State-aware orchestration now in Preview for Fusion projects

As your data estate grows, costs and complexity grow with it. Across the dbt ecosystem, teams built over 33 billion models in the past year—more than four per person on Earth. Scale brings real costs in compute and cognitive load. Data growth does not have to mean complexity.

Today, we're excited to announce that state-aware orchestration is now available in Preview for dbt platform projects running on Fusion.

What is state-aware orchestration?

State-aware orchestration is a Fusion-powered innovation that fundamentally rewrites how data pipelines work—by only running what actually needs to run.

In a traditional dbt run, every model in your DAG is rebuilt regardless of whether its inputs have changed.

With traditional dbt orchestration, models are rebuilt even when they don't need to be

With state-aware orchestration enabled, dbt moves from being stateless to stateful. Instead of simply running exactly what was specified in a job, dbt maintains a real-time fingerprint of both model code and data state. The system pinpoints exactly which models need refreshing based on when new data is produced upstream or when code changes occur. Models without any upstream changes are automatically reused, eliminating unnecessary processing and only building models that will actually produce a different result.

With state-aware orchestration, stale models are reused

Advanced configuration for fine-tuned control

With state-aware orchestration, you get immediate savings by simply enabling it, but the real power comes with the advanced configuration options. Instead of rigidly scheduling jobs, you can now express your data freshness requirements directly in your dbt project.

These intent-based configurations allow you to specify exactly how fresh each model needs to be and under what conditions it should rebuild. For example, you can tell a model to:

• Wait for all upstream dependencies with updates_on: all - only rebuilding when all source data is fresh
• Define maximum staleness windows with build_after: 6h - ensuring models rebuild at least every six hours regardless of upstream changes
• Prioritize critical models with varying freshness requirements based on business importance

The system intelligently orchestrates your pipeline based on these declared intents rather than rigid schedules. This means models will only run when they'll actually produce different results, delivering immediate cost savings without compromising data freshness.

with tuning of SAO settings, intelligent rules will be observed for higher reuse

Smarter testing, same standards

With column‑aware testing, checks only run when the underlying data actually changes, eliminating unnecessary compute. State‑of‑the‑art static analysis identifies tests that are guaranteed to pass and safely skips them, while intelligent batching combines multiple checks into fewer, more efficient operations. The result is the same rigorous quality bar with with far fewer redundant table scans and compute, and no loss in accuracy.

The business impact: Cost savings without compromise

For data leaders and platform owners who need to control costs while delivering high-quality insights, state-aware orchestration delivers measurable bottom-line benefits:

• Immediate ROI with zero-implementation effort: Simply flip a switch to enable state-aware orchestration. No rewrites or restructuring: just immediate 10% reduction in compute costs based on our beta results.
• Data freshness on your terms: Fine-tune configurations to meet your business SLAs without over-scheduling. Declare your freshness requirements, and Fusion handles the rest: delivering an additional 15%+ in cost savings.
• Smarter testing: Fusion intelligently skips tests when they do not need to run. The result is an estimated 4% additional annual cost savings.
• Total impact: 29%+ reduction in annual dbt-related compute costs requiring only minor changes.

Estimated cost reductions with Fusion and SAO

Real-world results from dbt Labs

At dbt Labs, we've implemented state-aware orchestration in our own internal analytics project with impressive results. With a data estate containing around 1,500 models and many jobs on various schedules, we saw immediate impact. After enabling Fusion with state-aware orchestration, we immediately saw around 35% of our models being reused daily, resulting in 9% cost savings on our compute bill, just from turning it on, with no tuning required.

Then things got interesting. When we added tuned configurations to optimize our data freshness requirements, we achieved even more dramatic results.

results from dbt Labs' internal project

When we implemented tuned configurations to optimize our data freshness requirements, the results were impressive and surprisingly easy to achieve. Here's what we did:

• Simplified freshness configurations: We created just two freshness tiers - daily (our global default for analytical purposes) and hourly (for operational needs that require more frequent updates).
• Streamlined job management: We reduced our job complexity from numerous over-scheduled jobs to just two: one hourly job that refreshes both daily and hourly data appropriately, and a weekend job for cleanup and targeted backfills.

Many teams will ask how difficult this transformation was to implement. The answer surprised even us: just a few lines of YAML at the project definition level, some model-specific overrides, and configuring the two jobs.

Here's an example of a model-specific override:

-- fct_dbt_invocations.sql
{{ 
    config(
        materialized = 'incremental',
        unique_key = 'invocation_id',
        freshness = {
            'build_after': {
                'updates_on': 'all'
            }
        }
    ) 
}}

And here's what our daily grain configuration looks like today:

--dbt_project.yml
models:
  +freshness:
    build_after:
      count: 1
      period: day
    updates_on: any

With this simple setup, we dramatically reduced both our cognitive overhead and our compute bill.

Developer experience benefits

Beyond the business value, state-aware orchestration transforms how data teams work:

• Faster development cycles: Reduced run times mean quicker iterations and faster time-to-insight.
• Simplified job management: No more wrestling with complex scheduling. Define freshness requirements once, and Fusion handles the orchestration. Jobs are also aware of each other, preventing concurrent writes to the same table, eliminating a significant pain point for many data teams.
• Fewer false alarms: With more precise orchestration comes less noise: reducing alert fatigue and helping teams focus on actual issues.
• More time for new initiatives: When your team spends less time managing jobs and monitoring costs, they can focus on delivering business value through data.

Get started

State-aware orchestration is available in Preview for Enterprise customers running their projects on Fusion. Enabling it is as simple as flipping a switch in your project settings.

With state-aware orchestration, your data infrastructure can now be as intelligent as the insights it delivers: running smarter, not harder.

Ready to rewrite the rules for your data platform? Learn more in our documentation or contact your account team to get started.