AWS buys the company behind DuckDB — and leaves the project with its foundation

AWS has signed a definitive agreement to acquire DuckLabs, the roughly 30-person Amsterdam company behind DuckDB, the in-process analytical database that grew out of the Dutch research institute CWI. The interesting part is what is not being bought: the DuckDB project itself stays open source under the independent DuckDB Foundation, under the MIT licence, with co-creators Hannes Muhleisen and Mark Raasveldt continuing to steer its technical direction from inside AWS. That split — company acquired, project governance left with a neutral foundation — is worth watching as a template for how cloud providers absorb open source teams without absorbing the code.

Technically, the fit is about scaling down rather than up. DuckDB was built on the observation that the overwhelming majority of real-world SQL work touches a terabyte or less, and that engines tuned for petabyte batch jobs handle that badly. Running in-process with vectorised execution, it queries Parquet, CSV and JSON directly on local disk or in S3 — and it turns out to suit AI agents well, because agents poke at small slices of data the way analysts do. AWS is already running it in anger: Amazon Quick has served over 2.5 billion queries on a DuckDB-backed engine since October 2025, cutting average query latency by around 30 percent.

For teams building on AWS, the practical read is that DuckDB is about to become a first-class citizen alongside S3 Tables, Athena, Redshift, EMR and Glue — so the cheap, embedded, single-node end of the analytics stack is no longer a workaround, it is part of the platform. Worth revisiting any pipeline where you reached for Spark because it was the default rather than because the data demanded it.

This is a summary by our content curator. Read the original at AWS Big Data Blog: https://aws.amazon.com/blogs/big-data/aws-and-ducklabs-building-the-future-of-analytics-together/.