Your data stack starts here. Every pipeline, transformation, and dashboard feeds into a warehouse. Getting this decision wrong costs you a painful migration 18 months later. Getting it right means the rest of your stack falls into place.

The three serious options for startups are Snowflake, BigQuery, and Redshift. Databricks is worth knowing about but overkill until you have a data engineering team.

The three options

Snowflake

Snowflake is the default for a reason. It works with any cloud, separates compute from storage (so you only pay for queries when they run), and has the broadest connector and tool support in the ecosystem. dbt, Fivetran, Airbyte, Looker, Tableau — everything integrates with Snowflake first.

Pick Snowflake if:

  • You are cloud-agnostic or multi-cloud
  • You want the path of least resistance for tooling
  • You have budget and want a fully managed experience

Watch out for: Credits disappear faster than expected. A badly written query against a large table can burn a week of budget in minutes. Set resource monitors on day one.

BigQuery

BigQuery is Google’s serverless warehouse. The pricing model is different — you pay per byte scanned rather than per compute-second, and storage is very cheap. If your team already uses Google Workspace and Google Cloud, BigQuery is a natural fit.

Pick BigQuery if:

  • Your stack lives in GCP
  • You want a generous free tier to start (10 GB storage, 1 TB queries/month free)
  • You value Dataform for transformations over dbt

Watch out for: The on-demand pricing model punishes full-table scans. Partition your tables and cluster aggressively from the start, or your first month’s bill will be a surprise.

Redshift

Redshift is Amazon’s warehouse. It has been the enterprise default for years and is deeply integrated with the AWS ecosystem — S3, Glue, EMR. The Serverless tier removed the old provisioned cluster overhead, making it more accessible to smaller teams.

Pick Redshift if:

  • Your infrastructure is committed AWS and your data already lives in S3
  • You have existing AWS credits or enterprise agreements
  • You need tight integration with AWS Glue or other AWS data services

Watch out for: Redshift has the steepest learning curve of the three. Performance tuning (sort keys, dist keys) is a skill in itself. If you don’t have an engineer familiar with Redshift, budget time to get up to speed.

What most startups should do

Start with Snowflake. The per-second compute model, the broad tooling support, and the managed experience mean you spend engineering time on your product, not on database administration.

Switch to BigQuery if you are deeply embedded in GCP and want to keep your infrastructure consolidated. The free tier makes it genuinely free to get started.

Consider Redshift only if AWS is your primary cloud and you have engineers already comfortable with it.

Before you move on

Before setting up ingestion (Step 2), make sure you have:

  • A warehouse account created
  • A dedicated database and schema for raw data (e.g. raw.salesforce, raw.postgres)
  • Resource monitors or budget alerts configured
  • A service account / user for your ingestion tool with write permissions to the raw schema only

The raw schema matters. You want your ingestion tool writing to a separate schema from your transformed data. This is the foundation of the ELT pattern — raw in, transformed out.