The Same Platform, Twice: AWS to Azure and Fabric

Level 7: Warehouses, Lakehouse & Dimensional Modelingeasy24 mincross-cloud service mappingADLS Gen2OneLakeFabric WarehouseAzure Data Factorymarket readingpercentage aggregation

Translate every layer of the data platform between AWS, Azure, and Fabric names, and read where Azure demand actually concentrates.

Five layers, two vendors, one set of ideas

Everything this course has taught you is vendor-neutral underneath. Object storage is object storage, a catalog is a catalog, a warehouse is a warehouse. What changes between clouds is the label on the box.

Table

The same layers under three vendor vocabularies. Learn the concept, then swap the noun.
concept	AWS	Azure	Microsoft Fabric
Object storage	Amazon S3	ADLS Gen2	OneLake
Catalog and governance	Glue Data Catalog	Microsoft Purview	OneLake catalog
Serverless SQL over files	Amazon Athena	Synapse serverless SQL	SQL analytics endpoint
MPP warehouse	Amazon Redshift	Synapse dedicated SQL pool	Fabric Warehouse
Pipeline orchestration	Step Functions and MWAA	Azure Data Factory	Fabric Data pipeline
Event streaming	Kinesis Data Streams	Azure Event Hubs	Eventstream

The same layers under three vendor vocabularies. Learn the concept, then swap the noun.

That table is why every simulated seed in this course is both clouds at once. An object inventory is an S3 Inventory report and an ADLS Gen2 listing. A pipeline run log is a Glue job history and an ADF activity run history. A shard-metrics table is Kinesis and Event Hubs. You have been practicing Azure this whole time without the Azure nouns.

Where the Azure demand actually is

Say the numbers honestly. In a scrape of 943 US Glassdoor data engineering postings published in April 2026, AWS appeared in 40.3%, Azure in 34.3%, and GCP in 12.3%. That is a single-source sample and the same report swung hard year over year, so the safe claim is that AWS and Azure are roughly comparable in US postings with AWS ahead.

The split that matters more than the national average is the segment split. AWS skews startups and product tech companies. Azure skews large enterprises: healthcare systems, banks and insurers, manufacturing, and the public sector. If you are interviewing in the Midwest with auto suppliers, health systems, and regional banks, you will meet Azure vocabulary more often than the national number suggests. That is exactly what the Apply exercise measures.

The Microsoft data stack in 2026

Four sentences get you through the small talk:

ADF (Azure Data Factory) is the workhorse orchestrator and copy engine, the ADLS Gen2 lake is where files land, and Azure Databricks does the heavy Spark work. This trio is the current junior-loop reality.
ADLS Gen2 is object storage with a hierarchical namespace bolted on, so directory renames are real operations instead of key rewrites.
Microsoft Fabric is the bundle Microsoft is steering everyone toward: OneLake as the single storage account, Delta-backed tables, and Power BI in the same product. Microsoft claimed roughly 70% of the Fortune 500 as paying Fabric customers by mid-2026, but Fabric-titled data engineering roles usually ask for 3 to 5 years, so treat Fabric as the forward anchor rather than the thing you claim to have run.
DP-700 (Implementing Data Engineering Solutions Using Microsoft Fabric) is the current Microsoft data engineering certification. DP-203 retired on March 31, 2025 with no transition path, so naming it dates you instantly.

This lesson is the vocabulary layer only. Level 8 goes inside ADF itself: pipelines, activities, datasets, linked services, integration runtimes, triggers, and the metadata-driven pattern that copies 200 tables without building 200 pipelines.

Common mistake: treating Azure as a separate subject you have not studied. It is the same five layers you already know, and answering "I have not used Azure" instead of "that is ADLS Gen2, which is object storage with a hierarchical namespace" throws away knowledge you already paid for.

Interview nuance: at an Azure shop, answer the concept question in their vocabulary. If they ask where raw files land, say ADLS Gen2 or OneLake, not S3, then explain the concept identically. Naming DP-700 rather than DP-203 signals you are current, and mentioning that Synapse is folding into Fabric signals you follow the platform.

On a real platform this differs. Here you query one small cloud_service_map table that a platform team would keep in a wiki. In a real migration that mapping lives in an architecture decision record and gets messier at the edges, because the services are not one-to-one: ADF covers what Step Functions, Glue jobs, and DMS split between them on AWS. The layer thinking still holds, which is the point.

Sample data for this example

CREATE TABLE cloud_service_map (
  concept       TEXT,
  layer         TEXT,   -- storage | catalog | compute | orchestration | streaming | serving
  aws_service   TEXT,
  azure_service TEXT,
  fabric_item   TEXT    -- '' when Fabric ships no equivalent item
);
INSERT INTO cloud_service_map (concept, layer, aws_service, azure_service, fabric_item) VALUES
  ('Object storage',            'storage',       'Amazon S3',            'ADLS Gen2',                  'OneLake'),
  ('Catalog and governance',    'catalog',       'Glue Data Catalog',    'Microsoft Purview',          'OneLake catalog'),
  ('Serverless SQL over files', 'compute',       'Amazon Athena',        'Synapse serverless SQL',     'SQL analytics endpoint'),
  ('MPP warehouse',             'compute',       'Amazon Redshift',      'Synapse dedicated SQL pool', 'Fabric Warehouse'),
  ('Spark batch processing',    'compute',       'AWS Glue ETL',         'Azure Databricks',           'Fabric notebook'),
  ('Hadoop cluster (managed)',  'compute',       'Amazon EMR',           'Azure HDInsight',            ''),
  ('Pipeline orchestration',    'orchestration', 'Step Functions',       'Azure Data Factory',         'Fabric Data pipeline'),
  ('Event streaming',           'streaming',     'Kinesis Data Streams', 'Azure Event Hubs',           'Eventstream'),
  ('BI serving',                'serving',       'Amazon QuickSight',    'Power BI',                   'Power BI in Fabric');

Worked example (SQL)

-- The platform, layer by layer, in three vocabularies at once.
SELECT layer, concept, aws_service, azure_service, fabric_item
FROM cloud_service_map
ORDER BY layer, concept;

Apply

Your turn

The task this lesson builds to.

Write a query that returns each vertical's Azure share of postings as (vertical, azure_pct), highest share first, over de_job_postings(posting_id, employer, vertical, primary_cloud).

azure_pct is the percentage of that vertical's postings whose primary_cloud is 'azure', rounded to 1 decimal. Keep every vertical, including the ones with no Azure postings.

3 hints and 1 automated check are waiting in the workspace.

Practice

Make it stick

A second problem on the same idea, plus 2 bonus drills.

Write a query that returns every compute-layer concept in all three vocabularies as (concept, aws_service, azure_service, fabric_item), ordered by concept, over cloud_service_map(concept, layer, aws_service, azure_service, fabric_item).

Keep only rows where layer is 'compute'. Some concepts have no Fabric equivalent and store an empty string, so flag those by returning the text no Fabric equivalent in fabric_item instead of the blank.

3 hints and 1 automated check are waiting in the workspace.