A lakehouse keeps the lake's storage and adds the warehouse's table guarantees with two thin metadata layers.
The bottom two layers are the lake you already know. The table format is the new idea: a table is whatever its latest metadata file says, a list of data files plus the schema and partitioning that describe them. A write creates new data files and a new metadata file, and nothing becomes visible until the catalog swaps its pointer from the old metadata file to the new one in one atomic step. That compare-and-swap gives you ACID commits (readers see a table before or after a write, never halfway), time travel (old metadata files still describe old versions), schema and partition evolution without rewriting files, row-level updates and deletes, and engine independence: Spark 129 writes, Trino 403,499 and DuckDB 61,228 read, and no engine owns the storage.
The price is operational: metadata and small files need maintenance (Iceberg in Production), and the catalog becomes critical infrastructure (Iceberg Catalogs).