At Snowflake Summit 2026, dbt Labs walked away with the 2026 Data Integration Partner of the Year award from Snowflake—its fourth consecutive win. The company also snagged the CoCo Adoption Award for driving adoption of Snowflake's coding agent and control plane. This isn't just a trophy moment; it reflects a deeply integrated technical partnership that's reshaping how joint customers build production data pipelines.
Consider the numbers: over 75% of Snowflake customers with dbt accounts rely on the platform for transformation and context management. More telling, 90% of those joint customers actively use Snowflake Cortex AI. That suggests dbt's semantic layer, Fusion engine, and MCP server are becoming the default orchestration and governance layer for AI workloads. The architecture here is straightforward: dbt provides the structured, governed data foundation; Snowflake provides the compute and AI services. The result is a pipeline where data moves from raw ingestion to AI-ready assets without the usual governance gaps.
From an engineering perspective, the key trade-off is between flexibility and control. dbt's approach enforces a transformation-as-code paradigm, which introduces schema-on-write discipline. For teams running agent-based systems, this reduces drift and improves reproducibility—critical when you're feeding models that can't tolerate schema changes mid-flight. The Snowflake Marketplace integration, with transactions growing 30% year-over-year, lets teams consume and publish dbt-ready datasets directly, cutting down ETL boilerplate.
WHOOP's case study illustrates the practical payoff. They used dbt to enforce data quality and governance as they migrated to Snowflake, then built a financial forecasting model using Snowpark. The bottleneck shifted from data cleaning to strategic modeling—exactly the outcome you'd expect when you decouple transformation from consumption. For data engineers, the lesson is clear: invest in a transformation layer that scales with your AI ambitions, but be ready to manage the metadata overhead that comes with it.
Source: dbt Labs Blog