AWS Bought DuckDB's Maintainers, Not DuckDB

AMSTERDAM, 29 AUG 2026 — Amazon Web Services has agreed to acquire DuckLabs, the company behind the open-source analytical database DuckDB. The project itself stays MIT-licensed under an independent foundation, and the people who set its technical direction now work for Amazon.

Most coverage of the deal has carried the first half of that sentence and not the second.

What was bought, and what was not

AWS acquired DuckLabs, the Amsterdam company that employs DuckDB's creators and does the bulk of its development. It did not acquire DuckDB itself, whose intellectual property sits with the independent DuckDB Foundation and remains available under the MIT licence.

Hannes Mühleisen and Mark Raasveldt, who created the database and co-founded the company, continue to lead the team and set the project's technical direction. They stay in Amsterdam. Financial terms were not disclosed, and the deal is expected to close subject to customary conditions.

DuckDB is an in-process analytical database, which means it runs inside the calling program rather than as a server you connect to. That design choice is why it spread. A developer can scan a few hundred million rows in a laptop process, without provisioning a warehouse, waiting for a cluster or paying for either.

The licence is the protection, not the foundation

The public reassurance rests on governance — an independent foundation oversees the project, and AWS says it will keep it that way. Foundations are worth having. What actually holds here is the licence.

MIT is irrevocable for code already published. Every version of DuckDB released to date can be copied, modified and redistributed by anyone, for any purpose, in perpetuity, and no acquisition changes that. If the project's direction became unacceptable to its users tomorrow, the code could be forked tomorrow.

A foundation, by contrast, is a set of people and by-laws. It can be restructured, its board can change, and its independence is a description of current arrangements rather than a property of the artefact. We have written before about where a piece of software actually keeps its licence, and the answer is rarely the place the announcement points to.

Whether the foundation stays independent matters less than whether a fork could be sustained if it stopped being, and that depends on the maintainers rather than the paperwork.

Maintainer attention is the asset

Anyone may fork an open-source project. Very few can maintain the fork. A fork inherits the code and none of the people who understand it, and for a database written in C++ with a vectorised execution engine, that gap is not closed by enthusiasm.

AWS bought the small group of people who hold that understanding, all on one payroll. The commitment that the founders will continue to set technical direction reads as a guarantee of continuity, and it is also a description of where the influence now lives. Nobody needs to instruct anyone. A maintainer employed by a cloud provider will, over years, find the problems that provider has interesting more tractable than the ones it does not.

This is the same shape as Stripe buying the layer that routes AI model requests. In both cases the acquired thing is small, neutral infrastructure that a much larger company would prefer not to have owned by someone else.

This is not a private reading. Ashish Chaturvedi of HFS Research warned developers not to mistake an unchanged licence for an unchanged project, on the blunt ground that “paychecks bend roadmaps”. He advised treating AWS as DuckDB's owner in practice when planning multi-cloud architecture.

Why Amazon wanted it

AWS has been working with DuckLabs since 2024, and DuckDB already powers the dashboarding engine inside Amazon Quick, where the company says it has cut query latency by about 30 per cent across 2.5 billion queries since October 2025. This was not a speculative purchase. AWS bought something already load-bearing in its own infrastructure.

The stated plan is to pair DuckDB's handling of everyday queries at a terabyte or less with S3's exabyte scale, and to look at integrating it across Redshift, Athena, EMR and Glue. That is Amazon closing the gap at the small end of its own analytics range, where its answer for years has been to spin up a cluster for a job that never needed one.

There is a second motive that the announcement does not state and that follows from what the tooling is now used for. An AI agent doing exploratory analysis issues many small queries and needs them answered in milliseconds. A warehouse billed per query suits neither the volume nor the latency. An engine running inside the process does.

The company caught in the middle

MotherDuck is a separate, venture-backed company that has raised around US$100m to sell DuckDB as a managed cloud service, founded by Jordan Tigani, previously of Google BigQuery. Its product runs on AWS infrastructure, and the engine underneath it is now maintained by an AWS subsidiary.

Its public response has been to welcome the deal and to start selling something it had previously declined to sell. MotherDuck says it is now offering enterprise support for DuckDB, having stayed out of that market to avoid competing with DuckLabs, and that it does so with the founders' explicit blessing now that they are moving to Amazon.

This could be read as a company making the best of an awkward week. It is also the clearest evidence of what actually changed. The reason MotherDuck avoided enterprise support was that the maintainers sold it. That constraint is gone, because the maintainers now have an employer with different priorities — and support contracts for a free database are not among them.

What to watch

The commitments made this week are real and they are also the easy ones. Licences and foundation stewardship cost nothing to promise on day one.

The signals that matter come later. Will DuckDB's integrations with competing services — it reads from Google BigQuery and Sheets today — receive the same maintenance as the AWS ones? Will the promised technical advisory board, which would give other stakeholders a formal channel, actually be constituted? Will performance work that helps every user keep landing at the same rate as work that helps S3?

None of those will be announced. They will be visible in the commit history, which stays public whoever signs the cheques.