AWS Buys DuckLabs as AI Agents Change How Data Gets Read

Most companies running DuckDB today never chose it. They bought analytics products with it already inside, and it never came up. I keep finding it in products I cover. DuckDB is a free analytical database, which means software built to answer questions across a large pile of data. Normally, there is no server to install and no cluster to run. A developer drops it into a program like any other code library, and it reads data directly from cloud storage.

For years, that approach was super convenient for developers. It was a way to work fast without waiting for infrastructure. AI agents have since made it more of a cost question. An agent explores data much the way a person does, poking at small sets before it settles on what it wants. It just does that far more often and throws away most of the answers. On a cloud data warehouse, that pattern gets expensive fast. DuckDB runs inside the application instead, and AWS just bought the company behind it.

The warehouse was never priced for the way agents read

Traditionally, a cloud data warehouse charges you for compute resources that are provisioned and running, regardless of question difficulty. Serverless tiers soften that, though you still get a bill for the compute each query uses. None of it was a problem when the people asking questions were people. A team of analysts sends a few hundred deliberate queries a day, and the pricing was built around exactly that. Point a fleet of agents at the same warehouse, and the bill scales with thousands of small throwaway questions and answers that will never even be used.

DuckDB doesn’t have to worry about keeping a cluster alive. The queries run where the agent is already working, so nothing has to start up and nothing sits idle between questions. You still pay for the machine and for the storage requests the queries make. The idle capacity charge disappears. DuckDB reads the data where it already sits in cloud storage, so there is no copy step and no second version to maintain. One rentable server now carries enough memory, cores, and network bandwidth to handle most of these small queries on its own.

DuckDB does not solve governance on its own. An agent reading straight out of storage still has to be told what it is allowed to see, and those controls live in the services around the engine.

DuckDB was already growing past the laptop before AWS arrived

DuckLabs has previewed DuckDB 2.0, due this fall, and calls it the year of DuckDB as a server. Its Quack extension lets one DuckDB instance serve data to other machines over a network. Most people still embed DuckDB inside a single program. Running one shared copy is where it is heading.

A fleet of agents wants exactly that setup. One copy of the engine, every agent connected to it, all reading the same data in cloud storage, and nothing to stand up when more agents arrive. DuckLabs was building toward that before the deal, and AWS knew, because it was already a customer. It sponsored DuckDB’s Apache Iceberg extension. Iceberg is a standard way of organizing data in cloud storage so different query engines can read the same tables. The two teams also shipped support for S3 Tables, which are Iceberg tables AWS stores and maintains for you.

As a customer, AWS could pay for a feature and wait for DuckLabs to schedule it. Now the people setting the schedule work for AWS.

The Foundation keeps the copyright and DuckDB stays free

The DuckDB Foundation owns the copyright, and AWS is not buying the Foundation. Only the copyright holder can change a license, so nobody at AWS can make DuckDB stop being free. Anything running on DuckDB today keeps running, at no cost, under the same terms.

MotherDuck sells a managed cloud service built on DuckDB and has worked alongside DuckLabs from the start. No independent company has more riding on where the project goes. It responded by beginning to sell enterprise support for DuckDB, a business it had left alone because DuckLabs depended on the revenue. That is a bet on DuckDB staying broadly useful.

DuckDB ships extensions that reach into services on competing clouds, and that matters to anyone running across more than one. The Foundation is adding a technical advisory board that gives community members a formal say in technical direction. There is also a plan to let extensions signed by developers outside DuckLabs run in DuckDB. I’d love to see that board seated with people well outside AWS and DuckLabs. We’ll see what happens there.

AWS now sits underneath analytics it does not run

The deal adds no product to the AWS catalog. It puts AWS engineers behind an engine that runs inside other companies’ software, on competing clouds, and on laptops that will never touch AWS. AWS ends up funding the thing its competitors ship inside their own products, and it still comes out ahead.

AWS has been working to make S3 a place where you query data directly, and S3 Tables was a big step in that direction. DuckDB is the engine that does the reading. That puts pressure on the market tier that sells compute and governance on top of stored data. AWS gets there without building a new warehouse to do it.

Odds are, AWS ships a managed DuckDB service, which puts it in direct competition with independent vendors already selling one. DuckDB has a quieter path in, too, showing up inside services customers already pay for, so queries get faster without anyone having to migrate anything. I’d take the second one. Most people will never see any of this happen. DuckDB will get faster inside products they already bought, and agent workloads that would have been too expensive to run become routine. Nobody buying those software products picked the engine inside them. The teams that built the products did, and AWS just paid to prove them right.

Mike Leone
VP & Principal Analyst | + posts

Mike Leone is a principal analyst at Moor Insights & Strategy covering data platforms and analytics, data infrastructure and storage, and data governance and enterprise data strategy. He brings 15 years of analyst experience from his work at Enterprise Strategy Group, where he rose to practice director for data management, analytics, and AI. Mike's work is grounded in a strong technical and strategic foundation, including early roles in software and hardware engineering.

Recent Posts