Before we write a line of pipeline code, a client almost always asks the same thing: Snowflake, Redshift, Big Query or Databricks?
I’ve led these migrations off Sybase, Oracle and on-prem ETL, and my honest answer is that the platform’s monthly bill is the smallest part of the decision. The expensive, risky part is the move itself and getting the fit right.
Here’s the short version of how I choose, and what a cloud data warehouse migration really costs.
First: Are you building a warehouse, or a lake house?
Settle this before comparing price-per-credit.
- A warehouse is optimized for structured data and BI clean tables, fast SQL, dashboards.
- A lake house keeps data in open formats (Parquet, Delta, Iceberg) on cheap object storage and runs both SQL and ML over it.
The four, in one breath each
- Snowflake: Zero-ops and multi-cloud (AWS/Azure/GCP). Separated storage and compute, no indexes to tune. The safe default when nothing else decides it.
- Amazon Redshift: Unbeatable adjacency if you’re already on AWS: analytics next to your S3, IAM and VPC, with minimal data movement and egress.
- Google BigQuery: the most serverless of the four. No cluster to size; brilliant for bursty, ad-hoc analytics on GCP, with SQL-native ML built in.
- Databricks: the lakehouse: open formats, Spark, best-in-class ML, and the strongest hedge against storage lock-in. A platform, not just a warehouse.
Snowflake, Redshift and Big Query are warehouse-first.
Databricks is lake house-first.
Everyone’s converging, but the center of gravity still shapes the bill and the developer experience.
What the migration actually costs
The pricing calculators hide the real number. The platform’s compute bill is rarely the expensive part of a cloud data warehouse migration the engineering effort and risk of moving are. From experience that’s:
- Schema & code conversion: DDL and especially stored procedures and SQL-dialect differences. On one bank migration that meant re-engineering 800+ stored procedures.
- Historical backfill: Bulk-loading years of data through a staging layer, plus any egress.
- Validation & reconciliation: Proving the new numbers match the old ones, table by table. Trust is the deliverable.
- Parallel running: You pay for both systems while you validate the new one.
- Team retraining: New quirks, new tooling, new cost-control habits.
Want the full framework?
This is the condensed version. The complete guide with the full comparison table, a migration-cost breakdown, a phased-rollout diagram and a decision flowchart is here: Cloud Data Warehouse Migration: Snowflake vs Redshift vs BigQuery.
