AWS buys DuckDB's engineers; the Foundation keeps the license

A foundation that holds a license and a company that holds a payroll are two different kinds of control over a codebase, and only one of them is changing hands. DuckDB, DuckLake and Quack stay with the DuckDB Foundation under the MIT license ^[5], so code already published cannot be pulled back or relicensed away from the people using it ^[15]. What AWS would buy, if the deal closes, is the employment of the more than 30 people who decide what gets written next ^[3], plus the continued technical leadership of the two researchers who started the project at CWI in 2018 ^[4][7].

The money says more about governance than the license does. DuckLabs was created in 2021 because contracts for feature development were showing up ^[10], and its revenue was enterprise support, technical advice, feature development and engineering partnerships ^[11]. Paid feature work is a roadmap channel, and it was open to anyone with a budget. Inside AWS that channel becomes an internal allocation decision, arbitrated by someone the announcements do not name, at a price they also do not state ^[6]. Runtimewire frames the test as whether the roadmap stays broadly governed once AWS employs the core team ^[14]. The material needed to answer it has not been published.

There is one useful precedent in the work already done. The March 2025 integration gave DuckDB support for Amazon S3 Tables through Apache Iceberg REST catalogs ^[8], and that is a published interface rather than a private one, so effort spent against it is reusable by anything else implementing the same spec. That is the shape of AWS-funded roadmap work that costs non-AWS users nothing. The shape that would cost them something is a capability that only resolves against AWS endpoints.

The track record being used to justify the deal does not have one agreed length. AWS dates the technical relationship to 2024 ^[16], DuckLabs described more than a year of collaboration ^[17], and Andy Warfield, an AWS vice president and distinguished engineer, put it at roughly two years ^[18]. Those reconcile only if the work began in the opening weeks of 2024, because an August 26 announcement leaves about twenty months at the outside ^[19]. A small gap, but the collaboration history is doing argumentative work here.

The bottleneck the founders described was commercial rather than technical: they did not want DuckLabs standing in the way of companies building businesses on DuckDB ^[13], and AWS answers that with a customer base, infrastructure and a support organisation they were never going to build themselves ^[9]. The demand side looks real, though the figure they cite, more than a million downloads a day, is company-supplied and the announcement does not say how it was measured ^[12]. What remains unresolved is the Foundation's position. Its strongest lever is that a fork stays legal ^[15], and a fork needs funded engineers, which is precisely the asset being sold ^[3].