A Turning Point for the Modern Data Stack
In mid-September 2026, two thousand data professionals gathered in Las Vegas for the dbt Summit. The announcements from Fivetran and dbt Labs that emerged from the event mark a clear inflection point in data engineering: the general availability of dbt v2, active warehouse cost control through selective execution, and Git-versioned dashboards now treated as first-class code.
A Rust Engine That Is Ten Times Faster
dbt v2 drops Python entirely in favor of a ground-up rewrite in Rust. The impact is immediate: a project with ten thousand models now parses in real time, with a reported 10x speed improvement over the previous version.
This performance leap fundamentally changes the development loop. Errors, column validation, and lineage checks are caught before any warehouse execution, collapsing debugging from multi-minute cycles into a near-instant feedback experience. Generally available adapters cover BigQuery, Databricks, DuckDB, Redshift, and Snowflake; ClickHouse and Spark follow in beta. The era of two parallel engines — Core Python alongside Fusion Rust — is over. The open-source version remains published under the Apache 2.0 license.
Cutting Warehouse Costs Through Selective Execution
Also reaching general availability is dbt State. This orchestration feature identifies, before each run, only the models that have actually changed — by cross-referencing warehouse metadata against source SQL — and recomputes the strict minimum. Published customer results are telling: a 59% reduction in scheduled job costs for a healthcare services provider, and 25% savings on both BigQuery spend and processing time for a European telecom operator.
Those numbers land against a difficult backdrop. According to the dbt Labs annual survey, 57% of data teams report rising warehouse and compute costs — yet only 36% say their budgets are growing at the same rate. The pressure on data leadership to do more with less has rarely been this acute.
Dashboards Finally Move Into the Git Repository
The announcement with arguably the greatest long-term implications is dbt Charts, now in public beta. It brings dashboards into the same Git repository as SQL code: dashboards are declared in YAML, versioned, testable, and subject to the same code review workflow as any data transformation. The long-standing divide between data engineering teams and BI tooling — two worlds with separate lifecycles and governance practices — is beginning to close structurally, rather than being papered over with organizational workarounds.
Running in parallel, the Fivetran Context Layer — available in early access — offers an open-source infrastructure layer that standardizes how data and metadata are exposed to external consumers, whether analytics tools or automated pipelines.
What This Means for Data Teams
These announcements share a single underlying logic: one fast, open engine; active warehouse cost governance; and a steady convergence of SQL code, tests, and dashboards into a unified, version-controlled pipeline. For organizations struggling to align infrastructure spend with business value delivered, the timing could hardly be better.