State of Play 2026: Data Catalogs and Metadata Management

Where the market stands

• March 19, 2025. Coalesce acquired CastorDoc. CastorDoc is now Coalesce Catalog.
• May 21, 2025. Acryl Data, the company behind the open-source DataHub project, raised $35 million in Series B funding led by Bessemer Venture Partners and rebranded as DataHub.
• July 2025. ServiceNow closed its acquisition of data.world, a data catalog and data governance platform.
• July 24, 2025. Collibra acquired Deasy Labs, a startup that discovers and enriches unstructured data.
• November 18, 2025. Salesforce completed its acquisition of Informatica, bringing Informatica's catalog, integration, governance, quality, privacy, metadata management and master data management services to the Salesforce platform.
• November 24, 2025. Snowflake announced a definitive agreement to acquire the Select Star team and platform technology. Snowflake said Select Star's integrations with PostgreSQL, MySQL, Tableau, Power BI, dbt and Airflow would expand Horizon Catalog's view of a customer's data estate.
• February 19, 2026. Apache Polaris, an open-source catalog for Apache Iceberg tables that Snowflake first open-sourced in mid-2024, graduated to a top-level project of the Apache Software Foundation.

Google's Dataplex Universal Catalog is now Knowledge Catalog. Its billing items keep the Dataplex name.

Two kinds of catalog

The word covers two different products.

• Table catalogs. These tell query engines where tables are and who may read them. Several now implement the Apache Iceberg REST Catalog specification, so any compliant engine can use them. Databricks open-sourced Unity Catalog under the Apache 2.0 license on June 12, 2024, with support for the Iceberg REST Catalog and Hive Metastore interfaces. Apache Polaris is a vendor-neutral implementation of the Iceberg REST specification, and Snowflake builds Horizon Catalog on it. The AWS Glue Data Catalog exposes an Iceberg REST endpoint, with access governed by IAM or Lake Formation.
• Discovery catalogs. These harvest metadata from many systems for people to search, describe and certify. Alation, Atlan, Collibra, Coalesce Catalog, Informatica Cloud Data Governance and Catalog and ServiceNow's Data Catalog, built on data.world, are sold this way. DataHub and OpenMetadata are open-source discovery catalogs, both under the Apache 2.0 license. DataHub started at LinkedIn in 2019.

The data platforms now reach into discovery. Snowflake says Horizon Catalog governs data and AI assets across Snowflake and external catalogs. Databricks says Unity Catalog manages data and AI assets, including models, agents and MCPs, across any cloud or platform.

Decide which table catalog your engines will share before you choose a discovery catalog. The discovery catalog then needs a connector to it.

How lineage gets into the catalog

Lineage can arrive in three ways: parsed from queries the platform runs, collected from pipelines through an open standard, or registered by hand.

• From the platform. Unity Catalog captures lineage automatically, down to the column, for queries run on Databricks. Its lineage system tables keep a rolling one-year window. Lineage is not preserved when a catalog, schema, table, view or column is renamed. Snowflake says Horizon Catalog tracks column-level lineage across Snowflake and external databases.
• From pipelines. OpenLineage is an open framework for collecting lineage, and a graduate project of the LF AI & Data Foundation. It defines datasets, jobs and runs, with extensible facets. Its documentation lists integrations for Apache Airflow, Apache Flink, Apache Hive, Apache Spark, dbt, Great Expectations and Trino. Marquez, another LF AI & Data project, collects and visualizes that metadata.
• By registration. Unity Catalog's external lineage lets you add systems outside Databricks, such as Salesforce or Tableau, to the same graph.

Ask each vendor which of your pipelines it parses itself and which must emit OpenLineage events. Then test column-level lineage on your own transformations, including one that crosses from the warehouse into a BI tool.

Metadata for AI agents

Catalog vendors now ship Model Context Protocol (MCP) servers, so AI assistants and agents can search the catalog and read lineage.

• Atlan. A hosted MCP server, enabled for all Atlan tenants. Users sign in with OAuth and act with their own permissions, or automation uses a shared API key.
• Collibra. The MCP server reads assets, glossary terms, lineage, data contracts and classifications, and can edit assets, contracts and classifications. Every action uses Collibra's existing permission model. Setup requires an OAuth 2.0 application.
• DataHub. A managed MCP server on DataHub Cloud version 0.3.12 and later. Users of the open-source DataHub Core host it themselves. It supports search, table- and column-level lineage and query analysis.

Check whether the agent sees only what the signed-in user may see, and whether it can change metadata as well as read it.

What each costs where prices are public

• Microsoft Purview Unified Catalog. Prices in USD, the same in every Azure region Microsoft lists. Billed per governed asset, at $0.0165 per asset per day, prorated by the days governed. An asset counts only when it is linked to a governance concept, such as a data product, critical data element or glossary term. Assets collected in its Data Map but not linked to a governance concept are not counted. Microsoft's example: 200 governed assets for 30 days cost $100 without discounts.
• Databricks Unity Catalog. Included with the Databricks Premium and Enterprise tiers. No separate license is required.
• Google Knowledge Catalog. Billed in data compute unit (DCU) hours, in USD at the pay-as-you-go rate. In the us-central1 (Iowa) region, standard processing for metadata discovery costs $0.06 per DCU-hour, with 100 DCU-hours free each month. Premium processing for lineage, data quality and profiling costs $0.089 per DCU-hour. Metadata storage costs $0.002739726 per GiB-hour in any region. Metadata ingested automatically from Google Cloud services such as BigQuery is stored free, and catalog API calls are free.
• AWS Glue Data Catalog. In us-east-1, the first million objects stored and the first million requests each month are free. Beyond that, storage costs $1 per 100,000 objects per month and requests cost $1 per million. Crawlers cost $0.44 per DPU-hour, billed by the second.
• Amazon SageMaker Catalog. In USD, one global rate in AWS's price list. $10 per 100,000 requests after 4,000 free each month per account, $0.40 per GB of metadata storage after 20 MB free, and $1.776 per compute unit after 0.2 free. Core APIs such as Search are free.
• Collibra. On AWS Marketplace, a 12-month subscription to the Collibra cloud platform is listed at $170,000 (list price as published), with private offers available. The 24- and 36-month terms show no price.
• Alation. On AWS Marketplace, a 12-month subscription to the Alation Data Catalog is listed as starting at $60,000 (list price as published), with private offers available. Alation's own pricing page lists no prices.
• Atlan. The pricing page lists no prices and routes buyers to sales.
• Collate, built by the creators of OpenMetadata, offers a free tier for 5 users and 500 data assets. Its Premium and Enterprise plans list no prices.

What an open-source catalog requires

DataHub and OpenMetadata carry no license fee. Each needs its own database and search engine.

• DataHub. The quickstart starts MySQL, OpenSearch and Kafka alongside DataHub's own services. DataHub says the quickstart is not for production and recommends Kubernetes. DataHub Cloud is the managed version.
• OpenMetadata. Production needs at least PostgreSQL 15 or MySQL 8.0.42, OpenSearch 3.4.0 or Elasticsearch 9.3.0, and Java 21. The documentation sizes the server at a minimum of 4 vCPUs, 16 GiB of memory and 100 GiB of storage, and sizes the database and search engine separately.

Before choosing open source, name the engineers who will run upgrades, connectors and the search cluster, and price their time against a subscription.

Questions to settle first

Settle these before shortlisting.

• Which table catalog will your query engines share, and does the discovery catalog connect to it?
• Which sources must be covered on day one, and which have no native connector?
• Which pipelines need column-level lineage, and does each emit it or need parsing?
• Will AI agents read the catalog, and with whose permissions?
• Who will run an open-source catalog, or who will curate a commercial one, after launch?
• If the vendor changes hands, what do your contract terms allow?

Ask to contribute

Vendors in this market can ask to answer the six questions for State of Play 2026: Data Catalogs and Metadata Management. Use an email address on your company’s own domain: answers come from the company itself, not from an agency. We email a link to confirm the request.