Snowflake pricing looks simple at first: pick an edition, pay per credit. The invoice is where it gets harder to read. Credits burn at different rates by warehouse size and generation. Serverless features run on their own meters, and AI features now bill in a separate credit type, AI Credits.
Plenty of teams only learn how the bill works the month it jumps. A Gen1 Medium warehouse left running for 48 hours over a weekend costs $576 on Enterprise Edition, or about $778 on Gen2. A busy table with 90 days of Time Travel can carry several times its own size in historical storage.
Every Snowflake rate below comes from the company’s Service Consumption Table, effective October 29, 2026. We spend extra time on storage retention, since that’s where Snowflake spend and data protection overlap.
How Snowflake pricing works
Snowflake pricing is a consumption-based model: you pay for compute in credits, for storage per terabyte per month, and for data moved out of a region. A suspended warehouse costs nothing.
The price of one credit depends on your edition, your cloud and region, and whether you buy on demand or through a pre-purchased capacity contract.
Here are the on-demand list prices in AWS US East (Northern Virginia):
Storage bills separately, on the daily average of compressed data, and VPS is only sold through an order form, which also sets its deployment fee.
Rates rise in most regions outside the US, and capacity contracts lower the credit price by a negotiated discount.
How Snowflake credits work
A credit is Snowflake’s unit of compute. Snowflake’s consumption table now calls them Platform Credits, to separate them from AI Credits.
Virtual warehouses consume credits by size while they run. Billing is per second, with a one-minute minimum each time a warehouse starts or resumes.
Serverless features like Snowpipe and automatic clustering burn credits on Snowflake-managed compute. Cloud services (logins, metadata, query planning) also uses credits, but you only pay for the part above 10% of each day’s warehouse credits. AI features bill in AI Credits, which cost the same on every edition.
On-demand vs. capacity pricing
On demand, you pay list price for what you used, billed month to month with no upfront commitment.
A capacity contract means buying a set dollar amount of usage up front, usually for a year or more. Snowflake applies the platform credit discount in your order form to the on-demand price for your cloud, region, and edition. Discounts aren’t published. They’re negotiated in each order form, and generally grow with the size and length of the commitment.
The credit discount doesn’t apply to AI Credits. And on contracts under $1.2 million a year, the capacity rates for storage ($23 per TB in US East) and AI Credits ($2.00) match the on-demand list, so the platform credit discount is the whole saving.
How much does a Snowflake credit cost by region?
A Snowflake credit costs $2.00 to $3.25 on Standard Edition, depending on region, and up to $9.75 on VPS in GCP’s Middle East Central 2 (Dammam) region. US regions on all three clouds (plus AWS Mumbai and Azure Central India) sit at the bottom of the range, and most of Europe, Asia Pacific, and South America cost more.
These are on-demand list prices from the Service Consumption Table, matching what the pricing page shows when you pick a platform and region:
Storage barely moves across most regions, while credit prices swing by more than half. US government regions only offer Business Critical and VPS.
If you’ve seen $40 per TB quoted for on-demand storage in US commercial regions, that figure is out of date. The current table lists $23 for both on-demand and the first capacity tier.
6 parts of a Snowflake bill
A Snowflake invoice has six cost buckets, and warehouse compute is usually the biggest.
Virtual warehouse compute
Warehouse credits double with each size. Snowflake now offers two generations of standard warehouse, and Gen2 burns more credits per hour:
At $3 per credit, a Gen1 Medium warehouse costs $12 an hour and a Gen2 Medium on AWS costs $16.20.
A standard warehouse created without a generation setting defaults to Gen2 wherever the region supports it, so new warehouses land on the higher rate by default. Snowflake says most queries finish faster on Gen2, so the higher hourly rate can pay for itself, but only if your runtimes actually drop.
Multi-cluster warehouses (Enterprise and up) bill the per-size rate for every running cluster. Snowpark-optimized warehouses and the newer Interactive warehouses have their own rate tables, and Interactive warehouses bill a 60-minute minimum each time they start.
Serverless features
Serverless features run on Snowflake-managed compute. Each one bills one credit per compute-hour (the equivalent of an X-Small warehouse for an hour), times a feature multiplier. Some also add a per-unit charge.
Table 5 of the consumption table lists multipliers for common features:
Snowpipe charges text files like CSV and JSON on their uncompressed size, and binary formats like Parquet on their observed size. Loading 2,000 GB of CSV in a month uses 7.4 credits, or $22.20 at $3. Snowflake also lists a newer Optima Clustering option, billed at 0.007 credits per uncompressed GB.
Cloud services and the 10% daily adjustment
Cloud services is the layer that handles authentication, metadata, query compilation, and access control. It bills at 4.4 credits per hour of use, but you only pay for the part that exceeds 10% of that day’s warehouse credits.
Say your warehouses use 150 credits on a given day and cloud services uses 25. Snowflake waives 15 (10% of 150) and bills the other 10.
The adjustment is calculated per UTC day, and serverless usage doesn’t count toward the warehouse base. Snowflake’s own four-day example runs 400 warehouse credits and 48 cloud services credits, with 413 credits billed in the end. Workloads made of thousands of tiny queries or heavy metadata operations are the ones that cross the line.
Storage
Storage bills at a flat rate per terabyte per month, based on the average compressed bytes stored each day. That’s $23 per TB in US commercial AWS regions. The total includes table data, files sitting in internal stages, and data held for Time Travel and Fail-safe.
Zero-copy clones start free because they share the original table’s micro-partitions. Each change to a clone creates new micro-partitions that the clone owns and pays for.
Capacity storage discounts only apply to order forms designated as a "Qualifying Order"; everyone else pays the Tier 1 rate. On qualifying orders in US East, the rate drops from $23 per TB to $21.47 at $1.2 million in annual contract value and to $13.80 at $40 million and up.
Snowflake also lists cool and cold archive storage tiers ($4 and $1 per TB per month in US East, with retrieval fees), though archive writes and retrieval processing are still marked as preview.
Data transfer
Snowflake doesn’t charge for data ingress, and transfers within the same region are free. Egress bills per TB when data leaves for another region or another cloud.
From AWS US East, that’s $20 per TB to another AWS region and $90 per TB to a different cloud or the internet. Rates climb in Asia Pacific: from AP Sydney, both routes cost $140 per TB.
Charges apply to unloading data to another region, replication, external functions, external network access from UDFs, and some cross-region Iceberg reads and writes. Query results returned to a Snowflake client or driver don’t count, even across regions.
AI features billed in AI Credits
Snowflake bills its AI features (AI Functions, Cortex Agents, Cortex Search, the Cortex REST API, and AI document parsing) in AI Credits. An AI Credit costs $2.00 with global routing or $2.20 with regional routing, and the price is the same on every edition.
Units vary by feature. From Table 6 of the consumption table:
- AI_COMPLETE with claude-4-sonnet: 1.80 AI Credits per million input tokens and 9.00 per million output tokens
- Cortex Search: 6.3 AI Credits per GB of indexed data per month, billed while the service is resumed
- AI_PARSE_DOCUMENT (layout mode): 3.66 AI Credits per 1,000 pages
- Cortex Analyst (standalone API): 67 Platform Credits per 1,000 messages, since it still bills in Platform Credits
Fifty million input tokens and five million output tokens on claude-4-sonnet come to 135 AI Credits, or $270 with global routing. The warehouse running those queries still bills Platform Credits on top.
Teams pointing AI agents at Iceberg tables can read how we approach AI-ready Iceberg lake architecture.
Monthly Snowflake cost examples: On-demand vs. capacity
Both examples use AWS US East list prices and illustrative usage. Swap in your own numbers.
Example 1: An on-demand Enterprise account
A mid-size analytics team runs three Gen1 warehouses, a few serverless features, and some AI functions, at $3 per credit:
The three warehouses account for $4,680, about 82% of the bill. Storage, transfer, and serverless features together come to under $800.
If the same warehouses ran on Gen2 with no change in runtime, their 1,560 credits would become 2,106, or about $1,638 more a month.
Example 2: A $300,000 capacity contract
A larger team signs a one-year, $300,000 capacity contract on Enterprise Edition. We’ll assume a 15% platform credit discount for illustration (real discounts are negotiated), which brings the credit price to $2.55:
The contract costs $300,000 for usage that would have cost $325,920 on demand, saving $25,920 even with $14,580 of capacity left unused.
All of that saving comes from the 15% platform credit discount. Unused capacity only carries into the next term if your contract allows it, and usage past the commitment is typically billed at on-demand rates until you add capacity.
Hidden Snowflake costs that inflate the bill
The four charges below come from settings someone chose once and never revisited, the same kind of hidden cloud costs that accumulate across a whole cloud estate.
Time Travel and Fail-safe storage
Time Travel keeps changed and deleted data for 1 day by default. On Enterprise Edition and up, you can set permanent tables to any value up to 90 days. After Time Travel ends, Fail-safe holds the data for another 7 days, and you can’t change that window.
Both bill as storage on the data that changed. As a rough estimate, take a 5 TB table that rewrites about 250 GB a day. At 90-day Time Travel, it can carry roughly 22.5 TB of history plus another 1.75 TB in Fail-safe. That’s about $560 a month in retention for a table that costs $115 to store.
Transient and temporary tables have no Fail-safe period and cap Time Travel at one day. Use them for staging and scratch data, and set Time Travel per table so only the data that needs long retention gets it.
Some teams stretch Time Travel to 90 days so it can stand in for backup. It wasn’t built for that. Time Travel history lives in the same account as production, Fail-safe recovery goes through Snowflake Support on a best-effort basis, and nothing survives past day 97.
Upstream of Snowflake, the databases and object storage that feed it need protected copies outside the production account, with retention you control. Eon keeps those in a logically air-gapped vault in open formats like Parquet and Iceberg, so Snowflake queries the history directly with no ETL or restore, and global deduplication cuts their storage cost by up to 50%.
Cloud Backup Posture Management (CBPM), the category we created, continuously discovers new databases and buckets and applies policy by data type, so new sources are covered without tagging.
Serverless features nobody turned off
Automatic clustering, search optimization, and materialized views bill at a 2x multiplier based on how much the underlying data changes. They keep billing on a busy table even if nobody queries it anymore.
Cortex Search works the same way. A resumed search service bills 6.3 AI Credits per GB of indexed data every month, whether anyone searches or not.
Resource monitors only cover warehouses, so they won’t catch any of this. Check the ACCOUNT_USAGE.METERING_DAILY_HISTORY view grouped by service type every month, and set a budget for serverless and AI spend.
Oversized, always-on, and Gen2 warehouses
New warehouses suspend after 600 seconds of inactivity by default. A warehouse that gets a query every nine minutes never suspends at all.
Teams often default to Medium or Large when an X-Small or Small would handle their dashboard queries. On Enterprise, dropping a Gen1 Medium to Small cuts the rate from $12 an hour to $6.
Multi-cluster warehouses with the same minimum and maximum cluster count run every cluster the whole time the warehouse is on. New Gen2 and multi-cluster warehouses also turn on the Query Acceleration Service by default, with a maximum scale factor of 2, so they can lease extra serverless compute on top of the warehouse rate.
Cross-region and cross-cloud transfers
Replication to a second region bills twice: classic replication compute at a 2x multiplier, plus per-TB transfer. A 5 TB monthly change set replicated from AWS US East to another AWS region adds $100 in transfer alone, before the compute.
Unloading to a bucket in a different region, sharing data with consumers in another cloud, and reading Iceberg tables across regions all add egress. Keep Snowflake in the same region as your main data sources and consumers wherever you can, and track the charges you can’t avoid with multi-cloud cost management tools.
How to lower your Snowflake bill
Start with the four changes below. Each one applies a broader cloud cost optimization strategy to the way Snowflake bills.
Right-size warehouses and test Gen2
Snowflake recommends running representative queries on several sizes before settling on one. If a warehouse finishes its work well inside your service level, drop a size and check again.
Split workloads into separate warehouses (ELT, BI, data science) so each can be sized on its own and its spend tracked by team through tagging and chargeback.
Run the same test for Gen1 vs. Gen2. Keep Gen2 only where the faster runtimes cover the 35% higher hourly rate (25% on Azure).
Set auto-suspend to match your idle time
Snowflake suggests 5 to 10 minutes or less for auto-suspend. ELT warehouses that run in batches can go down to 60 seconds. BI warehouses can justify a few minutes more, since suspending drops the local cache and the first queries after a resume run slower.
Don’t set auto-suspend shorter than the usual pause between queries. Each resume bills a full minute, so a warehouse that suspends and wakes up every few minutes pays that minimum again and again.
Add resource monitors and budgets
A resource monitor sets a credit quota for a time period and acts at thresholds you choose: notify, suspend after running queries finish, or suspend immediately. Only the ACCOUNTADMIN role can create one.
Put a monitor on every warehouse with a suspend action near 100% of its expected monthly spend, then use budgets to watch the serverless and AI lines.
Buy capacity once usage is predictable
Snowflake doesn’t publish a minimum, but capacity contracts typically start around $25,000. Below that, on-demand is usually simpler.
Size a commitment from six to 12 months of real usage, modeled in Snowflake’s pricing calculator with your edition, cloud, and region, and leave room for growth on demand. Ask for rollover terms at signing, since unused credits are forfeited at term end by default unless you negotiate a carryforward.
Snowflake vs. Databricks vs. BigQuery vs. Redshift pricing
All four platforms bill for compute by usage, but they meter it differently: Snowflake by credits per warehouse-hour, Databricks by DBUs, BigQuery by data scanned or slot-hours, and Redshift Serverless by RPU-hours.
Sources: Databricks SQL pricing (Premium tier, AWS), BigQuery pricing, and Redshift pricing.
BigQuery on-demand is cheap for occasional queries over small data and expensive for frequent scans of wide tables. Redshift Serverless suits AWS-centered teams with spiky workloads. Databricks Classic and Pro look cheaper per DBU, but the VM bill comes on top.
Snowflake’s per-warehouse-hour model is simple to forecast for steady BI and ELT workloads, since each warehouse size has a fixed hourly rate.
Should you pay for Snowflake?
For teams running steady SQL analytics and ELT on structured data, yes. The pricing works best when warehouses run on a schedule and stop.
Snowflake is a good fit if you:
- Run BI dashboards, ELT pipelines, and ad hoc SQL that a few right-sized warehouses can handle.
- Want compute and storage on one bill, with no clusters or VMs to manage.
- Share data with partners or customers who also use Snowflake.
Look at other options if you:
- Run heavy Spark, ML training, or streaming workloads that fit Databricks’ model better.
- Mostly run occasional queries on small data, where BigQuery on demand may cost less.
- Need warehouses running around the clock at high concurrency, where BigQuery Editions commitments or Redshift reserved nodes can come out cheaper.
Snowflake cost control: Sizing and retention
Snowflake pricing rewards teams that size warehouses to the work and set retention on purpose. Warehouses were about 82% of our sample bill, so right-sizing and auto-suspend are where savings start.
Retention is the other lever. The 5 TB example earlier showed long Time Travel costing several times what the live table costs to store, all of it sitting in the same account as production. For the databases and buckets that feed Snowflake, Eon keeps an isolated, queryable copy outside it.
Are the databases feeding Snowflake protected outside production? Book a demo to see which of them no backup policy covers.
Frequently asked questions
How much does Snowflake cost per month?
Snowflake costs anywhere from under $100 a month to six figures, depending on warehouse hours, edition, and data volume. An X-Small Gen1 Standard warehouse running an hour a day with 1 TB stored costs about $83 a month in US East. Our sample Enterprise account with three warehouses and 20 TB comes to about $5,700.
Is Snowflake very expensive?
No, Snowflake isn’t expensive at list price for workloads that run and then suspend: a Gen1 Medium warehouse costs $12 an hour on Enterprise, or $16.20 on Gen2 in AWS. Bills get expensive when warehouses stay on, Time Travel runs long, or serverless features keep billing in the background.
Is Snowflake free or paid?
Snowflake is a paid service. New accounts get a 30-day trial with $400 in free credits (it ends early if the credits run out), and after that you pay per credit for compute and per TB for storage.
Who is Snowflake’s biggest competitor?
Snowflake’s annual report lists the big cloud providers, AWS, Microsoft Azure, and Google Cloud, first among its competitors. At the product level, that means Amazon Redshift and Google BigQuery, plus Databricks, which competes with Snowflake for SQL analytics and data engineering workloads.
How much does one Snowflake credit cost?
One Snowflake credit costs $2.00 to $9.75 on demand. Standard Edition in US regions is the cheapest at $2.00, and VPS in GCP’s Middle East Central 2 (Dammam) region is the most expensive.
Does Snowflake charge for Time Travel?
Yes. Data held for Time Travel and Fail-safe counts toward storage at your normal per-TB rate, calculated on the data that changed. Long retention on tables that change often can cost more than storing the live table.




