What is dbt (dbt Labs)?

dbt (data build tool) handles the "T" in ELT pipelines — it transforms raw data that already lives in your warehouse by compiling SELECT statements into the tables and views your analysts actually query. Built by dbt Labs (founded 2016, Philadelphia), it brings software engineering practices to analytics: version control, automated testing, dependency graphs, and documentation, all expressed in SQL. Over 50,000 teams use dbt today, making it the de facto standard for analytics engineering. In late 2025, dbt Labs announced an all-stock merger with Fivetran, combining the leading data ingestion platform with the leading transformation layer into a combined entity approaching $600M ARR.

Key Takeaways

• dbt transforms data already in your warehouse — it does not extract or load, so you always pair it with a tool like Fivetran or Airbyte.
• Every transformation is a single SELECT statement; dbt handles materialization, dependency ordering, and schema management automatically.
• Built-in data quality tests (not_null, unique, referential integrity) run alongside every build, catching issues before they reach dashboards.
• The dbt Analytics Engineering Certification has become a recognized hiring credential, appearing in job posts at salaries from $114k to $215k.
• The 2025 Fivetran merger created a single company owning both the ingestion and transformation layers of the modern data stack.

How dbt Works in a Modern Data Stack

The pattern mirrors how developers deploy application code: write locally, test in CI, deploy to production. A dbt project is a folder of .sql files, each containing exactly one SELECT statement that defines a model. You reference other models with `{{ ref('model_name') }}` instead of hardcoded table names, and dbt builds the full dependency DAG automatically. When you run `dbt build`, it resolves that graph, runs tests on each node, and materializes results as tables or views in your warehouse. The Jinja templating layer lets you write reusable macros — think stored procedures, but portable across Snowflake, BigQuery, Databricks, and Redshift without dialect rewrites.

One practitioner detail worth knowing: dbt's compile phase processes Jinja and resolves refs, but it does not validate SQL syntax. Errors only surface when the warehouse actually tries to execute the query. Plan your CI strategy accordingly — catching type mismatches and column renames requires running `dbt build` against a development environment, not just `dbt compile`.

dbt vs. SQLMesh vs. Dataform

dbt remains the dominant choice for talent availability, ecosystem integrations, and documentation depth. If your hiring pool and tech stack are already oriented around it, switching carries real cost.

SQLMesh is the strongest technical challenger. It parses raw SQL via SQLGlot, enabling compile-time validation and automatic column-level lineage that dbt currently lacks without the Fusion Engine. A Databricks benchmark showed SQLMesh 9x faster and cheaper on equivalent workloads. Choose SQLMesh when pipeline scale and CI compute costs are primary concerns — and when you can tolerate a smaller hiring pool. The situation became more complicated in 2025: Fivetran acquired SQLMesh before merging with dbt Labs, so both tools now technically share a parent company.

Dataform is Google's answer to dbt, now embedded inside BigQuery Studio. It solves the same problems with a near-identical mental model but is only a realistic option for teams fully committed to GCP.

Pricing

dbt Core (the open-source CLI) is free and will remain so — the Fivetran merger explicitly committed to preserving the open-source license. dbt Cloud Developer is also free for one seat, covering up to 3,000 successful model builds per month, which is plenty for individual contributors and learning environments.

The Starter plan costs $100/user/month for up to five seats, with 15,000 model builds included and overages billed at $0.01 per additional build. This is where teams hit surprises: run dbt on every pull request in CI, and a pipeline with 200 models can accumulate 4,000+ builds per month from PRs alone — before any production runs. Enterprise is contracted pricing with 100,000+ monthly builds, multi-project Mesh support, and advanced Insights features including the Semantic Layer. Enterprise+ adds PrivateLink, IP restrictions, and hybrid deployment for regulated environments.

dbt in the Fractional Talent Context

dbt is one of the cleaner tools to hire fractionally because the work has a natural project shape: stand up a new data stack, refactor a legacy transformation layer, or migrate from raw SQL procedures to governed dbt models. Most of these engagements run three to six months and produce a self-documenting codebase the internal team can maintain afterward.

On Pangea, companies typically post dbt roles alongside Snowflake or BigQuery — the warehouse is almost always part of the ask. Senior dbt engineers are often paired with data platform architects for greenfield builds, or brought in as individual contributors to accelerate a team that has the warehouse but lacks transformation discipline. The analytics engineer title that dbt effectively created now anchors a recognizable career path, so fractional talent in this space tends to be well-defined and easier to evaluate than generalist "data engineer" roles.

The Fivetran Merger: What It Means for Teams Using dbt

The October 2025 all-stock merger between dbt Labs and Fivetran is the most significant consolidation event in the modern data stack since Snowflake's IPO. The combined company sits at ~$600M ARR with a16z as a shared backer; George Fraser (Fivetran CEO) leads the unified entity, with dbt Labs founder Tristan Handy as President.

For existing users, the near-term impact is minimal — both products keep their names, open-source commitments stand, and no disruptive changes are planned. The longer-term story is tighter end-to-end integration: 80–90% of Fivetran customers already use dbt, so the combination formalizes a de facto partnership. The more interesting wrinkle is SQLMesh: Fivetran had already acquired Tobiko Data (SQLMesh's creator) before the dbt merger closed, meaning the new parent company now owns both the incumbent and its most credible open-source challenger. How product teams reconcile that is an open question in 2026.

The Bottom Line

dbt has done something rare: it invented a job title (analytics engineer) and made itself the prerequisite skill for that entire category. For companies running a cloud data warehouse, dbt is the standard way to build a governed, testable, version-controlled transformation layer — not one option among many. The Fivetran merger adds distribution muscle without changing the core value proposition. For freelancers and fractional engineers, dbt proficiency combined with Snowflake or BigQuery experience is one of the most consistently in-demand skillsets in data infrastructure today.