprise AI is real and growing. Research published earlier this year found a significant portion of enterprise AI budgets were being consumed by over-provisioned model usage, where businesses paid frontier model prices for tasks that a smaller model would handle equally well.

Snowflake’s approach of embedding intelligent routing directly into the data platform layer, rather than expecting application teams to solve this themselves, is the direction the market has been heading. If you are already building data pipelines, AI agents, or analytics workflows inside Snowflake, this update means cheaper operations without a rebuild.

A few things to think through for your own setup:

Audit your current model usage. Before assuming dynamic routing will save money, understand what your Cortex AI queries are actually doing. If most of your load is already well-suited to smaller models, the savings could be substantial. If your workloads are genuinely complex across the board, the impact will be more modest.

Test routing decisions on representative data. The automatic routing is based on task complexity signals, but edge cases exist. Run your typical query mix through the new gateway and validate that quality holds up on tasks the system routes to lighter models.

Consider governance alongside cost. The model selection controls are just as important as the cost savings for teams in finance, healthcare, or legal. Build your provider allowlist with compliance in mind from the start, not as an afterthought.

Plan for your AI agent traffic. If you are running AI agents that call Cortex AI Gateway frequently, dynamic routing changes your cost model in ways worth projecting before you scale up.

The broader trend here is data platforms taking on more of the operational complexity of running AI at scale. A year ago, managing model selection was squarely an application concern. Increasingly it is becoming platform infrastructure, and Snowflake’s move is a clear signal of where that market is going.

For data teams trying to demonstrate AI ROI to leadership, having a built-in mechanism to control and reduce AI costs without sacrificing output quality is a meaningful argument to have in your back pocket.

---

Enterprise DNA helps business leaders make sense of data and AI trends. If you are building an AI-powered data operation and want expert guidance, explore Omni Advisory or learn the fundamentals with EDNA Learn.