The people shaping DuckDB’s 2026 rise

The CWI Research Foundation

Mark Raasveldt and Hannes Muhleisen began DuckDB in 2018 as researchers at CWI Amsterdam. They built an in-process OLAP engine that runs inside the host application process. This eliminates the need for a separate database server or complex installation. The engine uses a columnar-vectorized execution model to process batches of values. This design reduces the CPU overhead found in traditional row-based systems.

It scales beautifully.

The engine relies on columnar storage to optimize scan speeds. By storing data column by column, DuckDB minimizes the amount of data the CPU must read. This allows the system to skip columns not required by a query. DuckDB 1.5.5, released in July 2026, demonstrates this efficiency. The project reached 40,000 GitHub stars in August 2026. While SQLite handles point lookups in fractions of a millisecond, DuckDB outperforms it on analytical scans. A 850,000-row aggregation took DuckDB 2.90 ms, while SQLite required 2,722 ms.

The architecture draws from decades of database research. The execution engine takes inspiration from the MonetDB/X100 paper. The optimizer uses techniques from dynamic programming and unnesting research. This portability allows the engine to run on everything from resource-constrained edge devices to large memory servers with 100+ CPU cores. Because it has no external dependencies, deployment remains simple. DuckDB offers APIs for C, C++, Go, Python, R, Rust, Java, and Node.js. The project development began as the team worked as public servants in the Netherlands. They release the software under the permissive MIT license. The intellectual property remains with the DuckDB Foundation.

Scaling through MotherDuck and AWS

Jordan Tigani leads MotherDuck, a Seattle-based startup that provides a collaborative serverless analytics platform. Tigani previously worked as a founding engineer at Google BigQuery and a product officer at SingleStore. MotherDuck raised $100 million in total funding to expand its services. This includes a $52.5 million Series B round led by Felicis in 2023. The company reached a $400 million valuation. Amazon acquired DuckLabs in early September 2026 to bring the development team into Amazon Web Services. Muhleisen and Raasveldt will continue to lead the project’s technical direction from Amsterdam. The DuckDB Foundation retains control over the MIT licensed intellectual property.

Ryan Boyd and Tino Tereshko joined the founding team to build the platform. Boyd brings experience from Google, while Tereshko previously served as a vice president of product at Firebolt. MotherDuck targets existing DuckDB users, data analysts, and software developers. The platform provides read scaling via dedicated DuckDB instances and various instance sizes for configuration. It also converts natural language questions into traceable SQL queries with sandboxed compute. The startup has 30 employees and plans to increase to 45 people by the end of the year.

Amazon’s acquisition signals a shift toward analyzing data directly in cloud storage. This strategy puts pressure on companies like Snowflake and Databricks.

The engine remains versatile.

The Economics of the Duck Stack

You know that cloud data warehouses often generate high bills due to complex pricing. Most businesses do not manage petabyte-scale data. For these users, the scale-up approach of DuckDB provides a cheaper alternative to distributed architectures. Snowflake and Amazon Redshift charge for compute time using 60-second minimum billing increments. This penalizes interactive workloads where queries might only take a few seconds.

While Snowflake and Amazon Redshift impose a 60-second minimum billing increment that penalizes short queries, DuckDB allows users to run analytical queries on their local machines or in the cloud without these hidden architectural costs.

Companies face heavy costs for moving data between cloud providers. Egress fees often range from $90 to over $150 per terabyte. DuckDB avoids these costs by reading Parquet files directly from object storage. This ability to query data where it lives bypasses the need for expensive data movement.

The single-process design of DuckDB limits its ability to handle multiple concurrent writers. It fails at high-concurrency transactional workloads. DuckDB remains the best option for analytical processing on datasets measured in gigabytes. Will the Quack protocol eventually allow DuckDB to replace fully distributed cloud warehouses?