Data fabric

What is a data fabric?

A data fabric is an architecture for connecting data that is spread across systems, clouds, platforms, and business domains. It does not mean moving everything into one database first. It means adding a smart layer that knows where data lives, what it means, who owns it, how it is related, and how it may be used.

The core ingredient is metadata: data about data. A data fabric collects metadata about sources, tables, reports, pipelines, ownership, usage, quality, sensitivity, and lineage. It then uses that metadata to make data easier to discover, connect, govern, and automate.

Gartner describes data fabric as a design concept for an integrated layer of data and connecting processes, using continuous analysis of metadata to support reusable datasets across environments. In less analyst-heavy language: a data fabric is the connective tissue over a messy data estate.

Data fabric versus Microsoft Fabric

The names are easy to confuse, but they are not the same thing.

Data fabric is an architecture concept. It can be built with many tools and many vendors. It is about metadata-driven integration and governance across distributed data.

Microsoft Fabric is a Microsoft SaaS analytics platform. It brings experiences such as Data Factory, Data Engineering, Data Warehouse, Real-Time Intelligence, Data Science, and Power BI together around OneLake and Fabric capacity.

You can use Microsoft Fabric as part of a data-fabric-style architecture. You can also build a data fabric with Snowflake, Databricks, IBM, Denodo, Collibra, Atlan, OpenMetadata, Purview, custom services, or a mix of tools. The concept and the Microsoft product share a word, not an identity.

The parts of a data fabric

Data catalog
The catalog is the searchable map of assets: datasets, reports, APIs, tables, data products, owners, definitions, and classifications.

Active metadata
Active metadata is metadata that is continuously collected and used. It is not just documentation. It can drive recommendations, quality checks, impact analysis, policy enforcement, and automation.

Lineage
Lineage shows how data moves from sources through transformations into reports, models, and applications. It lets teams understand impact before changing a table and trace incidents after a failure.

Semantic layer or knowledge graph
A semantic layer or knowledge graph captures relationships and meaning: this CRM account is the same customer as that billing entity, this invoice belongs to that order, this KPI uses these measures.

Data virtualization
Data virtualization lets users query several sources through one logical layer while data stays physically distributed. It is useful for discovery and some operational or federated analytics, but it does not replace every warehouse workload.

Governance and policy
A fabric needs rules for access, sensitivity, retention, quality, and ownership. The point is to find data faster and use it in the right way.

Automation
Automation uses metadata to reduce manual work: suggest joins, flag stale assets, recommend owners, apply classifications, route access requests, or warn that a schema change will break a dashboard.

Data fabric versus data mesh

Data fabric and data mesh are related but different.

A data fabric is technology-led. It focuses on metadata, integration, automation, discovery, and governance across distributed systems.

A data mesh is organisation-led. It focuses on domain ownership: sales, finance, operations, or product teams own their data as products and publish them for others to use.

The two can work together. A mesh says who owns and serves the data. A fabric helps make those data products discoverable, connected, governed, and reusable across the organisation.

A practical example

Imagine a wholesaler with customer data in a CRM, invoices in accounting, inventory in an ERP, web orders in an ecommerce system, and campaign data in a marketing platform.

Without a fabric, answering a customer 360 question often means exports, spreadsheets, manual matching, and guesswork. The same customer appears under several names and IDs.

With a data-fabric approach, the catalog knows the relevant assets, the metadata layer knows owners and definitions, the semantic layer links the customer entities, virtualization or pipelines expose the joined view, and governance controls who may see which fields. The underlying data may still live in several systems, but the organisation can reason about it as one connected estate.

What to watch out for

It is not one product in a box
Vendors may sell a catalog, integration platform, virtualization layer, or governance suite as a data fabric. A real fabric is an architecture assembled from capabilities, processes, and ownership.

Bad metadata gives bad results
If sources are undocumented, owners are missing, and definitions conflict, a fabric can connect the wrong things faster. Start with trustworthy metadata for the most important domains.

Virtualization has limits
Querying data where it lives is powerful, but heavy analytics over large volumes may still belong in a warehouse or lakehouse. A fabric complements those platforms; it does not magically remove performance physics.

Automation needs oversight
Automatic classification, join suggestions, and policy routing are useful, but sensitive data and access rights still need human accountability.

Architecture does not fix ownership
If nobody owns customer data, a data fabric will not invent that responsibility. Governance and stewardship must be part of the design.