Fivetran and dbt Labs Merge

The Fivetran dbt Labs merger is done. The 2 companies closed their all-stock deal on June 1, 2026. George Fraser stays CEO, Tristan Handy becomes President, and the combined company, operating as Fivetran + dbt Labs, says it serves more than 100,000 data teams.^[1] The deal was first announced on October 13, 2025. At that point the companies projected the combined business would be "approaching" $600M in ARR.^[2] The closing release also named the dbt Core v2.0 alpha as a first joint launch and described it as opening the dbt Fusion engine runtime under Apache 2.0.^[1]

My read: the license split will affect your actual work more than the change of ownership will. Your ingestion and transformation layers now share a vendor. The dbt code you write runs on an open runtime, and a proprietary engine sits on top of it. You need to know which part is which, because interviewers have started asking about it in system design rounds.

This piece covers 4 things: what the companies stated, which of those numbers you can lean on, what actually moved to Apache 2.0, and how to talk about vendor consolidation when an interviewer brings it up.

What the Fivetran dbt Labs Merger Announcement Actually States

Here is the deal stripped of the press-release adjectives. Every row comes from the companies' own releases or from trade coverage of those releases.

The company's positioning is "data infrastructure for trusted AI agents." In the closing release, Handy argues that trust in AI systems comes from high-quality tooling built on open standards.^[1] That is his opinion. It may even be correct, but it's a sales thesis and should be read as one.

I've been through a few acquisitions of tools my pipelines depended on. In the first 6 months, nothing changed. The changes showed up at the next renewal, and in which GitHub issues stopped getting responses. Expect the same kind of slow signal here.

How to Read 100,000 Teams and $600M

Both headline numbers come from the companies themselves. Deal terms weren't disclosed,^[3] and none of the material I reviewed includes an audited filing behind either figure. That doesn't make the numbers false. It means they are claims, and you should cite them as claims.

100,000+ data teams. The release doesn't define a "data team." It also doesn't say how a team using both Fivetran and dbt gets counted. If a shared customer counts once, the figure is a union of the 2 customer bases, which is smaller than their sum. I couldn't confirm the counting method either way, so I treat the number as a rough indication of size rather than something to compute with.

$600M. The October 2025 wording was "approaching" $600M in ARR.^[2] That was a forecast for the combined company, made before the deal closed. If you see it described as current revenue, the writer has gone further than the companies did. ARR, revenue, and "approaching" are 3 different statements.

This distinction is worth getting right because candidates repeat vendor metrics in interviews all the time. If you say "Fivetran has $600M ARR" as though you checked it, a sharp interviewer will notice. If you say "the companies projected roughly $600M at announcement," you sound like someone who reads source material.

dbt Core v2 and the Apache 2.0 Fusion Runtime: What Got Opened

The press-release version says the Fusion runtime was open-sourced. dbt Labs' own post is more precise, and the precision matters.

According to dbt Labs, a subset of code it calls "the runtime" is now licensed under Apache 2.0 as dbt Core. The company had previously committed to releasing that code under the Elastic License 2.0 (ELv2). The same post describes dbt Core's code as completely open source under Apache 2.0.^[4] It also says Core and Fusion no longer run as separate engines. They now share a single runtime, and Fusion adds advanced capabilities on top, some free and some premium.^[4]

Fusion itself remains a commercial product. dbt's licensing documentation describes Fusion as proprietary to dbt Labs, licensed under the dbt Product Licensing Agreement.^[5] In 2025, dbt Labs said Fusion's source-available parts would ship under ELv2, a license that bars offering the product as a hosted managed service.^[6] dbt's pages haven't always described Fusion's license the same way. Read the current text before you put it into a procurement doc.

The accurate phrase is "a subset of the Fusion runtime moved to Apache 2.0," not "Fusion went open source." If you mix those up in an interview, you'll get corrected.

dbt Core v2 Ships in 2 Distributions

The upgrade docs describe 2 distributions of Core v2.0. Fusion is recommended for most users. An Apache-2.0-only distribution is available for customers with license constraints. Both use the same project language.^[7]

In practice, your models, macros, and YAML should work on either distribution. Licensing, and whichever premium features you come to depend on, decide which one you run. If your employer's legal team keeps an approved-license list, the Apache-only distribution exists for you.

The Alpha Is an Alpha

dbt Labs calls v2.0 a first alpha.^[4] The dbt-core README warns that behavior, APIs, and on-disk formats may change before the stable release. It also says v2.0 writes Parquet artifacts while still producing a JSON manifest.json for backward compatibility.^[8]

If you have CI checks, lineage scripts, or a homegrown docs site that parses manifest.json, those are where an upgrade will break things quietly. I once spent 2 days tracing a "missing models" bug in a lineage dashboard. It turned out the artifact schema had shifted under us and nobody had pinned the version. Pin your versions. Run the alpha in a branch. Keep it out of production until the label changes.

Modern Data Stack Consolidation: Fivetran's 13-Month Buying Spree

The dbt deal is the largest of several moves. In roughly 13 months, Fivetran assembled every layer of a data stack except compute:

• Activation: On May 1, 2025, Fivetran agreed to acquire Census, a reverse ETL and data activation provider, for cash and equity on undisclosed terms.^[9]
• Transformation, part 1: On September 3, 2025, Fivetran announced it had acquired Tobiko Data, the company behind SQLMesh and SQLGlot.^[10]
• Transformation, part 2: On October 13, 2025, Fivetran and dbt Labs announced their all-stock merger,^[2] which closed June 1, 2026.^[1]

So one company now owns the ingestion connectors, dbt, SQLMesh, and a reverse ETL tool. SQLMesh and dbt overlap directly, and the materials I reviewed don't include a public roadmap commitment for how the 2 will coexist. I won't guess. If your team picked SQLMesh specifically to get away from dbt, follow both projects' release notes closely.

The broader pattern is something I've seen before. The modern data stack broke apart into 15 single-purpose vendors, every one of them priced like the main product, and now it's consolidating again. Neither phase means the industry is in trouble. Bundling and unbundling repeat on a cycle; your job is to keep your architecture able to survive either one.

Takeaway: Consolidation changes who sends the invoice. Your data model is still yours. Keep your layers separable, and know which parts of your stack run on open licenses and which run on a contract.

What Changes for Data Engineer dbt Skills

Short answer: very little that's worth worrying about. If you already write solid dbt, you write solid dbt. The project language is the same across both v2 distributions,^[7] and the skills behind good dbt work were never tied to a particular vendor.

The dbt skills that earn you offers are the same ones they've always been:

• Grain discipline. Know the grain of every model. You can always aggregate up; you can never disaggregate down.
• Incremental logic that survives late data. Merge keys, lookback windows, and what happens when a backfill runs twice.
• Testing that catches real failures. A not_null on a primary key is the minimum. Row-count reconciliation against the source is where the silent 40% drop gets caught.
• DAG design. Staging, intermediate, and marts layers that a new hire can follow at 2am.

None of these depend on who owns dbt Labs. Concepts transfer across tools; tool knowledge doesn't transfer across concepts. If your dbt interview prep is mostly Jinja trivia, switch to dbt interview questions that test modeling judgment. Then go back to the fundamentals in data modeling interview questions, because that's where most dbt candidates actually lose points.

The new skill is license literacy, and it takes about an afternoon to learn. Be able to explain open source (OSI-approved licenses such as Apache 2.0) versus source-available licenses (such as ELv2, which restricts offering the software as a managed service^[6]). Be able to say which dbt features in your project come from the Apache runtime and which come from Fusion. On your resume, name the distribution you ran. "dbt (Fusion)" and "dbt Core" now mean different things to the person reading it.

Answering Vendor Consolidation Questions in Data Engineering System Design Rounds

I've sat on panels where the "which tools would you use?" question decided the round. The candidate who named 6 vendors lost to the one who named 2 and explained what would happen if each disappeared. With Fivetran owning both ingestion and transformation, interviewers have a concrete example to ask about.

A strong answer to "you're building ingestion plus transformation, walk me through your choices" covers 5 points:

• Separate the layers first. Ingestion lands raw data; transformation models it. Explain how you'd swap either one without rewriting the other. This is the core of ETL vs ELT design, and it doesn't change because 2 companies merged.
• Name the license of each component. "Transformations run on dbt Core v2's Apache 2.0 distribution, so the runtime is portable. If we adopt Fusion features, that part is under a commercial license."
• Separate written commitments from roadmap hopes. A license is a written commitment. Future feature investment in the open distribution is a hope. Say which one you're relying on.
• Describe the exit path. Raw data in your own warehouse, models in SQL under version control, no business logic locked in a vendor UI. If pricing doubles, you can migrate in a quarter instead of a year.
• Attribute vendor numbers. "The companies say 100,000+ teams" works. "Fivetran has 100,000 customers" stated as fact doesn't.

That structure works for any vendor question, which is why it's worth practicing. The Data Engineering system design guide covers the pipeline-architecture framing more fully. If you want to practice saying all of this out loud under time pressure, the mock interview simulator on DataDriven is the best practice I've found for it. Explaining trade-offs to a timer is a different skill from knowing them, and most candidates find that out mid-round.

What Data Engineers Should Watch Next

The facts so far are an ownership change, a leadership arrangement, and an alpha release. The next 12 months will show what they amount to. Here's what I'm tracking:

• The stable v2.0 release. Until it ships, the README's warning about changing APIs and on-disk formats applies.^[8] Wait for the stable release before migrating anything you get paged for.
• The licensing pages. Check whether dbt's docs settle on one consistent description of Fusion's license.^[5]
• Parity between distributions. See whether new capabilities keep landing in the Apache-only distribution, or mostly in the premium side of Fusion.^[4]
• SQLMesh's direction. With 2 overlapping transformation frameworks under one owner, the release notes will tell you more than the press releases.
• Your renewal. Bundles are cheaper until they aren't. Price ingestion and transformation separately before you sign anything combined.

For your job search, prepare to explain the license split in 2 sentences and to describe your exit path for any tool in your stack. Then get back to grain, incrementals, and tests. Vendors will keep merging; you'll still be debugging late-arriving data under whichever logo is on the invoice.

References

1. Fivetran, "Press release: Fivetran and dbt Labs Complete Merger to Create the Data Infrastructure for Trusted AI Agents," June 1, 2026. fivetran.com
2. Fivetran and dbt Labs via Business Wire, "Fivetran and dbt Labs Unite to Set the Standard for Open Data Infrastructure," October 13, 2025.
3. TechTarget, "Fivetran, dbt Labs complete merger to form data layer for AI," reporting on the Fivetran closing release. techtarget.com
4. dbt Labs, "dbt core v2 is here." docs.getdbt.com
5. dbt Labs, "dbt licensing." docs.getdbt.com
6. dbt Labs, "New code, new license: understanding the new license for the dbt Fusion engine," 2025. getdbt.com
7. dbt Labs, "Upgrading to v2." docs.getdbt.com
8. dbt Labs, "dbt-core" repository README. github.com
9. Fivetran, "Fivetran Signs Agreement to Acquire Census, Delivering the First End-to-End Data Movement Platform for the AI Era," May 1, 2025. fivetran.com
10. Fivetran via Business Wire, "Fivetran Acquires Tobiko Data to Power the Next Generation of Advanced, AI-Ready Data Transformation," September 3, 2025.