A modern data architecture is not one standard blueprint. It combines storage, governance, data movement, transformation, and workload-specific services in a way that fits an organization’s data, teams, and operating model. A lakehouse, warehouse, and domain-oriented data products can coexist; the right design depends on what the organization needs to run and who can responsibly operate it.
What a modern data architecture includes
Modern data platforms often bring together a data lake, a warehouse or other purpose-built stores, shared governance, data operations, and services for analytics, discovery, data science, or machine learning. AWS describes its reference architecture as an S3-based lake with governance and a DataOps layer that moves data to purpose-built analytics services. That is AWS’s design for its own ecosystem, not a universal standard.
As an Amazon Associate I earn from qualifying purchases.
The broader problem is that organizations may hold data in several systems, each suited to a different purpose. As those systems grow, movement and coordination between them can become complex. AWS’s Analytics Lens describes three useful movement patterns:
- Inside-out: send data from a lake to a specialized store for a particular workload.
- Outside-in: bring data from an application-oriented store into the lake for broader analysis.
- Around the perimeter: move data between specialized stores.
These patterns help expose where data originates, where it is used, and where copying or integration is required. They do not mean that a lake-centered architecture eliminates duplication or data movement.
#1 Best Overall
When a warehouse or a lakehouse is a better fit
A lakehouse combines characteristics associated with data lakes and warehouses. Google Cloud describes it as bringing together the flexibility and cost savings of a data lake with data management and performance associated with a warehouse. The term covers different implementations, however, and capabilities vary by platform.
Microsoft’s Fabric guidance illustrates why teams may choose different centers of gravity. Its examples are vendor guidance, not universal thresholds for choosing one design over another.
| Design orientation | Workloads it can suit | What to consider |
|---|---|---|
| SQL-first warehouse | Structured reporting and SQL-based transformation. | Microsoft’s example keeps data structured across bronze, silver, and gold warehouses. Consider whether SQL is the team’s preferred transformation approach and whether the data is primarily structured. |
| Lakehouse-first | Spark-based engineering, data science, machine learning, and substantial unstructured or semi-structured data. | Microsoft describes Spark-oriented processing, distributed computation, notebooks, and machine learning as relevant capabilities. These can add operational and skills requirements that should be assessed for the actual team. |
| Hybrid: lakehouse engineering with warehouse consumption | Workloads that need lakehouse capabilities in engineering layers and a warehouse-oriented business-facing endpoint. | Microsoft documents a Fabric arrangement with lakehouses for bronze and silver and a warehouse for gold. Its alternative uses a lakehouse at every layer. |
Microsoft also identifies Azure Databricks as an alternative in cases where teams want deeper open-source machine-learning integration, finer Spark execution control, or multicloud portability. Treat that as Microsoft’s product guidance; validate portability and operational fit in the target environment rather than assuming either from a product description.
What bronze, silver, and gold mean
Medallion architecture describes logical layers of data quality and processing. It is a design pattern, not a mandatory platform layout or a guarantee of data quality. Azure Databricks documentation, updated September 11, 2026, states: “Following the medallion architecture is a recommended best practice but not a requirement.”
- Bronze: source data retained in raw form. Keeping original copies can support reprocessing and audit; Microsoft recommends preserving originals as a source of truth and keeping original formats in bronze where practical.
- Silver: data that has been cleaned and made more consistent through validation, standardization, deduplication, and enrichment.
- Gold: curated, consumption-ready data shaped for uses such as reporting and analytics.
For example, raw sales, inventory, and customer records could enter bronze. Silver processing might standardize formats, convert currencies, remove test records, and match records. Gold could then contain dashboard tables, customer-lifetime-value data, or inventory forecasts. This is an illustration of the pattern, not a required three-stage pipeline for every system. Microsoft commonly uses Delta tables in silver and gold in its Fabric guidance.
Layers only describe intent. Reliable results still depend on suitable validation, transformation controls, and tests; the layer names alone do not establish that the data is correct.
Rank #3
How data mesh changes ownership
Data mesh is an organizational and architectural approach, not simply a way to distribute storage. In Google Cloud’s description, teams that know a business domain best develop and own its data products within organization-wide governance. Domains can both produce and consume those products, while central functions provide shared services such as governance, cataloging, discovery, and cross-domain oversight.
Outdated Drivers Are Slowing You Down
One free scan finds every outdated or missing driver and matches the right update for your exact hardware.Free scan · exact hardware matchWindows Errors? Fix Them Before They Spread
Repair common Windows errors and clear accumulated junk for a smoother, more stable PC - no reinstall needed.Free scan · no reinstallA data product needs a usable interface for consumers. Depending on the use case, that interface might be a table, view, file, topic, or stream. The operating model must also make ownership of the product lifecycle and quality clear, and give consumers a workable way to discover and access products.
A central platform remains important in a mesh: shared policy, services, and interoperability help domains exchange data without requiring each team to solve common problems independently. Google’s reference implementation is explicitly a learning demonstration, not a production-ready implementation.
Rank #4
Mesh and hub-and-spoke are different ownership balances
A mesh places more product responsibility with domains and relies on central platform functions for shared capabilities and governance. AWS’s hub-and-spoke pattern concentrates more enterprise assets and governance centrally while preserving some business-unit autonomy. AWS presents hub-and-spoke as an option for organizations with less mature governance and as a possible step toward a more distributed mesh; that is AWS’s model guidance, not a universal maturity rule.
Choose based on who can own data quality, product lifecycle, policy, and support in practice. Domain labels or distributed storage, by themselves, do not create a functioning mesh.
Recommended Free Tools
How to choose a design for your organization
Compare candidate designs against the same operational questions rather than looking for a universal winner. The following checks draw on the workload, movement, governance, ownership, skills, and interoperability considerations highlighted in provider guidance.
Best Value
- Map the workload shape. List structured reporting, exploratory analysis, batch and streaming ingestion, machine learning, and unstructured or semi-structured data needs. Identify which are central and which are occasional.
- Trace data movement. Record which systems own operational data, which analytics consume it, and where copying, federation, or specialized stores may be needed. Use the inside-out, outside-in, and perimeter patterns to make exchanges visible.
- Set the governance requirements. Decide how access controls, cataloging, policy enforcement, lineage, and cross-domain exchange will work. Distributed ownership still requires organization-wide governance that teams can apply.
- Choose an ownership model. Decide whether a central team will own most platform and data-product work, whether domains will own products on a shared platform, or whether a hub-and-spoke balance better fits current responsibilities.
- Check team capability and support burden. Compare the available skills and expected responsibilities for SQL, Spark, data operations, platform engineering, product ownership, and ongoing support. A technically suitable pattern can still be a poor fit if no team can operate it.
- Test interoperability in the target environment. Examine formats, interfaces, engines, and platform constraints. A vendor’s description of a design as open is not proof of practical portability.
- Start with a useful initial design and evolve it. Establish clear ownership, governance, and workload fit for the first platform scope. Add distributed product ownership, additional stores, or other patterns when real needs and organizational capability justify them.
What the available guidance can—and cannot—settle
A provider’s architecture documentation is useful for understanding that provider’s patterns and services, but it is not a neutral cross-vendor ranking. The reviewed AWS, Google Cloud, Microsoft, and Databricks guidance does not establish which architecture will deliver the best performance, total cost, or staffing burden for equivalent workloads across providers. Those outcomes depend on the organization’s data, implementation, workload, and operating model.
The dates also differ across the guidance: Google Cloud’s mesh page was last reviewed September 3, 2024; Microsoft’s Fabric medallion page reports an update of July 17, 2026; and Azure Databricks’ medallion page reports an update of September 11, 2026. AWS’s cited pages did not show a publication date in the reviewed page content. Use each provider’s documentation to understand its own recommendations, then validate the design against the organization’s requirements.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
