The Data Cloud and Managed Platforms

• Describe the services a 'data cloud' bundles (storage/catalog/compute/governance)
• Compare compute-and-storage separation vs classic warehouses
• Evaluate vendor lock-in, egress and interoperability trade-offs

• It removes the 'build five tools and glue them' tax: catalog, warehouse and governance come integrated.
• Elastic, separated compute means you pay for what you run, not a fixed cluster that idles.
• Data sharing across org boundaries becomes an exchange (no one emails CSVs).
• Serverless consumption lets small teams get enterprise-grade quality without a platform team.

You don't wire a house to your own power plant — you buy metered electricity from a grid you share. Data clouds meter queries and storage the same way and sell you a governance switchboard on top.

• Storage in a vendor-native format (or open formats like Parquet via Iceberg/Delta on object storage).
• Compute clusters spin up per query and scale horizontally; warehouses = virtual compute endpoints.
• Catalog centralizes schemas, tables, volumes, functions and lineage across the org.
• Governance layer unifies row/column access, masking, tagging and compliance policies.
• Data sharing/cleanrooms allow governed access to other people's data (and your own) without copying.
• Integrations/connectors ingest from SaaS and databases; open catalogs expose it to Spark/Athena/Trino.

• A ten-person fintech stands up Snowflake, loads SaaS data via a connector, and serves CFO reports on day one — no platform team, no self-hosted orchestration.
• A retailer shares a governed product catalog with suppliers via a cleanroom, so exact revenue numbers cross orgs without copying files.
• An analytics team runs ad-hoc BigQuery queries on inbound events while a marketing warehouse auto-scales to zero at night.

• Ground decisions in total cost of ownership: storage, compute, egress and idle-cluster behavior.
• Prefer open table formats (Iceberg/Delta) so a future exit doesn't ransom your data.
• Standardize on one catalog for lineage and access even if compute is multi-engine.
• Set up per-role cost allocation so each team sees its own consumption bill.

• Assuming 'serverless' means free — runaway queries and overnight cubes burn real money.
• Copying everything into one vendor and then discovering egress is expensive to leave.
• Letting each team provision its own isolated workspace, fragmenting the catalog.
• Skipping cost alerting until the monthly invoice shocks the finance team.

• Slow query → check if compute autoscale met the query or concurrency throttled it to a queue.
• Cost spike → run the vendor usage explorer by warehouse/time-of-day; overnight backfills are the usual suspect.
• Data invisible → the object is in storage but the catalog hasn't refreshed (external table metadata drift).
• Access denied → check the platform's tag/masking policy layer, which can hide columns even with table grants.