AWS is acquiring DuckLabs, the company behind DuckDB. Both parties say DuckDB remains open source.

That claim is worth examining rather than accepting, because the phrase has preceded several relicensings in recent years. In this case the structure behind it is unusually solid.

Why the structure matters

DuckDB is MIT-licensed and its intellectual property is held by the DuckDB Foundation, a Dutch non-profit, deliberately separate from the commercial entity. DuckLabs was the company employing many of the core developers and selling services around the database; it did not own the database.

This arrangement was set up specifically so that an acquisition of the company would not be an acquisition of the project. Compare the alternative: where a single company holds the copyright, a change of ownership can mean a licence change on the next release, with the community’s only recourse being a fork from the last permissive commit. That is the path that produced the relicensing dramas around several well-known databases.

MIT is also about as permissive as licences get, and it is irrevocable for code already released. Every existing version of DuckDB stays MIT forever regardless of what anyone decides later.

What could still change

The licence is safe. The things that are not guaranteed by a licence:

Development direction. The people writing DuckDB will be AWS employees. Nothing forces the roadmap to keep prioritising the standalone single-node use case over integration with AWS services. Priorities follow paychecks, usually without anyone deciding they should.

The pace of the open version. A common pattern is that the open-source core stays open while the surrounding work, the connectors, management tooling, and performance features, appears first or only in the commercial offering.

Community contributions. Corporate stewardship changes who reviews patches and what gets accepted. Sometimes for the better, with more resources and faster reviews. Sometimes not.

Why people care about DuckDB specifically

DuckDB earned its following by being genuinely good at something awkward: analytical queries on local data without a server. It is an in-process OLAP database, the analytical counterpart to SQLite, and it made a lot of data work simpler.

# query a parquet file directly, no server, no import
duckdb -c "SELECT count(*) FROM 'data.parquet' WHERE status = 'active'"

For anyone who has stood up a database cluster to answer a question about a 2GB CSV, that is a meaningful improvement. It runs comfortably on a laptop or a small VPS, which is a large part of why the self-hosting community adopted it.

The reasonable position

Cautious optimism seems right. The foundation structure is real protection, not a press release, and it is the arrangement people should be asking for whenever a company builds a business around an open-source project.

The thing to watch is not the licence, which is settled. It is whether the next two years of commits keep serving the person running DuckDB on their own machine as well as they serve the person running it inside AWS.