Cloud data management is the discipline of keeping data accessible, queryable, and governed across a stack that usually includes a lake, a warehouse, a pipeline layer, and a catalog. The shape of the stack has converged in the last five years; the disagreement now is about which layer is the source of truth and which is the consumer.
Table of contents
- The three layers
- The lake-warehouse-lakehouse pattern
- The data catalog
- The pipeline layer
- The governance layer
- How to choose
- FAQ
The three layers
A cloud data management stack in 2026 has three layers, each with a different job:
- Storage layer. Where the data lives at rest. Object storage (S3, GCS, Azure Blob) for the lake, a managed warehouse (Snowflake, BigQuery, Redshift) for the structured data.
- Compute layer. Where the queries run. Spark, Presto, Trino, the warehouse’s native compute, the lakehouse’s native compute.
- Catalog layer. Where the schema, the lineage, and the access control live. Glue Catalog, Hive Metastore, Unity Catalog, Polaris.
The lake-warehouse-lakehouse pattern
The pattern that has emerged in 2026:
- Data lake. Object storage, raw data in its native format, queried by Spark or Trino. The cheapest place to store data at rest; the most expensive place to query.
- Data warehouse. Managed service (Snowflake, BigQuery, Redshift), structured data, queried by SQL. The most expensive place to store data; the cheapest place to query.
- Lakehouse. Object storage with a transactional layer (Delta Lake, Apache Iceberg, Apache Hudi) that gives the lake the query performance and consistency of the warehouse. The middle ground that most teams have converged on.
The data catalog
The catalog is the source of truth for what data exists, who owns it, who can read it, and where it came from. The catalog is the layer most teams underinvest in. The team that has a good catalog answers “what tables exist?” in 10 seconds. The team that does not have a good catalog answers it in a week.
The catalog the team should pick: Unity Catalog (Databricks), Glue Catalog (AWS), Polaris (Apache). The right answer depends on the warehouse and the lake the team is using.
The pipeline layer
The pipeline layer is what moves data from the source to the lake, from the lake to the warehouse, and from the warehouse to the consumer. The pattern in 2026 is event-driven pipelines (Kafka, Pulsar) for the real-time stream, batch pipelines (Airflow, Dagster, Prefect) for the batch, and reverse ETL (Hightouch, Census) for the consumer-facing data.
The right answer depends on the latency the consumer needs. The team that needs the data in 5 seconds uses a streaming pipeline. The team that needs the data in 5 hours uses a batch pipeline.
The governance layer
The governance layer is what enforces the access control, the retention, the PII handling, and the audit. The governance layer is also the layer most teams underinvest in. The team that has a good governance layer answers “who has access to this table?” in 10 seconds. The team that does not have a good governance layer answers it in a week.
The governance layer is also the layer most likely to be a regulatory requirement. The team that is in healthcare, finance, or any regulated industry has to have a good governance layer; the team that is not still benefits from one.
How to choose
The choice rule: pick the stack the team can actually operate. The team that picks the most modern lakehouse architecture and cannot operate it is in worse shape than the team that picks a simpler stack and runs it well.
The right starting point for most teams: an object-storage lake, a managed warehouse, a batch pipeline, and a catalog. The team can add streaming, lakehouse, and reverse ETL as the workload grows.
FAQ
What is cloud data management?
The discipline of keeping data accessible, queryable, and governed across a stack that usually includes a lake, a warehouse, a pipeline layer, and a catalog.
What is the difference between a data lake and a data warehouse?
A data lake stores raw data in its native format on object storage. A data warehouse stores structured data in a managed service. Lakes are cheaper to store; warehouses are cheaper to query. The lakehouse pattern combines the two.
What is a lakehouse?
A lakehouse is object storage with a transactional layer (Delta Lake, Iceberg, Hudi) that gives the lake the query performance and consistency of a warehouse. The middle ground most teams have converged on.
How do I get started with cloud data management?
Start with object storage as a lake, a managed warehouse, a batch pipeline, and a catalog. Add streaming, lakehouse, and reverse ETL as the workload grows.
If you are sizing the data tier for a new project, the RunxBuild hosting calculator is the place to model the line items. The storage, the queries, the pipeline, the catalog - each one is a separate number, and the team’s mental model for the platform is the sum of those numbers. The RunxBuild dashboard is where the team sees the actual data flow in one place.