A data lake is a large-scale storage system that holds raw data in its native format, structured, semi-structured, or unstructured, at low cost on cloud object storage. It applies structure only when data is read, not when it is written. In commercial real estate it stores raw leases, rent roll exports, T-12 files, and market feeds before any are modeled.
How Does a Data Lake Work?
A data lake works by storing data as-is on object storage such as Amazon S3, Google Cloud Storage, or Azure Data Lake Storage, with no schema enforced on ingestion. A lease PDF, a CSV rent roll, and a market data feed all land in the same store untouched. Structure is applied on read, when a query or extraction job interprets a file at the moment it is used.
This schema-on-read model makes the lake cheap and flexible: it holds anything, including data no one has modeled yet. Object storage is inexpensive, for example roughly $0.021 to $0.023 per gigabyte per month on major cloud hot tiers, per 2026 vendor pricing, though retrieval, egress, and request fees can add 30 to 70% to a theoretical storage bill. Data lake growth is driven by unstructured data, which is expected to account for over 80% of new data created as annual global data creation exceeds 180 zettabytes by 2026, per industry reporting via Market.us.
Property | Data lake |
Data shape | Raw, any format |
Schema timing | Schema-on-read, applied at query time |
Primary use | Storage, exploration, ML, extraction inputs |
Storage | Cloud object storage, low cost |
Risk | Can degrade into a data swamp without governance |
Why a Data Lake Matters
A data lake matters because much of what a CRE team receives, scanned leases, PDFs, images, and varied exports, is unstructured and cannot fit a rigid schema on arrival. The lake keeps every source document in its original form, cheaply, so nothing is lost before anyone decides how to use it. It is the raw material layer that extraction and modeling draw from later.
The risk is governance. Without cataloging and validation, a lake degrades into a data swamp where analysts cannot trust which files are complete or current. Because unstructured formats will make up over 80% of new data, per industry reporting, the volume the lake absorbs keeps rising, so metadata and provenance controls are what separate a usable lake from a dumping ground.
Example
A data lake is clearest when varied raw files land together, then are read on demand. A CRE team stores a deal's source documents untouched, then extracts only when needed.
File in lake | Format | Read on demand for |
lease_suite200.pdf | Unstructured PDF | Lease abstraction |
rent_roll_2026q2.csv | Semi-structured CSV | Underwriting model load |
t12_propertyA.xlsx | Spreadsheet | Expense analysis |
costar_market_feed.json | Semi-structured JSON | Comp benchmarking |
None of the four shares a schema, and none is transformed on the way in. Storing all four raw, say 4 files totaling 40 megabytes, costs under a cent per month at roughly $0.021 per gigabyte, since 40 megabytes is about 0.039 gigabytes, or about $0.0008. When underwriting begins, only the rent roll CSV is read and transformed into model rows; the other three wait until a query needs them. The lake paid almost nothing to keep every option open.
Variations and Edge Cases
Data lakes vary by governance and architecture. The cases below change whether a lake stays trustworthy or turns into a liability.
Case | Behavior |
Governed lake | Cataloged, with metadata and access control |
Data swamp | Ungoverned lake where files cannot be trusted |
Lakehouse | Adds schema enforcement and ACID over lake storage |
Raw and curated zones | Separate tiers for untouched and cleaned data |
Cold or archive tier | Cheaper storage for rarely accessed files |
Data Lake vs Data Warehouse
A data lake is often confused with a data warehouse, but they sit at opposite ends of the structure spectrum. A data lake holds raw data of any format at low cost and applies structure on read. A data warehouse holds structured, curated data under a schema enforced on write, so it is fast to query and trusted for reporting.
The lake keeps everything cheaply, including unmodeled and unstructured data; the warehouse holds a clean subset optimized for fast analytics. A CRE team often uses both: raw leases and feeds land in the lake, and the modeled data it underwrites on flows into the warehouse. The emerging lakehouse combines the two, adding warehouse-style schema enforcement over a lake's low-cost storage.
Frequently Asked Questions
What is a data lake?A data lake is a large-scale storage system that holds raw data in its native format, structured, semi-structured, or unstructured, at low cost on cloud object storage. It applies structure only on read. In commercial real estate it stores raw leases, rent rolls, T-12 files, and market feeds before any are modeled.
What is the difference between a data lake and a data warehouse?A data lake holds raw data of any format at low cost and applies structure on read. A data warehouse holds structured, curated data under a schema enforced on write, so it is fast to query and trusted for reporting. The lake keeps everything cheaply; the warehouse holds a clean, modeled subset for analytics.
What is a data swamp?A data swamp is a data lake that has lost governance, where files accumulate without cataloging, metadata, or validation, so analysts cannot trust which datasets are complete or current. Metadata and provenance controls are what keep a lake usable rather than letting it degrade into a swamp.
Related Terms
Data Warehouse
ETL
Data Pipeline
Structured Data Extraction
Data Validation