GCP

AWS has more job volume; GCP punches above its weight in data roles specifically because BigQuery is so widely adopted, including by companies that otherwise run AWS. Skills transfer heavily — object storage, IAM, and orchestration concepts are the same shape everywhere.

Dataflow for fully serverless pipelines and unified batch/streaming under the Beam model, especially when you do not want cluster operations. Dataproc when your team already writes Spark, needs its ecosystem, or wants to reuse code across clouds.

Often not — Pub/Sub BigQuery subscriptions land streams directly into tables, and scheduled SQL or Dataform handles transformation. Dataflow earns its complexity for heavy per-event enrichment, joins across streams, or exactly-once stateful processing.

Lake and warehouse governance: cataloging assets across projects, data quality scans, lineage, and policy at the domain level. It matters once multiple teams share the platform — before that, plain IAM plus good dataset conventions is usually enough.

BigQuery on-demand scans (fixable with partitioning and byte limits), Dataflow jobs left streaming with no traffic, and cross-region egress in multi-region architectures. Budgets plus alerts per project are ten minutes of setup that prevent most surprises.