What is a data lakehouse? A plain-English definition

A data lakehouse is a single store that keeps raw files cheaply, like a data lake, but manages them as reliable tables you can query with SQL, like a data warehouse. One copy of the data serves both reporting and data science. In Microsoft Fabric, a lakehouse is an item you create in a workspace, stored in OneLake.

Why did the lakehouse appear?

Before it, larger organisations usually ran two systems. A data lake held raw files of any shape cheaply but had no real guarantees: a half-written load could leave readers with inconsistent data. A data warehouse held clean, structured tables for reporting but was a poor fit for messy or very large data. The Databricks glossary describes the cost of running both: duplicate copies, extra infrastructure and data that is often stale by the time it reaches a report.

The lakehouse closes that gap with an open table format. In Fabric that format is Delta Lake, which Microsoft Learn says adds ACID transactions, schema enforcement and "time travel" (querying an earlier version of a table) on top of ordinary files.

How is a lakehouse different from a data warehouse?

In Fabric the line is narrower than the names suggest. According to Microsoft's lakehouse vs warehouse decision guide (updated September 2026), both store data in Delta format in OneLake and share the same SQL engine. The differences are how you build and what you store:

	Lakehouse	Warehouse
Main build tool	Spark notebooks (Python, SQL, R, Scala)	T-SQL
Data types	Structured and unstructured	Structured
Multi-table transactions	No	Yes
SQL access	Read-only SQL analytics endpoint	Full read and write T-SQL
Best for	Landing and cleaning data, data science	BI reporting by SQL-first teams

A common pattern is the medallion architecture: raw data lands in a bronze layer, is cleaned into silver, and business-ready tables sit in gold. Microsoft also notes you can land and transform data in a lakehouse, then expose curated tables to a warehouse.

What does a worked example look like?

Take a distributor with sales in an ERP, web orders in Shopify exports and stock levels in a supplier's daily CSV.

1. Land: a pipeline or Dataflow Gen2 copies each source into the lakehouse Files area overnight, unchanged.
2. Clean: a notebook or dataflow standardises product codes and dates, then writes Delta tables into the Tables area. Fabric registers them automatically, so they are immediately queryable.
3. Model: the gold tables become the fact and dimension tables of a star schema.
4. Report: a Power BI semantic model reads them in Direct Lake mode. Per Microsoft Learn, a Direct Lake refresh copies only metadata, so new data appears without a long scheduled refresh.

The product-code fix happens once. Every report, and any later AI or forecasting work, reads the same cleaned table.

Does a small business need one?

Usually not on day one. A lakehouse lives inside Microsoft Fabric, and Direct Lake needs a Fabric capacity: Pro licences alone are not enough. As of October 2026, Microsoft's guardrails for the smallest F2 capacity allow up to 300 million rows per table and a 10 GB model in Direct Lake, which is far beyond what most SME reports need.

The honest test, and this is our opinion rather than a Microsoft rule: if your data is a few well-kept sources feeding a handful of reports, Power BI on its own is cheaper and simpler. A lakehouse earns its place when several messy sources need cleaning once and reusing many times. Our Fabric vs Synapse vs "just Power BI" guide walks through that decision, and Fabric pricing covers what a capacity costs.

We design lakehouses and pipelines as part of our data engineering work, within the wider data practice.

Frequently asked questions

What is a data lakehouse?

A data lakehouse is a single store that keeps raw files cheaply, like a data lake, but manages them as reliable tables you can query with SQL, like a data warehouse. One copy of the data then serves both reporting and data science, instead of being copied between two separate systems.

What is the difference between a lakehouse and a data warehouse?

In Microsoft Fabric both store data in Delta format in OneLake. The lakehouse handles structured and unstructured data and is built mainly with Spark notebooks; the warehouse handles structured data, is built with T-SQL and supports multi-table transactions. Many teams use both.

Do I need a lakehouse to use Power BI?

No. Power BI Pro with Import mode connects to Excel, SharePoint, SQL Server and many other sources without any lakehouse. A lakehouse becomes useful when several sources need cleaning once, in one place, before many reports use them.

Can Power BI read a Fabric lakehouse directly?

Yes. Every Fabric lakehouse gets a read-only SQL analytics endpoint, and Power BI semantic models can use Direct Lake mode to load its Delta tables into memory without a full scheduled refresh. Direct Lake needs a Fabric capacity (an F SKU); it is not available on Pro alone.

---

Easy Insight is a UK consultancy for AI, web, apps and data — senior specialists only, no juniors.