Building a Data Mesh on AWS, Step by Step

Three AWS paths, one multi-account pattern, and where to start your first data product.

Blog
8-MINUTE READ
[alt_text prompt_guidance="USE SEO keywords or synonyms based on post title"]

A data mesh on AWS is a multi-account architecture in which domain teams publish governed data products to a shared catalog, and other teams subscribe to them through controlled, cross-account access. AWS supports three ways to build one: Amazon DataZone (the foundation of Amazon SageMaker Catalog), the open-source data.all framework, or a custom build on AWS Lake Formation and the AWS Glue Data Catalog. The services are the easy part. The harder decisions are who owns each dataset and where governance lives. This guide walks architects through the account layout, the service choices, and a pilot-first rollout.

If you need the concept before the implementation, start with our plain-English explainer on what a data mesh is.

Key takeaways

  • A data mesh on AWS separates producer, governance, and consumer accounts, so each domain owns its data while one catalog enforces shared rules.
  • AWS documents three implementation paths: Amazon DataZone (managed), data.all (open source), and a custom Lake Formation build. The right one depends on your team's capacity to operate a platform.
  • Start with one domain and one data product. A pilot proves the ownership model before you scale it.

What Does a Multi-Account Data Mesh Architecture on AWS Look Like?

A multi-account data mesh architecture on AWS splits responsibilities across accounts: producer accounts hold domain data, a governance account holds the catalog and policies, and consumer accounts subscribe to approved data products. AWS describes this pattern in its February 2026 Big Data Blog post on a data mesh with Amazon SageMaker Catalog. Producers publish existing assets from Amazon S3 and the AWS Glue Data Catalog. Consumers discover and subscribe to them.

The Four Roles AWS Defines

The AWS Well-Architected Data Analytics Lens defines four roles in a data mesh. Naming them early prevents most ownership disputes later.

  • Data owner: a domain expert responsible for a data domain in a decentralized data lake account.
  • Data steward: manages federated governance and the auditability of metadata.
  • Data engineer: acts as the data producer, building data products that are discoverable, secure, and reusable.
  • Data consumer: discovers, requests access to, and uses data products.

Which AWS Services Make Up a Data Mesh?

A data mesh on AWS combines storage, catalog, access control, and discovery services. AWS Prescriptive Guidance lists Amazon S3, the AWS Glue Data Catalog, AWS Identity and Access Management (IAM), AWS Key Management Service (KMS), and AWS Lake Formation as the common foundation across all three implementation options.

LayerAWS serviceRole in the mesh
StorageAmazon S3Holds each domain's data in its own producer account
MetadataAWS Glue Data CatalogStores table definitions and supports data product discovery
Fine-grained accessAWS Lake FormationEnforces who can see which tables, columns, and rows
Identity and encryptionAWS IAM and AWS KMSControls access by role and encrypts shared data
Discovery and sharingAmazon DataZone / Amazon SageMaker CatalogProvides the business catalog, subscription requests, and approval workflow

What Are the Three Ways to Implement a Data Mesh on AWS?

AWS documents three implementation paths: a fully managed option with Amazon DataZone, an open-source option with data.all, and a custom build on AWS Lake Formation. Each one covers the same four concerns: domain ownership, data products, self-service access, and federated governance. The table compares them. The first four columns follow AWS Prescriptive Guidance, and the last column is D3Clarity's own guidance.

OptionTypeDiscovery and self-serviceAccess approvalBest fit
Amazon DataZone / Amazon SageMaker CatalogManaged AWS serviceDataZone portal to search, filter, and discover data productsBuilt-in workflow where data owners review and approve requestsTeams that want the fastest path with the least platform operations
data.allOpen-source frameworkWeb front end with sign-in through SSO or Amazon CognitoBuilt-in workflow where data owners review and approve requestsTeams that want to inspect and extend the platform and have engineers to run it
Custom build on AWS Lake FormationBuild it yourselfCustom platform components that your team buildsLake Formation manages access to the AWS Glue Data Catalog in a governance accountTeams with workflow or compliance requirements a packaged option cannot meet

Option 1: Amazon DataZone Data Mesh with Amazon SageMaker Catalog

Amazon DataZone is the managed option. It provides a portal where users search for data products, a business catalog where producers publish them, and a built-in workflow where data owners approve access requests. As of 2026, AWS documents that Amazon SageMaker Catalog is built on Amazon DataZone and is accessed through Amazon SageMaker Unified Studio. If you are starting fresh, this is the shortest path to a working mesh.

Option 2: data.all

data.all is an open-source framework that adds a web front end, a catalog, and an access-request workflow on top of your AWS accounts. It uses Amazon Cognito for sign-in and AWS CloudFormation, AWS Lambda, and Amazon API Gateway for its backend. Choose it if you want an open-source platform that you can inspect and extend, and you have engineers to run it.

Option 3: A Custom Build on AWS Lake Formation

A custom build uses the AWS Glue Data Catalog as a central catalog in a governance account and AWS Lake Formation to manage access to it. This option relies on AWS Lake Formation data governance: Lake Formation centrally manages access to catalog entries, while producer and consumer accounts stay separate. Your team builds the self-service layer itself. This gives the most control and requires the most engineering. It fits organizations with unusual workflow or compliance requirements that a packaged option cannot meet.

Can You Build a Data Mesh Without Rewriting Your Applications?

Yes. The AWS February 2026 pattern lets existing applications keep reading data after it moves under catalog governance. In that design, a consumer application assumes the IAM role of its consumer project and reads the subscribed data. The application code does not change. Lake Formation enforces the permissions behind the scenes. This matters for architects with legacy workloads that cannot be refactored on the mesh timeline.

How Do You Implement a Data Mesh on AWS, Step by Step?

Start with a fit check, then pilot one domain end to end before you scale. Follow these steps:

  1. Confirm the fit. Check your organization against AWS's suitability criteria in the section below.
  2. Pick one pilot domain. Choose a team with a clear owner and a real consumer waiting for its data.
  3. Stand up the governance account. Set up the catalog, Lake Formation permissions, and encryption standards before any data is published.
  4. Publish one data product. Give it a named owner, documentation, a refresh schedule, and quality checks.
  5. Onboard one consumer. Run the full subscribe-and-approve workflow and note where it slows down.
  6. Expand domain by domain. Reuse the templates and guardrails from the pilot.

D3Clarity helps teams plan this sequence as part of data management and governance work.

When Is a Data Mesh on AWS the Wrong Choice?

A data mesh is the wrong choice when your organization has one small, well-managed data team and no real sharing bottleneck. The AWS Well-Architected Data Analytics Lens says a data mesh suits organizations that have:

  • A well-established data strategy.
  • Decoupled, autonomous business units.
  • A need to share data across units or with external partners.
  • A need for consistent governance across multiple teams.
  • Quick delivery cycles and agile practices.

If most of these do not apply yet, a well-governed central data lake is usually the better first step.

Frequently Asked Questions About Data Mesh on AWS

What is a data mesh on AWS?

A data mesh on AWS is a multi-account architecture where domain teams own and publish data products, and a shared governance layer controls access. Producer accounts hold the data, a governance account holds the catalog, and consumer accounts subscribe to approved data products. It can be built with Amazon DataZone, data.all, or AWS Lake Formation.

How do you implement a data mesh on AWS step by step?

Implement a data mesh on AWS in six steps: confirm the fit, pick one pilot domain, set up a governance account, publish one data product, onboard one consumer, and expand domain by domain. Start small so you can test ownership, access approvals, and documentation before you scale the model across the organization.

Which AWS services do you need for a data mesh?

The common foundation is Amazon S3, the AWS Glue Data Catalog, AWS IAM, AWS KMS, and AWS Lake Formation. On top of that, most teams add Amazon DataZone or Amazon SageMaker Catalog for discovery and access approval. The open-source data.all framework is an alternative to DataZone for the same layer.

Is Amazon SageMaker Catalog the same as Amazon DataZone?

Not exactly. As of 2026, AWS documents that Amazon SageMaker Catalog is built on Amazon DataZone. DataZone provides the underlying governance platform, and SageMaker Catalog is how users reach it inside Amazon SageMaker Unified Studio. For data mesh purposes, they serve the same function: a governed catalog with data products and subscription workflows.

Do you have to move your data to build a data mesh on AWS?

No. A data mesh is an ownership and governance model, not a storage migration. AWS's February 2026 pattern publishes existing Amazon S3 data and AWS Glue tables to a catalog. Consumer applications keep working by assuming a project IAM role, so application code does not need to change.

The Bottom Line

Implementing a data mesh on AWS comes down to three decisions: how you split accounts, which of the three implementation paths fits your team, and which single domain you pilot first. Get those right, and the services fall into place. If you want a second set of eyes on your design, talk to D3Clarity about building a data mesh on AWS.

Earin Persson
VP of Channel Sales & Service Delivery Operations

Related Content

PODCAST
DATA
Why IoT Data Needs Different Architecture
Explore how IoT data changes architecture, from streaming sensor events to rollups that support smarter business decisions and insights.
PODCAST
Strategic Mindset vs Technical Depth: Finding Balance

No meta description found

Data & AI
Secure Cloud