Data lakehouse

A data lake stores everything as it arrives: tables, but also documents, log files and images. That is flexible and cheap, but without structure it soon turns into a swamp in which nobody can find anything. A lakehouse adds the rules of a data warehouse on top: fixed tables, version control and checks. Whether your data foundation for AI needs one depends on what the AI has to do, and for AI that works inside your processes the starting point is a single source of truth.

Platforms such as Databricks and Microsoft Fabric are built around this idea. The advantage is that reports, data science and AI applications use the same data, instead of each keeping its own copy. That makes a lakehouse a form of data platform.

For most mid-sized organisations a lakehouse is not the place to start. It pays off when you have large volumes or many kinds of data and want to run your own analyses on them. For AI that helps with day-to-day work, such as an AI agent that processes a request, good integrations with the source systems and reliable data usually matter more.