Lakehouse vs warehouse vs data lake
What is the actual difference between a data lake, a data warehouse, and a lakehouse?
A data lake gives you cheap storage without guarantees, a warehouse gives you guarantees without engine choice, and a lakehouse adds the guarantees to the lake so you keep both.
| Data lake | Data warehouse | Open lakehouse | |
|---|---|---|---|
| Where data lives | Files in your object storage | Inside the warehouse, proprietary internal format | Files in your object storage |
| ACID transactions | No | Yes | Yes, via the table format |
| Schema enforcement | On read, if at all | On write, strict | On write, via the table format |
| Engine choice | Any engine that reads files | The warehouse engine | Any engine implementing the format |
| Concurrent writers | Unsafe, silent corruption | Safe | Safe, snapshot isolation |
| Time travel and rollback | No | Varies, often limited retention | Yes, snapshot based |
| Storage cost | Lowest | Highest | Lowest, same as a lake |
| Cost of leaving | Near zero | High, requires export and conversion | Near zero |
| Typical failure mode | Becomes a data swamp | Becomes a bottleneck and a bill | Under-maintained tables and small files |
Choosing
- You need cheap retention and will process with your own code A lake is sufficient, and you accept no guarantees
- You have one team, one engine, and value operational simplicity above all A warehouse is the least work, and you accept the lock-in
- Several engines or teams need the same governed data A lakehouse, because engine choice is the whole point
- You already have a lake with correctness problems A lakehouse, since a table format retrofits over data you already have
Caveats
- Warehouses increasingly read and write open table formats, which narrows this gap considerably. The distinction is becoming about where the data lives by default rather than what the system can technically read.
- A lakehouse moves work rather than removing it. Compaction, snapshot expiry, and file sizing are your responsibility in a way they are not in a managed warehouse.