How does Snowflake compute pricing compare to BigQuery and Redshift?

Snowflake's Credit-Based Compute Model

Snowflake separates compute from storage and bills compute in credits consumed per second (with a 60-second minimum) based on warehouse size. Warehouse sizes range from X-Small (1 credit per hour) to 6X-Large (512 credits per hour), and credit prices vary by edition: Standard (~$2/credit on AWS US East), Enterprise (~$3), and Business Critical (~$4). Storage is billed separately at roughly $23/TB/month on-demand, with additional costs for Time Travel and Fail-safe storage at ~$40/TB/month.

The key advantage of Snowflake's model is the ability to suspend warehouses completely when idle, meaning you pay zero compute during inactive periods. However, credit consumption can be opaque — a single poorly tuned warehouse running large queries can burn through credits rapidly. Cloud Services compute (metadata operations) is free up to 10% of warehouse spend, after which it bills separately. Materialized view auto-refresh and Search Optimization Service can silently add 5-15% to credit bills.

Capacity contracts are Snowflake's primary negotiation lever. Multi-year commitments (1, 2, or 3 years) unlock 10-50% off list price, with larger commitments yielding deeper discounts. At $1M+/yr annual spend, discounts typically reach 25-40%; at $5M+, they can hit 40-55%. Multi-cloud commitments (AWS + Azure or AWS + GCP) add another 5-10% discount. Snowflake also offers migration credits for teams moving from BigQuery or Redshift, and Cortex AI credits can be negotiated as a separate line item with usage-based true-up.

BigQuery's Per-Byte and Slot Models

BigQuery offers two distinct compute pricing modes. On-demand charges ~$6.25/TB scanned (US multi-region, list price), with no compute SKU — you pay only for the data your queries process. Editions (capacity slots) provide predictable pricing: Standard (~$0.04/slot-hour), Enterprise (~$0.06), and Enterprise Plus (~$0.10), with autoscaling and 1-year or 3-year commitments offering 20-40% discounts.

The on-demand model is ideal for small, bursty workloads with disciplined SQL. A team of 5 analysts scanning ~500GB/day with 2TB stored would pay roughly $135/month — cheaper than Snowflake or Redshift for that pattern. However, on-demand costs can explode if junior analysts write SELECT * against partitioned 50TB tables, potentially costing hundreds of dollars per query. Partitioning, clustering, and custom quotas are essential guardrails.

BigQuery Editions provide cost predictability for larger workloads but have a critical gotcha: slot reservations bill continuously even when idle, unlike Snowflake's auto-suspending warehouses. A 100-slot baseline at Enterprise tier costs ~$2,900/month regardless of usage. Streaming inserts cost extra over batch loads ($0.01/200MB), though the Storage Write API is cheaper and now the default for high-throughput ingest. Long-term storage tier ($0.01/GB/month) kicks in at 90 days untouched, but a single SELECT bumps data back to active pricing ($0.02/GB/month).

Redshift's Node-Based and Serverless Pricing

Redshift has two distinct SKU families. Provisioned RA3 nodes separate compute from managed storage: ra3.xlplus at ~$1.086/hour, ra3.4xlarge at ~$3.26/hour, and ra3.16xlarge at ~$13.04/hour (US East, list). Managed storage costs ~$24/TB/month. Reserved Instances on RA3 provide 1-year (~30% off) or 3-year (~60% off) discounts, making this the most cost-effective option for predictable, always-on workloads.

Redshift Serverless bills per RPU-second at ~$0.375/RPU-hour list (US East), with a 60-second minimum and 8 RPU floor. This means even tiny workloads cost ~$60/month minimum if always-on. Serverless autoscales for variable workloads but has no Reserved Instance structure — though AWS announced negotiable per-RPU pricing for $500K+/yr commits at re:Invent 2025.

For AWS-native shops, Redshift often wins on data gravity alone. If your data already lives in S3 with Lake Formation and IAM integration, Redshift avoids costly cross-cloud egress fees ($0.08-0.12/GB to non-AWS clouds). However, hidden costs include Concurrency Scaling beyond the free hour (can 2-3x bills during reporting peaks), Spectrum queries against S3 ($5/TB scanned separate from Redshift compute), and cross-AZ traffic for replicas.

Detailed Workload Comparison Math

All figures below are list-price approximations for US-East regions, May 2026. Actual contract pricing varies 30-60% with commitments.

Workload 1: 5 analysts, ~50 ad-hoc queries/day, ~500GB scanned/day, 2TB stored

• Snowflake: XS warehouse ~2 hrs/day × $2 + storage ~$50/mo = ~$170/mo
• BigQuery on-demand: ~15TB/mo scanned × $6.25 + storage $40 = ~$135/mo
• Redshift Serverless: 8 RPU × ~3 hrs/day × $0.375 = ~$280/mo
• Winner: BigQuery on-demand — pay-zero-when-idle model dominates for small teams

Workload 2: Mid-market BI dashboard, 24/7 light read, 10TB stored

• Snowflake: S warehouse on auto-suspend, ~6 credit-hrs/day × $2 = ~$580/mo
• BigQuery Enterprise: 100 slot baseline ~$2,900/mo (or on-demand ~$1,500 if scans stay disciplined)
• Redshift: 1× ra3.xlplus reserved 1yr = ~$540/mo
• Winner: Redshift RA3 reserved — fixed hourly rate beats Snowflake's per-second billing for always-on workloads

Workload 3: Streaming ETL, 500GB/day ingest, hourly transforms, 30TB stored

• Snowflake: M warehouse 8 hrs/day × $2 × 4 credits/hr = ~$2,200/mo + storage $700 = ~$2,900/mo
• BigQuery: Storage Write API + 200 slots = ~$5,800/mo + storage $600 = ~$6,400/mo
• Redshift Serverless: 32 RPU × 8 hrs × $0.375 × 30 = ~$2,880/mo
• Winner: Snowflake/Redshift tie — per-second billing on suspended warehouse helps Snowflake, but Redshift's lower RPU cost is competitive

Workload 4: Enterprise ML feature engineering, 50TB scanned/week, 200TB stored

• Snowflake: L warehouse w/ Snowpark ~$18,000/mo + storage ~$4,600 = ~$22,600/mo
• BigQuery: Enterprise Plus 500 slots reserved ~$36,000/mo + storage $4,000 = ~$40,000/mo
• Redshift: ra3.4xlarge × 4 nodes reserved 3yr ~$5,600/mo + RMS $4,800 = ~$10,400/mo
• Winner: Redshift RA3 reserved — 3-year commitment on provisioned nodes crushes the competition for predictable large-scale workloads

Workload 5: AI inference + LLM-generated SQL, 1M queries/month via Cortex/Gemini/Bedrock

• Snowflake Cortex: Llama 3.1 70b ~$1.21/1M tokens out; expect ~$8K-25K/month at this volume
• BigQuery ML + Gemini integration: ~$0.0025/1K tokens for Gemini 1.5 Flash; generally ~$3K-12K/month
• Redshift ML (SageMaker passthrough): ~$5K-15K/month + SageMaker endpoint costs separately
• Winner: BigQuery + Gemini — cheapest top-tier token pricing by 3-4x over Snowflake Cortex

Hidden Costs That Bite Each Platform

Snowflake gotchas:

• Cloud Services compute bills after exceeding 10% of warehouse spend
• Materialized View auto-refresh consumes credits silently
• Marketplace and Cortex AI listings billed in credits — easy to forget
• Time Travel + Fail-safe storage stacks on base storage (90 days × large tables = real money)
• Cross-region/cross-cloud replication: egress + destination storage + replication credits
• Search Optimization Service adds 5-15% to credit bill on enabled tables

BigQuery gotchas:

• Slot reservations bill continuously — no auto-suspend like Snowflake warehouses
• Streaming inserts cost extra over batch loads; many teams forget to migrate to Storage Write API
• Materialized View re-compute costs hit slot capacity or on-demand bytes
• Data egress to non-Google clouds: $0.08-0.12/GB — kills multi-cloud architectures
• Long-term storage tier resets to active pricing on any SELECT against the table
• BI Engine reservations separate from query slots — easy to double-pay

Redshift gotchas:

• RA3 Managed Storage cheap (~$24/TB) but cross-AZ traffic for replicas isn't free
• Concurrency Scaling beyond free hour can quietly 2-3x bills during reporting peaks
• Serverless 8 RPU floor means even tiny workloads cost ~$60/month minimum if always-on
• Spectrum queries against S3 charge $5/TB scanned — separate from Redshift compute
• DataShare consumer-side compute billed to consumer; producers often surprised
• Backups beyond free retention period bill at S3 standard rates

Negotiation Levers in 2026

Snowflake (post-Sridhar Ramaswamy era, more aggressive on price):

• Multi-year capacity commits: 25-40% off list at $1M+/yr; 40-55% at $5M+
• Multi-cloud commit (AWS + Azure or AWS + GCP): adds 5-10% additional discount
• Migration credits if coming from BigQuery/Redshift — Snowflake funds POCs
• Cortex AI commit carve-out: negotiate AI credits as separate line item with usage-based true-up
• Procurement signal: mention Databricks evaluation in writing — discounts move 10-15%

BigQuery / Google Cloud:

• Committed Use Discounts on Editions: 20% (1yr) / 40% (3yr) standard
• Enterprise Agreement gets another 5-10% if bundled with GCP infra spend
• Gemini token commit: Google is buying market share — ask for AI credits parity vs. OpenAI
• Multi-region storage discount: data sovereignty asks (EU, India) move pricing
• Procurement signal: mention Snowflake-on-GCP — Google sales will discount to keep workload native

Redshift / AWS:

• 3yr Reserved Instance on RA3: ~60% off list, no upfront option available
• Enterprise Discount Program (EDP) — bundled S3, EC2, Redshift commit gets 15-25% across the board
• Serverless commit announced at re:Invent 2025: now negotiable per-RPU pricing for $500K+/yr commits
• Free DMS migration + Professional Services credits if migrating from competitor warehouse
• Procurement signal: mention Iceberg + open-table strategy — AWS will negotiate to keep you in Redshift vs. open-source Trino/Athena

The AI Workload Question

Snowflake Cortex prices LLM inference in credits — convenient billing but opaque. Llama 3.1 70b runs ~$1.21/1M output tokens, fine-tunes priced in credit-hours. Embedding functions are cheap, but generative LLM calls become expensive at scale. BigQuery + Gemini integration is currently the cheapest path for high-volume LLM enrichment — Gemini 1.5 Flash at ~$0.30/1M output tokens beats Cortex by ~3-4x for equivalent quality on summarization and classification tasks.

Redshift ML outsources to SageMaker — you pay SageMaker endpoint costs (instance-hour pricing) plus Redshift compute for the SQL wrapper. This is the worst $/inference of the three for ad-hoc generative work but best for batch scoring at predictable volume. The overcharge today: Snowflake Cortex is the most expensive per-token for top-tier models (GPT-4-class), justified by zero data movement. If your AI workload is 60%+ of platform spend, BigQuery + Gemini or Redshift + Bedrock will beat Snowflake on raw economics — but factor in egress and security review costs.

For vector search, Snowflake (native Cortex Search), BigQuery (Vector Search GA 2025), and Redshift (pgvector via Aurora integration) all offer solutions. Snowflake currently has the simplest TCO story for embedding-heavy RAG workloads kept entirely inside the warehouse.

Pricing Comparison Table

*All figures list-price approximations May 2026; actual contract pricing varies 30-60% with commits.*

Related questions

Which platform is cheapest for small, bursty ad-hoc queries?

BigQuery on-demand typically wins because you pay only for data scanned per query with no cost when idle. Snowflake requires a running warehouse, and Redshift Serverless charges for compute capacity even during idle periods.

How do Snowflake credits translate to real dollar costs?

Credits are consumed per second based on warehouse size (XS=1 credit/hr, L=8 credits/hr). Credit price varies by edition ($2-4/credit) and region. Multi-year contracts reduce per-credit cost by 30-50%.

Can BigQuery on-demand costs spiral out of control?

Yes — a single SELECT * against a 50TB table could cost hundreds of dollars. Partitioning, clustering, and custom quotas are essential guardrails, but SQL discipline from all users is required.

Is Redshift always cheaper for AWS-native workloads?

Not always, but data gravity often makes it most economical. Redshift avoids costly cross-cloud egress fees when data already lives in S3. Snowflake or BigQuery may still be cheaper for bursty workloads.

What hidden costs should I watch for with Snowflake?

Opaque credit consumption from poorly tuned warehouses is the main risk. Cloud Services costs, Materialized View auto-refresh, and Search Optimization Service can silently add 15-25% to bills.

How do multi-year contracts affect pricing for each platform?

Snowflake offers 30-50% off for committed capacity. BigQuery offers 20-40% off for 1-3 year slot commitments. Redshift provides 30-60% off on RA3 Reserved Instances but Serverless has no long-term discount structure.

FAQ

Which platform is cheapest for small, bursty ad-hoc queries? BigQuery on-demand typically wins here because you only pay for the data scanned per query, with no cost when no queries run. Snowflake requires a running warehouse (even if auto-suspended, you pay for the time it was active), and Redshift Serverless charges for compute capacity even during idle periods. For occasional queries under a few TB per month, BigQuery's per-query model is often the most cost-effective.

How do Snowflake credits translate to real dollar costs? Snowflake bills per second for warehouse compute, with credits consumed based on warehouse size (e.g., X-Small = 1 credit per hour, Large = 8 credits per hour). Actual cost per credit varies by cloud region and contract tier, typically ranging from $2–$4 per credit. Without a capacity commitment, list prices apply; multi-year contracts can reduce per-credit cost by 30–50%.

Can BigQuery on-demand costs spiral out of control? Yes, especially if queries scan large tables without filters or use SELECT *. BigQuery charges per TB of data processed, and a single poorly written query against a 50TB table could cost hundreds of dollars. Using partitioning, clustering, and cost controls (e.g., custom quotas) helps, but requires SQL discipline from all users.

Is Redshift always cheaper for AWS-native workloads? Not always, but data gravity often makes it the most economical choice. If your data already lives in S3 and you use Lake Formation, IAM, and other AWS services, Redshift avoids costly cross-cloud egress fees. Redshift Serverless pricing is per RPU-hour (typically $0.50–$1.00 per RPU-hour), while provisioned RA3 nodes have fixed hourly rates. However, Snowflake or BigQuery may still be cheaper for bursty or unpredictable workloads.

What hidden costs should I watch for with Snowflake? The main hidden cost is opaque credit consumption from poorly tuned warehouses. If you keep a large warehouse running for small queries, or run many concurrent queries on a single warehouse, credits burn faster than expected. Also, cloud services costs (e.g., data transfer out of Snowflake's region) are separate from compute credits. Monitoring warehouse-level usage is essential to avoid surprises.

How do multi-year contracts affect pricing for each platform? Snowflake offers the most aggressive discounts for committed capacity, often 30–50% off list price for 1–3 year terms. BigQuery has flat on-demand rates but offers flat-rate reservations (e.g., 500 slots for $X/hour) with discounts for 1–3 year commitments. Redshift provides Reserved Instance pricing for provisioned nodes (e.g., 30–60% off on-demand) but Serverless has no long-term discount structure. The best deal depends on your workload predictability and willingness to commit.

Sources

• Snowflake official documentation — pricing model, compute credits, and warehouse tiers
• Google Cloud BigQuery pricing page — on-demand and flat-rate compute costs
• Amazon Redshift pricing page — cluster-based and serverless compute pricing
• Gartner — cloud database management system market analysis and vendor comparisons
• AWS re:Post — community and official discussions on Redshift vs. Snowflake pricing
• Stack Overflow — user experiences and cost comparisons for Snowflake, BigQuery, and Redshift
• Google Cloud blog — BigQuery Editions and committed use discount announcements
• Snowflake investor relations — multi-year contract discount disclosures in SEC filings

Related on PULSE

• [Will Snowflake survive the AWS Redshift + Microsoft Fabric squeeze through 2027?](/knowledge/q1575)
• [What are the LLM fine-tuning compute requirements in 2027?](/knowledge/q12298)
• [What is the right way to compute true gross retention vs net retention when half your customers are on multi-year contracts with annual escalators?](/knowledge/q9518)
• [How does Snowflake onboarding compare to Databricks?](/knowledge/q1601)
• [How are 2027 buying committees using generative AI to compare vendor pricing before any contact?](/knowledge/q16659)
• [What specific negotiation tactics work in 2027 when enterprise buyers use AI to compare your pricing against 50 competitors in real time?](/knowledge/q13583)