Data platform company Snowflake claims its innovative AI model routing technology can deliver up to three times greater token efficiency, a crucial development for enterprises grappling with the escalating costs of deploying artificial intelligence at scale. The company's Vice President of Applied AI, Baris Gultekin, highlighted that automatically matching AI models to specific tasks is key to controlling expenses as AI agents become more prevalent across businesses.
Optimizing AI Workloads with Cortex AI Gateway
Snowflake's Cortex AI Gateway demonstrated this significant efficiency gain in a recent test, all while maintaining comparable output quality. The core principle behind this technology is its ability to automatically direct incoming tasks to the most appropriate AI models based on their capabilities and the unique requirements of each workload. This intelligent orchestration prevents the costly practice of relying on a single, often high-power and expensive, model for every single request.
Gultekin emphasized the shift in focus, stating, "The economics of AI are now boardroom math." He added that as companies invest more in AI, they are increasingly scrutinizing whether the value generated justifies the associated costs. This challenge is particularly acute as AI applications move from pilot projects to full production, where AI agents often make multiple model calls to complete complex tasks.
Addressing the Cost of Scaling AI
The issue of computational cost becomes magnified when simpler, less expensive models could suffice for routine operations, yet an advanced, powerful model is used indiscriminately. Snowflake argues that enterprises need to look beyond individual model call costs and instead prioritize "intelligence efficiency" – how effectively compute power, models, data, and context are converted into tangible business impact. Even minor inefficiencies can accumulate rapidly across thousands of workflows, making intelligent model selection paramount for economic viability.
Introduced in August 2026, Snowflake's dynamic model routing aims to provide businesses with a method to make AI deployments more economical without compromising the quality of the output. The system ensures that a model best suited for a particular task is utilized, rather than defaulting to a more expensive option for every query. For instance, a basic summary task does not require the same sophisticated reasoning capabilities as a complex analytical problem.
Beyond Cost: Balancing Performance, Governance, and Quality
The complexity of AI agents, which often involve multiple model calls and tool interactions, makes model selection even more critical. Enterprises must evaluate not only the cost per call but also the overall output quality, generation time, and the extent of rework required. A seemingly cheaper model might ultimately increase costs if its output demands substantial human correction.
"The goal should be to make AI useful in more places across the business. That means being disciplined about where additional intelligence makes a difference and where it only adds cost," said Baris Gultekin.
Snowflake views this model orchestration as a component of a broader ‘agentic control plane’, designed to manage models and agents efficiently while balancing cost, governance, and performance at scale. Through dynamic model routing within the Cortex AI Gateway, the system automatically selects from customer-approved models, considering factors like quality, cost, latency, reliability, security, and how well the model integrates with proprietary enterprise data.
Gultekin cautions against relying solely on benchmarks, stressing that real-world performance with unique enterprise data and operating rules is often different from controlled test environments. Governance is also critical; Snowflake's system restricts model choices to administrator-approved options and provides visibility into which model handles each request, adhering to existing data security and residency controls. Ultimately, providing the right data and selecting an appropriate model are fundamental to improving both outcomes and cost-efficiency in AI deployments.