AWS Is Buying DuckLabs, the Company Behind DuckDB — the Project Isn't in the Deal, and the IP Stays With the Foundation

Amazon announced on August 26 that it has signed a definitive agreement to acquire DuckLabs, the Amsterdam company behind the embeddable analytical database DuckDB, with the deal expected to take effect in early September and terms undisclosed. The boundary is the story: AWS is buying the company, not the open source project. DuckDB, DuckLake, Quack and the rest stay free and open source under their existing MIT licenses, stewarded by the independent DuckDB Foundation, which was incorporated in 2021 and holds most of the intellectual property. Founders Hannes Mühleisen and Mark Raasveldt continue to lead the team, and core engineering stays in Amsterdam. DuckLabs is a bit over five years old, past thirty employees, never externally financed, and funded by support and development contracts; the two sides have collaborated since early 2025 on S3 Tables and SageMaker Lakehouse. DuckLabs says DuckDB is now downloaded more than three million times a day.

Why the Structure Makes This Different From the Redis Episodes

When open source infrastructure gets acquired or relicensed, users fear one script: the company owns the copyright, commercial pressure arrives, the license becomes BSL or SSPL, and downstream forks. DuckDB's structure doesn't have that path. Most of the intellectual property sits with an independent foundation set up in 2021, and the company existed to provide support and custom development around the project. The foundation isn't part of the transaction, so there is no lever for changing the MIT license unilaterally. DuckLabs also says it intends to expand the foundation's role, set up an advisory board where commercial users can weigh in on the roadmap, and open up the extension ecosystem further. Those are statements for now, worth checking against actual behavior later.

The Variable Isn't the License, It's Maintainer Time

What does change is where attention goes. Thirty-odd engineers now sit inside AWS, with priorities and reviews coming from a new owner, and the S3 Tables and SageMaker Lakehouse line will naturally move up the queue. That isn't necessarily bad for users — integration with object storage and Iceberg has been a persistent ask. The thing to watch is the other end: workloads running on other clouds, the merge rate for community extensions, and whether features that compete with AWS products slow down. Worth noting that DuckLabs never raised outside money, which is rare in infrastructure software. It explains how the project could stay with a foundation without anyone having to answer to investors, and it means this exit carries none of the liquidation-preference machinery that usually shapes such deals.

Why It Matters If You Work on AI

DuckDB sits closer to the AI stack than its profile suggests: reading Parquet locally, cleaning and deduplicating data for RAG, querying a lake straight from a notebook, giving an agent a SQL engine that needs no server. A meaningful share of those three million daily downloads is exactly that kind of embedded use. Nothing to do in the short term — the license didn't change and the binaries are the same. Medium term, the sensible moves are ordinary ones: pin your versions, pay attention to where extensions come from, and if your data stack spans clouds, watch a few months of issues and PRs before drawing conclusions. A foundation holding the IP is real insurance, but what it insures against is a license change, not a change in how much maintenance the project gets.

via: Amazon's announcement, SiliconANGLE, DuckDB's project history page