Snowflake pricing 2026: the complete credit reference.
Every Snowflake credit rate across four editions, three clouds and 30-plus regions, plus Cortex AI credit cost and Reserved Capacity discount bands at $1M, $3M and $10M annual commit. The consumption model rewards edition and region discipline over warehouse sizing — and the five optimisation patterns below recover 20 to 35 percent of spend without renegotiation.
Snowflake credit prices run from $2.00 per credit on Standard in US East to $5.40 on VPS in EU — a 2.7x spread that puts edition and region choice above warehouse sizing in cost impact. Realised Capacity discounts land at 10–18% at $1M commit, 18–28% at $3M and 28–40% at $10M. The five optimisation patterns hit the bill harder than the negotiated discount does.
01 Key findings
Edition and region beat warehouse sizing on cost impact. The per-credit spread from Standard US East to VPS EU is 2.7x. The EU and APAC region premium over US runs 30 to 55 percent — a larger lever than most warehouse-tuning programmes deliver.
Bigger warehouses often cost less per query. Doubling warehouse size doubles credit consumption but frequently finishes in under half the time. Medium to Large gives the best credit-per-query economics; 2X-Large and above show sharply diminishing returns.
Auto-suspend is the single largest billed-hours lever. The default 600-second suspend leaks 5 to 12 percent of total credits to idle time on intermittent workloads; tightening to 60 seconds recovers most of it.
Capacity discounts scale with commit, not skill. $1M commit realises 10 to 18 percent, $10M reaches 28 to 40 percent, and three-year terms add 4 to 7 points. Storage and Cortex carry their own, steeper discount curves.
Cortex AI and Marketplace are the fast-growing blind spots. Cortex credit cost can become the dominant line item on AI-heavy estates; Marketplace apps consume the customer's credits but are bought by individual teams. Both need explicit quarterly control.
02 Snowflake credit model
Snowflake bills usage in credits. A credit represents one minute of compute on an X-Small warehouse at the Standard edition price point. Larger warehouses consume credits at higher multiples (Small = 2, Medium = 4, Large = 8, X-Large = 16, 2X-Large = 32, 3X-Large = 64, 4X-Large = 128). Higher editions consume credits at the same physical rate but at a higher credit price.
The customer pays per credit consumed plus a separate storage charge. The credit price varies by edition (Standard, Enterprise, Business Critical, Virtual Private Snowflake), by cloud (AWS, Azure, GCP) and by region. Realised cost is a function of three multipliers: warehouse size, edition and region.
03 The four editions
Snowflake sells four edition tiers. Edition is set at account creation but can be upgraded mid-contract. Per-credit prices below are AWS US East.
| Edition | Per-credit price (AWS US East) | Key features included |
|---|---|---|
| Standard | $2.00 | Encryption, role-based access, Time Travel up to 1 day |
| Enterprise | $3.00 | Standard plus multi-cluster warehouses, materialized views, search optimization, dynamic data masking, Time Travel up to 90 days |
| Business Critical | $4.00 | Enterprise plus HIPAA, PCI-DSS, customer-managed keys, AWS PrivateLink and Azure Private Link, Tri-Secret Secure |
| Virtual Private Snowflake (VPS) | $5.00 to $5.40 + commit minimum | Business Critical plus dedicated metadata store, isolated tenant deployment |
Most enterprise deployments land at Business Critical. Standard suits non-production workloads; Enterprise is the default for production without regulatory requirements. Business Critical is required for healthcare (HIPAA), payment data (PCI-DSS) and any workload needing customer-managed keys or Private Link. VPS is rare, typically justified only for sovereign-cloud or single-tenant regulatory requirements.
04 Pricing by cloud and region
Per-credit price across the most-used regions. The EU and APAC premium over US runs 30 to 55 percent — for workloads with no data-residency obligation, region choice alone is a 30 to 55 percent lever.
| Region | Standard | Enterprise | Business Critical |
|---|---|---|---|
| AWS US East (Virginia) | $2.00 | $3.00 | $4.00 |
| AWS US West (Oregon) | $2.00 | $3.00 | $4.00 |
| AWS EU (Ireland) | $2.60 | $3.90 | $5.20 |
| AWS EU (Frankfurt) | $2.70 | $4.05 | $5.40 |
| AWS AP (Sydney) | $3.10 | $4.65 | $6.20 |
| Azure East US 2 | $2.00 | $3.00 | $4.00 |
| Azure West Europe | $2.60 | $3.90 | $5.20 |
| GCP us-central1 | $2.00 | $3.00 | $4.00 |
| GCP europe-west4 | $2.60 | $3.90 | $5.20 |
Enterprise per-credit rate by region, indexed to the APAC high:
For workloads with EU data residency obligations under GDPR or Schrems II, the EU premium is unavoidable. For US-servable analytics workloads, it is pure discretionary cost.
05 Warehouse sizing economics
Warehouse size doubles credit consumption at each tier. The naive intuition is that bigger warehouses cost more. The counterintuitive truth is that bigger warehouses often cost less per query because they finish in less than half the time.
| Warehouse size | Credits per hour | Cost per hour (Enterprise, US East) | Typical query speedup vs X-Small |
|---|---|---|---|
| X-Small | 1 | $3.00 | 1.0x |
| Small | 2 | $6.00 | 1.9x |
| Medium | 4 | $12.00 | 3.7x |
| Large | 8 | $24.00 | 7.2x |
| X-Large | 16 | $48.00 | 13.5x |
| 2X-Large | 32 | $96.00 | 24x |
| 3X-Large | 64 | $192.00 | 38x (diminishing) |
| 4X-Large | 128 | $384.00 | 52x (diminishing) |
For most workloads, Medium to Large warehouses offer the best credit-per-query economics. Diminishing returns above X-Large mean 2X-Large and above are typically wrong unless the query is unusually parallelisable. Auto-suspend (set to 60 seconds in most cases) is the single largest billed-hours lever; the default 600-second auto-suspend leaks 5 to 12 percent of total credits to idle time on intermittent workloads.
Most Snowflake estates run X-Large or larger because someone once needed it for a heavy query and never resized. Resizing to Medium or Large and relying on auto-scaling for peaks typically cuts credit consumption by 25 to 45 percent on BI-style workloads.
06 Cortex AI and ML cost
Snowflake Cortex bundles LLM functions, Document AI and Cortex Analyst into Snowflake-native AI features. Cortex consumes credits per token or per request, on top of the credit cost of the warehouse running the query.
| Cortex feature | Pricing unit | Rate (Enterprise tier) |
|---|---|---|
| Cortex LLM Functions (Llama 3 70B) | Per 1M tokens | 0.79 credits in, 0.79 credits out |
| Cortex LLM Functions (Llama 3 8B) | Per 1M tokens | 0.19 credits |
| Cortex LLM Functions (Mistral Large) | Per 1M tokens | 5.10 credits in, 15.30 credits out |
| Cortex LLM Functions (Claude 3.5 Sonnet) | Per 1M tokens | 2.55 credits in, 12.75 credits out |
| Cortex Search | Per serving unit hour | 1 credit per active hour |
| Cortex Analyst | Per request | 0.067 credits per message |
| Document AI extraction | Per page | 0.0006 credits per token |
For AI-heavy Snowflake workloads, Cortex credit cost can become the dominant line item. Compare Cortex pricing against direct OpenAI, Anthropic or Bedrock pricing using our enterprise LLM cost comparison. The Snowflake case is strongest when the prompt context lives in Snowflake tables; it weakens when the context lives elsewhere and Cortex becomes a more expensive proxy.
07 Storage and data transfer
Storage is billed separately from compute. On-demand storage rates are typically $40 per TB per month on AWS, slightly higher on Azure and GCP. Capacity storage rates negotiated at $1 million commit drop to $23 per TB per month. Time Travel and Fail-safe storage are charged at the same rate but consume additional space (Time Travel: up to 90 days of changes for Enterprise edition; Fail-safe: 7 additional days, non-configurable).
Data transfer out of Snowflake is charged at $0.02 to $0.09 per GB depending on cloud and region. For replication across regions, Snowflake charges replication transfer at the same per-GB rate. Cross-region replication for disaster recovery is a frequently overlooked Snowflake line item.
08 Reserved Capacity bands
Snowflake Capacity contracts (committed annual spend) replace on-demand pricing with a prepaid commitment. Realised discounts in advisor-led Snowflake negotiations during 2024 to 2026:
| Annual commit (TCV) | Capacity discount range | Storage discount | Cortex discount |
|---|---|---|---|
| $250K to $1M | 5 to 12 percent | 20 to 35 percent | 0 percent |
| $1M to $3M | 10 to 18 percent | 35 to 45 percent | 0 to 10 percent |
| $3M to $10M | 18 to 28 percent | 45 to 55 percent | 10 to 20 percent |
| $10M+ | 28 to 40 percent | 55 to 65 percent | 20 to 30 percent |
Capacity contracts run 1, 2 or 3 years; three-year terms typically add 4 to 7 points over one-year terms. Unused Capacity carries forward subject to a cap (typically 20 to 30 percent of the year's commit), but year-three unused commit is forfeit on most paper. The negotiation lever is to extend carry-forward into year three and widen the cap.
09 Snowflake Marketplace cost
Snowflake Marketplace is the data exchange where third-party datasets, applications and Native Apps are sold, charged on top of Snowflake credits. The two cost patterns are Native Apps (apps installed into the customer Snowflake account, billed per consumption against the customer's Snowflake commit) and Data Products (datasets accessed via reader accounts, billed to the data provider's account).
Marketplace consumption is a fast-growing line item that finance teams routinely miss. Marketplace apps consume the customer's Snowflake credits but are negotiated separately by individual teams. A quarterly Marketplace audit is a standard control.
10 Cost optimisation patterns
Five patterns deliver most of the savings in Snowflake estates. Independent reviews typically identify 20 to 35 percent of total spend as recoverable through optimisation alone — most of it without Snowflake's agreement.
Right-size warehouses
Reduce X-Large defaults to Medium or Large for routine workloads and rely on auto-scaling for peaks. Savings: 20 to 35 percent of compute credits.
Tighten auto-suspend
Set auto-suspend to 60 seconds for most warehouses instead of the 600-second default. Savings: 5 to 12 percent of compute credits on intermittent workloads.
Result caching & reuse
Snowflake caches results for 24 hours; enabling result reuse eliminates redundant query cost. Savings: 8 to 18 percent on BI workloads.
Search Optimization & MVs
Use Search Optimization Service and Materialized Views selectively; both add cost but reduce query credits for repetitive lookups. Net savings: 10 to 25 percent on selective workloads.
Separate dev / test / prod
Split into separate accounts with Standard edition for dev where the Enterprise feature set is not required. Savings: 25 to 35 percent on dev compute credits.
11 Contract levers
Eight commercial levers move the Snowflake contract beyond headline discount: carry-forward cap raised from 20 to 35 percent of commit; carry-forward extension into year three; a price-protection clause locking per-credit rates for the term; termination for convenience after year one with pro-rated refund; edition flexibility; region flexibility; Cortex AI inclusion in the commit at a defined exchange ratio; and Marketplace consumption inclusion against the commit. Apply them situationally:
Lock per-credit rates for the term and secure carry-forward into year three with a widened cap — the customer, not Snowflake, otherwise carries multi-year commitment risk.
Negotiate edition and region flexibility so accounts can operate across Standard, Enterprise and Business Critical, and draw commit against any region without surcharge or re-quote.
Include Cortex AI and Marketplace consumption in the Capacity commit at a defined exchange ratio — the fastest-growing line items are otherwise billed outside the negotiated discount.
Cut Snowflake cost before renewal
Our Cloud & FinOps practice restructures warehouses, editions, regions and the Capacity contract — most savings need no Snowflake agreement.
12 Snowpark and Iceberg cost
Snowpark is the Snowflake developer framework for executing Python, Java and Scala workloads against Snowflake data. It runs on Snowflake warehouses and consumes the same credits as SQL queries, but at modestly higher per-second consumption due to JVM and Python container overhead. For workloads previously running on Databricks or external Spark clusters, Snowpark can shift compute spend back to Snowflake while reducing the data egress and tooling sprawl of multi-platform analytics.
Iceberg Tables let Snowflake read and write Apache Iceberg formatted tables on customer-managed cloud storage. Compute cost is unchanged but the storage charge moves from Snowflake Capacity storage to the customer's own S3, ADLS or GCS bucket. For very large estates where storage is more than 30 percent of the Snowflake bill, Iceberg can reduce storage cost by 40 to 60 percent versus Snowflake-managed storage, at the cost of additional governance complexity and slightly higher first-access query latency.
13 How contracts shape spend
Three structural choices change cost more than the headline discount: the on-demand to Capacity transition (most enterprises move once monthly spend exceeds $30,000); the term length (multi-year discount versus carry-forward risk); and the deployment topology (single account with role-based isolation versus multi-account with shared Capacity). The under-discussed clause is the Capacity overage rate.
Overage at Capacity rate Negotiate
Have overage consume against next year's Capacity at the Capacity rate, with a refresh credit if overage triggers an early renewal. This neutralises the on-demand premium once the commit is exhausted.
Overage at on-demand rate Default paper
On-demand pricing kicks in when Capacity is fully consumed, at rates 5 to 18 percent higher than Capacity. Without the clause, customers pay that premium for the remainder of the contract year.
For the full data platform comparison, see our Snowflake vs Databricks vs BigQuery guide, Databricks pricing pillar and analytics licensing guide. For the cloud-side commit interaction, see AWS EDP negotiation and Azure MACC versus CTP. For broader cost controls, see our cloud cost optimization guide and cloud contracts guide. To engage on Snowflake negotiation, see our cloud contract negotiation service.
The Licensing Edge
Weekly cloud and licensing intelligence for enterprise IT leaders. 3,000+ subscribers.