Cloud warehouses
🎯You'll be able to
Compare BigQuery and Snowflake, explain how they separate storage from compute, and estimate what a query or a month of use might cost.
🏠The everyday picture
Renting a hall by the hour is different from paying per guest. Cloud warehouses offer both kinds of deal, and the right one depends on how you use the hall.
📘What it really is
-
Storage and compute are separate: data sits in cheap storage; computers are added to run queries. You can store a lot and pay for compute only when you query.
-
BigQuery (Google Cloud): serverless. On-demand pricing charges by data scanned per query; you can also buy reserved capacity. On 1 November 2026 Google's pricing page listed $6.25 per TiB scanned on demand in the US multi-region, with the first 1 TiB a month free, and active storage at about $0.023 per GiB a month (check the current page; prices vary by region).
-
Snowflake: you create virtual warehouses (compute clusters) in sizes, billed in credits per second while they run, with auto-suspend when idle. Storage is billed separately.
-
Others: Amazon Redshift, Azure Synapse and Microsoft Fabric, Databricks SQL, ClickHouse Cloud. DuckDB and MotherDuck suit small data.
-
Partitioning and clustering (DT4 Lesson 5) cut what each query reads: BigQuery tables partitioned by date and clustered by store; Snowflake prunes micro-partitions automatically and can cluster large tables.
-
Controls: set budgets and alerts, limit the bytes a query may scan, auto-suspend compute, and look at the most expensive queries every month.
📝A worked example
Input (illustrative): SabziSathi at 50 times today's orders keeps about 1 GB of fact data a year. A dashboard query reads two columns, about 15% of the table, and runs 2,000 times a month.
Expected output: about 0.15 GB × 2,000 = 300 GB scanned a month, inside BigQuery's 1 TiB free allowance at the 1 November 2026 price list; storage of a few GB costs well under a dollar a month. These are estimates from the published prices, not a bill.
🌍Where it's used
- Startups: BigQuery's free allowance or a small Snowflake warehouse for the first analytics.
- Large companies: reserved capacity and many virtual warehouses for different teams.
- Data teams: monthly cost reviews of the top 20 queries.
📝Case scenario
An analyst at an edtech company ran select * on a 4 TB event table to look at ten rows, several times a day for a week. On-demand
pricing charged for the full scan each time. A limit doesn't reduce what BigQuery scans; selecting three columns from one day's
partition would have read a few gigabytes.
🖱️Try it
Estimate the monthly on-demand cost of a dashboard that scans 50 GB per refresh and refreshes every 15 minutes, at the 1 November 2026 price. Mark it illustrative. Success: about 4.8 TB a month (50 GB × 96 × 30 ÷ 1,000), which is about 4.4 TiB, so about 3.4 TiB beyond the free allowance, roughly $21.
⚠️Common mistakes
select *on large tables. Sign: big bills for small answers. Fix: select the columns you need, and filter on the partition column.- Compute left running. Sign: a Snowflake warehouse billed all weekend. Fix: auto-suspend after a few minutes idle.
- No budget alert. Sign: a surprise invoice. Fix: set budgets and per-query limits on day one.
🛠️For your project
The capstone's design notes include an illustrative monthly cost for SabziSathi in a cloud warehouse, using this lesson.
✅Quick check
What does BigQuery's on-demand price charge for?
Show the answer
The amount of data each query scans.
What does Snowflake's auto-suspend save?
Show the answer
Compute credits while a warehouse is idle.