Presented at OSACon Nov 2026

Open source has already taken every infrastructure layer below the query. Operating systems, web servers, orchestration, and the transactional database all followed the same path: a neutral interface appeared, the layer commoditized, and value moved up the stack. Analytics is the last holdout, and it is falling now.

Open table formats made storage layout a public contract. Arrow and DataFusion turned the query engine into a shared substrate that anyone can build on. Modern hardware erased the distributed-systems requirement for most real workloads. Object storage became a coordination primitive, which is why the same unbundling is now reaching streaming.

This session maps the trends driving the shift, shows where lock-in moved rather than disappeared, and lays out how to position for it: which layers to own, how to evaluate a project’s governance as an engineering risk, and how to build an exit you may never need to use.


Patrick McFadin is an Apache Cassandra PMC member and ASF Member, and co-author of Managing Cloud Native Data on Kubernetes (O’Reilly). He was at DataStax for over a decade, helping grow one of the largest open source data communities, and stayed through the company’s acquisition by IBM. He now runs McFadin Data and AI Advisory, an independent consultancy working with companies on data and AI infrastructure, advises several data infrastructure startups, and hosts O’Reilly’s What’s Happening in Data. Having watched consolidation from inside the company being consolidated, he has strong opinions about where lock-in goes when it stops being obvious.