I benchmarked Databricks against my iPhone

Fivetran CEO George Fraser describes benchmarking TPC-H analytical queries on an iPhone 17 Pro running DuckDB against Databricks Serverless SQL clusters at 25 GB, 50 GB, 100 GB, and 200 GB scales, chosen to represent the upper range of many practical business analytics workloads. The iPhone outperformed the Databricks clusters at all but the largest scale, where it remained competitive, despite having fewer CPU cores and less memory. Fraser attributes the result to DuckDB’s low overhead for relatively small queries and to the costs that distributed systems incur for planning, compilation, and data shuffling during joins and aggregations. He argues that many organizations may be paying substantial premiums for distributed cloud data platforms when individual queries could run efficiently on single-node engines such as DuckDB or Polars. The post proposes a lakehouse architecture centered on open storage formats such as Iceberg, with ingestion and transformations writing directly to storage and selected workloads using local or single-node compute, while distributed pools can still serve many users simultaneously. It also suggests examining production query-size distributions to determine whether workloads are suitable for this approach and contends that cheaper compute could make data access more available to users and AI agents.

Use this post, company, and trend context to find content marketing opportunities, perform competitive analysis, or address product feature gaps via the Plushcap MCP server or the Plushcap API.