About 20; min
Your company’s data lives in a warehouse — but which warehouse, and at what price? Snowflake, BigQuery, and Redshift are the three dominant cloud data warehouses in 2026, and their pricing models couldn’t be more different.
Snowflake charges by compute credits. BigQuery charges by bytes scanned. Redshift charges by node hours. Comparing them requires translating all three into the same language: dollars per query at your data volume.
Pricing Model Overview
| Feature | Snowflake | BigQuery | Redshift |
|---|---|---|---|
| Pricing model | Credits (compute) + storage | Per TB scanned + storage | Per node hour + storage |
| Free tier | $400 in credits (30 days) | 1 TB/month free queries | 2 months free trial |
| Storage (per TB/mo) | $23 (on-demand) | $20 (active), $10 (long-term) | $23.40 (managed) |
| Compute entry point | $2/credit (~$3.60/hr X-Small) | $6.25/TB scanned | $0.25/hr (dc2.large) |
| Auto-scaling | Yes (multi-cluster) | Automatic (serverless) | Yes (RA3 + Serverless) |
| Serverless option | No (always clusters) | Yes (default) | Yes (Redshift Serverless) |
Snowflake Pricing Deep Dive
Snowflake’s pricing hinges on warehouse size and runtime. An X-Small warehouse (1 node) costs 1 credit per hour at the Standard tier. A Medium warehouse (4 nodes) costs 4 credits per hour. Costs scale linearly with warehouse size.
Real Compute Costs
| Warehouse Size | Credits/Hour | Standard $/hr | Enterprise $/hr |
|---|---|---|---|
| X-Small | 1 | $2.00 | $3.00 |
| Small | 2 | $4.00 | $6.00 |
| Medium | 4 | $8.00 | $12.00 |
| Large | 8 | $16.00 | $24.00 |
| X-Large | 16 | $32.00 | $48.00 |
The key to controlling Snowflake costs: auto-suspend. Warehouses can automatically shut down after 1–10 minutes of inactivity. If your team runs queries for 4 hours per day on a Medium Enterprise warehouse, your daily compute bill is $48 — not $288 for a full 24 hours.
Storage Pricing
Snowflake charges $23/TB/month on-demand or $40/TB/month for pre-purchased capacity (which sounds worse but includes failover and replication). Most teams land between $23–$30/TB/month depending on their contract.
Time travel (querying historical data states) is included but uses storage. Standard gives you 1 day of time travel data; Enterprise gives 90 days. If you have 10 TB of data with daily changes, 90-day time travel could add 2–3 TB of storage costs.
BigQuery Pricing Deep Dive
BigQuery has two pricing models: on-demand (pay per TB scanned) and Editions (pay for compute slots). The right choice depends on query predictability.
On-Demand: Pay Per Query
At $6.25 per TB scanned, a query that scans 100 GB costs $0.63. Simple. But costs are unpredictable — a poorly written JOIN on a 5 TB table costs $31.25 per execution. Run that query 10 times while debugging? $312.50.
Cost control tips: partition tables by date, use clustering, and always preview query cost before running (BigQuery shows estimated bytes scanned). Teams that follow these practices typically cut on-demand costs by 60–70%.
Editions: Predictable Capacity
Editions pricing is based on slots (units of compute). Standard Edition starts at $0.04/slot-hour with autoscaling. A baseline of 100 slots running 8 hours/day costs $32/day or ~$960/month. You can burst above your baseline during peaks and pay per-slot for the extra capacity.
Editions makes sense when your monthly on-demand spend exceeds $1,500–$2,000. Below that, on-demand is simpler and often cheaper.
Redshift Pricing Deep Dive
Redshift has two modes: Provisioned (always-on clusters) and Serverless (pay per query). Provisioned is cheaper for steady workloads; Serverless is better for unpredictable or bursty usage.
Provisioned Cluster Costs
| Node Type | $/Hour | vCPU | RAM | Monthly (3 nodes) |
|---|---|---|---|---|
| ra3.xlplus | $1.086 | 4 | 32 GB | $2,346 |
| ra3.4xlarge | $3.26 | 12 | 96 GB | $7,042 |
| ra3.16xlarge | $13.04 | 48 | 384 GB | $28,166 |
Reserved instance pricing (1-year commitment) saves 32–36%. A 3-node ra3.xlplus cluster drops from $2,346/month to ~$1,596/month with a 1-year RI.
Serverless Costs
Serverless charges $0.375 per RPU-hour with a minimum of 8 RPUs (configurable up to 512). The minimum baseline costs $3/hour when active. Serverless automatically pauses after idle periods, so you only pay when queries are running.
For a team running queries 6 hours/day at 32 RPUs average: $72/day or ~$2,160/month. Compare that to a provisioned 3-node ra3.xlplus cluster at $2,346/month — Serverless is slightly cheaper with the added flexibility of zero management.
Cost Scenarios: 10 TB Data, Moderate Query Load
| Component | Snowflake Enterprise | BigQuery On-Demand | Redshift Provisioned (3x ra3.xlplus) |
|---|---|---|---|
| Storage (10 TB) | $230/mo | $200/mo | $234/mo |
| Compute (8 hrs/day, medium load) | $2,160/mo | ~$1,500/mo (est. 240 TB scanned) | $2,346/mo |
| Total | $2,390/mo | $1,700/mo | $2,580/mo |
100 TB Data, Heavy Query Load
| Component | Snowflake Enterprise | BigQuery Editions (Standard) | Redshift Provisioned (6x ra3.4xl) |
|---|---|---|---|
| Storage (100 TB) | $2,300/mo | $2,000/mo | $2,340/mo |
| Compute (12 hrs/day, heavy) | $10,800/mo | ~$5,760/mo (400 slots) | $14,083/mo |
| Total | $13,100/mo | $7,760/mo | $16,423/mo |
At heavy scale, BigQuery Editions is significantly cheaper — roughly half the cost of Redshift and 40% less than Snowflake. This gap widens further with reserved capacity commitments.
Feature Comparison Beyond Pricing
| Feature | Snowflake | BigQuery | Redshift |
|---|---|---|---|
| Cloud support | AWS, Azure, GCP | GCP only | AWS only |
| Data sharing | Best (Snowflake Marketplace) | Analytics Hub | Data sharing (limited) |
| Semi-structured data | VARIANT type (native) | Nested/repeated fields | SUPER type |
| ML integration | Snowpark | BigQuery ML (built-in) | SageMaker integration |
| Concurrency handling | Multi-cluster (auto-scale) | Automatic | Concurrency scaling (limited free hours) |
| Maintenance required | Low | None (serverless) | Medium (vacuuming, distribution keys) |
Which Should You Pick?
Pick Snowflake If:
- You’re multi-cloud (AWS + Azure, or considering switching providers)
- Data sharing with partners/customers is a primary use case
- You want workload isolation (separate warehouses for ETL vs analytics)
- Your team values the simplest DBA experience
Pick BigQuery If:
- You’re already on Google Cloud Platform
- Query volumes are unpredictable (on-demand pricing excels here)
- You want built-in ML without managing separate infrastructure
- You prefer zero cluster management
Pick Redshift If:
- You’re deeply invested in AWS and want tight S3/SageMaker integration
- Workloads are steady and predictable (reserved instances save 30%+)
- Your team has existing PostgreSQL expertise (Redshift is Postgres-based)
- You need Spectrum to query data directly in S3 without loading it
Best overall value at scale: BigQuery Editions — the slot-based pricing is the most cost-efficient for heavy analytics workloads, and zero maintenance is hard to beat.
Best for multi-cloud flexibility: Snowflake — the only option that runs identically across AWS, Azure, and GCP.
Best for AWS-native teams: Redshift — if your data lake is in S3 and your ML pipeline is in SageMaker, keeping the warehouse in AWS makes sense despite the higher price.
Best for small/unpredictable workloads: BigQuery on-demand — 1 TB/month free and pay-per-query means you only spend money when you’re getting value.
FAQ
Can I start free with any of these?
BigQuery gives 1 TB/month of free queries indefinitely. Snowflake offers $400 in trial credits (about 2 weeks of moderate use). Redshift has a 2-month free trial with a dc2.large node.
Which is cheapest for small startups?
BigQuery on-demand. If you’re scanning under 1 TB/month, it’s literally free. Snowflake and Redshift both require minimum spend to be useful.
Do I need a data engineer to manage these?
BigQuery: No — it’s fully serverless. Snowflake: Minimal — auto-suspend and auto-scaling handle most operations. Redshift Provisioned: Yes — you’ll need someone managing distribution keys, sort keys, vacuum operations, and cluster resizing.
Can I migrate between them?
Data migration is straightforward (export to Parquet/CSV, reload). The painful part is rewriting SQL — Snowflake and BigQuery use slightly different SQL dialects, and Redshift has Postgres-specific syntax. Budget 2–4 weeks for a full migration of a medium-complexity warehouse.
