Amazon Web Services is the largest cloud provider and, for data engineering, a complete stack on its own. Most AWS data architectures center on S3 as the lake and compose the rest of the services around it. Knowing which service plays which role is most of what you need to navigate an AWS data platform.

The core data services

• S3 - object storage; the foundation. Nearly everything reads from and writes to S3, and its storage classes + lifecycle rules control cost.
• Redshift - the managed columnar data warehouse for BI and large analytic queries; Redshift Spectrum queries S3 directly.
• Athena - serverless SQL (Trino/Presto under the hood) over S3 files; pay per query scanned, no cluster to run.
• EMR - managed Spark/Hadoop/Hive clusters for heavy batch processing.
• Glue - serverless Spark ETL plus the Glue Data Catalog, the metastore many other services share.
• Kinesis / MSK - streaming ingestion (Kinesis Data Streams) and managed Kafka (MSK).
• Lake Formation - fine-grained permissions and governance over the S3 lake.

A typical AWS data architecture

Events land via Kinesis or MSK; raw data is written to S3 (bronze); Glue or EMR transforms it into curated Parquet/Iceberg (silver/gold); the Glue Data Catalog records the schemas; Athena serves ad-hoc SQL and Redshift serves BI dashboards; orchestration runs on Managed Workflows for Apache Airflow (MWAA) or Step Functions. Identity and access run through IAM, which is the constant across every service.

-- Athena: query Parquet in S3 with no cluster, billed per data scanned
SELECT region, sum(amount)
FROM "lake"."events"
WHERE event_date = DATE '2026-06-01'
GROUP BY region;

Cost levers that matter

AWS data bills are driven by a few things: S3 storage class and request volume, data egress (moving data out of a region or to the internet is expensive), Athena's per-TB-scanned pricing (columnar formats + partitioning cut this sharply), and idle EMR/Redshift clusters. Keeping compute and storage in the same region, partitioning lake data, and using serverless options for spiky workloads are the standard savings.

Where it fits

AWS is the default when the organization is already on it or wants the broadest service catalog and ecosystem. Its data services are mature but numerous - the complexity is choosing and wiring the right subset. Azure and Google Cloud offer equivalent stacks; the decision usually follows existing data gravity, identity, and team familiarity rather than raw capability.