Amazon Web Services has signed a definitive agreement to acquire DuckLabs, the Amsterdam company founded by the creators of DuckDB, with closing expected in early September 2026. The detail that matters for anyone with DuckDB in a pipeline is what is not being bought: DuckDB, DuckLake and Quack stay free and open source under the MIT licence, governed by the independent non profit DuckDB Foundation. Hannes Muhleisen and Mark Raasveldt keep leading the team and the technical direction of the project, from Amsterdam. AWS says the two sides have worked together since 2024, and that DuckDB already sits inside Amazon Quick.
The short answer
AWS has signed a definitive agreement to acquire DuckLabs, the Amsterdam company behind DuckDB, with closing expected in early September 2026. DuckDB, DuckLake and Quack are not part of the transaction: they stay free and open source under the MIT licence, stewarded by the independent non profit DuckDB Foundation. Founders Hannes Muhleisen and Mark Raasveldt continue to lead the team and the technical direction of the project from Amsterdam. AWS says the two organisations have collaborated since 2024, and that DuckDB integrations inside Amazon Quick cut average query latency by around thirty percent.
Read the announcement twice and the interesting sentence is the one about what is not changing hands. AWS is buying a company in Amsterdam. It is not buying the database, and that looks like the point of the exercise rather than an afterthought.
What was actually signed
AWS has a definitive agreement to acquire DuckLabs, the company founded by Hannes Muhleisen and Mark Raasveldt, the two people who created DuckDB. DuckLabs says it expects to become part of AWS in early September 2026. Financial terms were not disclosed.
DuckDB, DuckLake and Quack are excluded from the transaction. They remain free and open source under the MIT licence, and they continue to be governed independently by the DuckDB Foundation, a non profit that predates this deal by years. The founders keep leading both the team and the open source technical direction, and the team stays in Amsterdam.
That combination, a permissive licence plus a separate foundation holding the project, is about as close as open source governance gets to a structural guarantee. It removes the two levers an acquirer would normally have. It cannot relicense what it does not own, and it cannot apply trademark pressure to a fork it does not control.
Why AWS wanted this specifically
The rationale AWS published is refreshingly narrow. Its argument is that the overwhelming majority of real analytical queries touch a terabyte or less, and that the distributed engines most teams reach for were built for a much larger size class. Below that threshold, the coordination overhead is the workload.
DuckDB is an embedded, in process OLAP engine with vectorised execution. There is no cluster, no scheduler and no JVM. You link it into your process, or you run a single binary, and the query executes where the data already is. For a great deal of day to day analytics work that is not a compromise, it is simply the correct shape.
AWS also disclosed something concrete rather than aspirational. It has been shipping DuckDB inside Amazon Quick since that product launched in October 2025, says it has processed more than two and a half billion queries, and credits DuckDB integrations with cutting average query latency by roughly thirty percent. The stated plan is to pair DuckDB at the everyday query size with the exabyte scale of S3 and the existing AWS analytics services.
There is a second argument in the post that is easy to skim past. AWS notes that AI agents interact with data the way people do, poking at it and running exploratory queries rather than issuing one carefully planned job. An engine with near zero startup cost per query fits that access pattern far better than one that spends seconds provisioning before it reads a byte.
What changes for you, and what does not
If DuckDB is in your build, nothing changes on closing day. Your licence is the same, your redistribution rights are the same, and your right to fork is the same. That is not reassurance, it is just what an MIT licence means.
There are two things worth doing anyway, and both are good practice independent of this news.
The first is pinning. Pin the DuckDB version you build against rather than tracking a floating tag, so that a change in release cadence, in either direction, does not arrive in your CI unannounced.
The second is extensions. DuckDB fetches extensions on demand from a repository by default, which means an ordinary query can quietly become a network dependency at runtime. If you rely on httpfs, parquet or the spatial extension in production, vendor them or host them yourself. That was already the right answer for air gapped and reproducible builds, and it is the one part of the stack where ownership of the distribution path is a practical question rather than a philosophical one.
The signal in the timing
This lands in the middle of a broader reshuffle of the cloud data layer, and it says something about where the value has moved. Buyers are not paying for query engines in the abstract. They are paying for the teams, and for the placement that decides which engine a developer reaches for by default.
It also sits alongside AWS committing to two million more Nvidia GPUs across 2027 and 2028 and Amazon closing Mechanical Turk after twenty one years, which together look less like unrelated news items and more like a company reallocating attention with some deliberateness.
For the rest of us the practical read is smaller and more cheerful. The fastest way to query a few hundred gigabytes on a laptop just acquired a very well funded sponsor, without acquiring a licence change. That is an unusual outcome, and it happened because someone set up the governance correctly years before there was a deal to survive.
Sources and further reading
- AWS to acquire DuckLabs, the Amsterdam based company behind DuckDB, About Amazon, August 25, 2026
- AWS and DuckLabs: Building the future of analytics together, AWS Big Data Blog, August 2026
- DuckLabs to Join AWS, Projects to Remain Open Source, DuckLabs, August 26, 2026
- Amazon to acquire DuckLabs, adding the team behind DuckDB, GeekWire, August 2026
- AWS buys DuckLabs to bring DuckDB's embeddable analytics to more enterprises, SiliconANGLE, August 26, 2026
Frequently asked questions
Does AWS now own DuckDB?
No, and the distinction is the whole story. AWS is acquiring DuckLabs, the commercial company that employs most of the core developers. The software itself is not part of the transaction. DuckDB, DuckLake and Quack remain free and open source under the MIT licence, and stewardship stays with the DuckDB Foundation, a non profit that is separate from DuckLabs and now separate from AWS. That structure was set up years before this deal, which is precisely why it can absorb one. If you embed DuckDB today, nothing about your licence, your redistribution rights or your ability to fork changes on closing day.
Should I change anything in my stack because of this?
Not on the strength of the announcement. The sensible response is the same audit you should run on any dependency that suddenly has a new corporate owner: pin the version you build against, know where your binaries come from, and check whether anything in your pipeline resolves DuckDB or its extensions at runtime rather than at build time. The extension repository is the piece worth thinking about, because extensions are fetched on demand by default. Vendoring the extensions you actually use is good hygiene regardless of who signs the payslips.
Why would AWS want an embedded database at all?
Because most queries are small and AWS says so plainly. Its own framing is that DuckDB is unusually fast for the very large majority of queries that touch a terabyte or less, the size class that distributed engines were never optimised for and where their coordination overhead dominates. AWS already ships DuckDB inside Amazon Quick, which it launched in October 2025 and which it says has processed more than two and a half billion queries, with DuckDB integrations cutting average query latency by about thirty percent. Buying the team turns an integration into a roadmap.
What are DuckLake and Quack?
They are the two other projects in what the team calls the Duck Stack, and both are named in the announcement as staying open source under the same terms. DuckLake is a lakehouse approach that keeps table metadata in a SQL database rather than in a pile of manifest files, which is a genuinely different bet from the designs that dominate the space. Quack is the newer of the three. All three continue under the DuckDB Foundation rather than moving into AWS, which is the part of the announcement that was clearly negotiated rather than assumed.
Has this kind of arrangement held up before?
Sometimes, and the variable is almost always governance rather than intent. An MIT licence plus an independent foundation is close to the strongest structural guarantee available in open source, because it removes both the relicensing lever and the trademark lever from the acquirer. What no structure can guarantee is engineering attention: a foundation can hold the copyright and still watch the release cadence slow if the paid maintainers get pointed elsewhere. That is the thing to actually watch over the next year, and it is measurable. Release frequency and issue response times will tell you more than any press release.