TurnClaude's answersChatGPT's answersGemini's answersintoauditable work

This matters enormously for the price. AWS explicitly is not acquiring the DuckDB open source project, which stays free and open source under the independent DuckDB Foundation on the MIT licence. DuckLake and Quack are excluded too.

So the purchased asset is a team, a brand, and a book of commercial contracts. No IP, no licence control, no ability to close the project. That caps the price far below what a "we bought DuckDB" headline would suggest.

DuckLabs is just over five years old, has more than thirty employees, was never financed with external capital, and remained wholly owned by the founders and staff, funding growth through support and development contracts. No preference stack, no board pushing for a mark, no fund-life clock.

I don't have their filed accounts, so I'll infer revenue from the cost base. Thirty-plus senior C++ database engineers in Amsterdam is roughly €4.5m to €5m fully loaded. They've been growing headcount without capital, so they're profitable. That puts revenue somewhere around €7m to €10m, call it $9m. This is support and prioritisation revenue rather than SaaS ARR, which matters for the multiple.

Revenue multiple. Services-flavoured revenue earns 3x to 8x. That's $27m to $72m. This is the floor, the price of the business as a business.

Per head. Elite systems talent in a non-competitive process runs $2m to $4m a head in 2026. Thirty-two heads gives $65m to $128m.

Closest comparable. Snowflake paid about $250m for Crunchy Data, which generated over $30m annualised. That was a hundred-person team of Postgres experts. Scale it by headcount and you get $80m. Scale it by revenue at the 8.3x price-to-ARR that implies and you get $75m. Two independent scalings landing in the same place is reassuring.

Scarcity premium. The other pole is Databricks paying roughly $1bn for Neon, which analysts put at possibly over thirty times revenue. That's what competitive tension does. DuckDB is genuinely one of one, downloaded millions of times a day, and AWS knew Databricks, Snowflake and Google could all have called.

Around $150m in total consideration, probably split as $80m to $100m of purchase price plus $50m to $70m of retention RSUs vesting over four years. Amazon has to lock thirty people in Amsterdam, and Mühleisen holds a professorship he's giving up, so the retention component is likely unusually large relative to the headline.

My 80% range is $60m to $350m. Below $60m the founders would have been leaving obvious money on the table. Above $350m you'd need real competitive tension, and the deal doesn't smell like an auction. AWS and DuckLabs had been working together since early 2025 on S3 Tables and SageMaker Lakehouse support, which reads as a bilateral, relationship-led process.

One thing I checked and then had to discard. DuckLabs expects to become part of AWS in early September, roughly a fortnight after signing, which is shorter than the thirty-day HSR waiting period. The 2026 size-of-transaction threshold is $133.9m. That looks like a hard ceiling until you remember DuckLabs is a Dutch B.V., and acquisitions of foreign issuers are usually exempt regardless of price. So the fast close is weak evidence rather than proof. It nudges me down, it doesn't cap me.

Two things would sharpen it a lot. Dutch companies file accounts at the KvK, so DuckDB Labs Amsterdam B.V. (company number 83657150) should have real turnover figures on file, which would collapse most of my revenue uncertainty. And Amazon's Q3 10-Q will show business acquisitions net of cash acquired, though a deal this size will vanish into the noise unless it's much larger than I think.