Roughly 159,000 dbt jobs run daily, most on an hourly schedule even when underlying data hasn't changed. That means thousands of models rebuild identical results, driving up cloud costs as data usage expands across dashboards, machine learning pipelines, and AI-generated queries.

dbt State, announced at the last dbt Summit as state-aware orchestration (SAO), now extends beyond production to staging, CI, and development. It also works outside the dbt platform, including on dbt Core v2.0 via dbt login, or as a plugin for v1.7 through v1.12 with pip install dbt-state. The tool tracks a hash of each table's data and code state in a control plane, then only builds models whose upstream data or code actually changed. If an unchanged version exists elsewhere, it clones that instead of rebuilding.

Early users report average warehouse compute savings of 30%. Fanatics Betting and Gaming, running about 11 projects with a couple thousand models on Snowflake and Airflow, first piloted dbt State on a low-risk governance project. Initial model reuse was 0.2%, but after configuring source freshness and model-level SLAs, reuse climbed to roughly 15%, with some overscheduled projects hitting 25% and an overall average near 8%. The team now focuses on DAG-based jobs and freshness requirements rather than juggling selectors and schedule tags.

Pricing is based on daily active target tables (DATT), counting unique reuses per day, so repeated rebuilds of the same table don't multiply charges. A 30-day free trial is available for all users.

Source: dbt Labs Blog