Snowflake does not charge a flat monthly subscription. You pay for what you consume across three separate meters, which is great for elastic workloads and painful for budgeting if nobody watches the spend. That unpredictability is exactly why disciplined cloud cost management matters from day one.

This guide breaks down every part of Snowflake pricing, gives you per-credit rates by edition, and walks through real dollar examples so you can estimate a bill before the invoice lands. The short version: you pay for compute in credits, storage in compressed terabytes, and data transfer in terabytes moved out.

Compute is where most of the money goes. A single credit can cost anywhere from about $2 to $6, depending on your edition, region, and whether you buy on demand or on a contract.

What Snowflake Pricing Is Based On

Snowflake's consumption model rests on three meters that run independently. Knowing which one is draining your budget is the first step toward any FinOps discipline.

• Compute: Virtual warehouses burn Snowflake credits while they run. Credits are billed per second with a 60-second minimum each time a warehouse starts or resumes, per Snowflake's compute documentation. A suspended warehouse costs nothing.

• Storage: You pay a flat monthly rate per compressed terabyte. Snowflake compresses data automatically on load, so your billed volume is usually far smaller than the raw file size.

• Data transfer: Ingress is free. You pay a per-terabyte fee only when data leaves a Snowflake account for a different region or a different cloud provider.

Because these meters are decoupled, a team can run tiny storage next to a massive compute bill, or the reverse. Treating Snowflake as one line item is the fastest way to lose track of the money, which is why granular cloud cost allocation pays off early.

Snowflake Credit Pricing by Edition

A credit is the unit of compute, and its dollar value depends on which edition you run. As a rough on-demand baseline for AWS US East, Snowflake's published rates land roughly here:

Two things move these numbers. Non-US regions usually carry a premium over the US baseline, and a capacity commitment can cut the rate well below on-demand. Mapping spend back to teams works much like SaaS unit economics: the headline rate means little until you tie it to the work it produces.

Warehouse Sizes and Credits Per Hour

The compute cost is driven by warehouse size. Each step up to the next size roughly doubles both the compute power and the credits billed per hour, as Snowflake sets out in its service consumption table.

The pattern compounds fast. Here is what the same warehouse costs if you run it 8 hours a day for 30 days on Standard at $2 a credit:

Oversizing is the single most expensive mistake in Snowflake. An X-Large costs four times as much as a Medium for the very same schedule, so catching it early depends on solid cloud cost forecasting tools rather than a month-end surprise.

Storage and Data Transfer Costs

Storage is the predictable part of the bill. On AWS US East, the on-demand rate is $23 per terabyte per month, as set out in Snowflake's cost documentation, and it drops lower for capacity customers.

Compression helps more than people expect. Because Snowflake compresses on load, a raw terabyte of files often shrinks to 200 to 300 GB of billed storage. Data transfer is the part that ambushes teams, so keep these rules of thumb in mind:

• Loading data into Snowflake is free.

• Moving data out to another region on the same cloud, or to a different cloud entirely, triggers a per-terabyte egress fee.

• Cross-cloud replication for disaster recovery quietly multiplies transfer charges.

These egress surprises behave like the spikes you watch for with cloud cost anomaly detection tools: small, recurring, and invisible until someone reconciles the invoice.

Serverless and Cortex AI Costs

Beyond warehouses, Snowflake runs serverless features that burn credits outside your warehouse allocation. Snowpipe, automatic clustering, materialized view maintenance, and search optimization all qualify. They do not show up in standard warehouse monitoring, so they are a common source of unexplained spend.

Cortex pushes this further by billing AI and LLM functions per token, much like the major model providers do. Every prompt and completion you send to a Cortex function lands on the same invoice as your warehouse credits. A chatbot that summarizes support tickets can quietly consume thousands of credits a month if nobody caps its tokens.

That metering is why understanding token economics inside the warehouse is now part of reading a Snowflake bill. The cost of a query scales with input and output tokens, so a verbose prompt template can double a workload's price without anyone noticing.

Teams running models across several platforms feel this most. When Cortex usage sits next to direct provider spend, the attribution problem is the same one solved by a multi-provider LLM cost management tool.

Without that visibility, a finance lead sees one large number and cannot tell whether the data team or a new AI feature drove it. Dedicated LLM cost allocation tools keep every token traceable to the team that triggered it, so AI spend never hides inside a generic data bill.

On-Demand vs Capacity Pricing

There are two ways to buy Snowflake, and the gap between them is large.

• On-demand: you pay the published per-credit rate with no commitment. Simple, flexible, and the most expensive per credit.

• Capacity: you pre-purchase a dollar amount of usage upfront. In exchange, the per-credit rate drops on a sliding scale, with discounts that deepen as your commitment and term grow.

The trade-off is classic. On-demand protects you from overcommitting, while capacity rewards predictable usage with a lower rate. The risk is buying capacity you cannot consume, so commit only after a few months of usage data and a forecast you trust.

That caution is the same instinct behind broader cloud cost optimization: never lock in spend you cannot defend with real usage evidence.

Hidden Costs and How to Control Them

The published rates are honest. The surprises come from how the platform gets used. The most common culprits:

• Idle warehouses: A warehouse left running between queries keeps billing. Aggressive auto-suspend, often set to 60 seconds, is the simplest fix.

• Oversized warehouses: As the example above showed, an X-Large costs four times a Medium. Right-size to the query, not the worst case.

• Serverless creep: Snowpipe, clustering, and Cortex accrue quietly. Track them as their own line items.

• Untagged usage: Without metadata, you cannot tell which team drove a spike. A disciplined set of tagging strategies makes every credit traceable.

Each leak is small on its own and easy to dismiss. Added together across a quarter, they are routinely the difference between a bill that lands on forecast and one that quietly doubles.

Governance turns those leaks from recurring surprises into managed line items. The same principles behind AI cost governance tools apply to a data warehouse: set budgets, alert on drift, and assign an owner to every dollar.

How to Track and Allocate Snowflake Spend

Knowing the rates is half the job. The other half is attributing spend back to the teams, products, and pipelines that generate it, then deciding who pays for it.

When that attribution is missing, every cost review becomes an argument about whose workload is to blame. That is a reporting problem, the same one that makes chargeback vs showback a recurring question in finance meetings.

Amnic gives engineering and finance teams a shared, read-only view of consumption across data and cloud platforms. A Snowflake credit spike and a compute bill sit in the same dashboard instead of two disconnected exports.

Strong cost allocation ties every credit to an owner. It turns a vague "the warehouse bill went up" into a specific conversation with the team responsible for it.

That ownership is the foundation for the metric leadership actually wants. Tracking unit economics converts the raw bill into a cost-per-customer or cost-per-pipeline number that a CFO can act on.

A pipeline costing $0.40 per processed record is a decision; a $40,000 monthly invoice is just a number. The discipline you build for Snowflake carries to every other consumption cloud, so the same playbook works when you start governing Vertex AI cost optimization tools or a growing fleet of warehouses.

The Bottom Line

Snowflake pricing is straightforward once you separate the three meters: credits for compute, compressed terabytes for storage, and per-terabyte fees for egress. The bill turns unpredictable not because the rates are hidden, but because consumption is elastic and easy to leave unmanaged.

Right-size your warehouses, suspend them aggressively, tag everything, and tie each credit back to an owner. Do that, and Snowflake's pay-for-what-you-use model works for your budget instead of against it.

FAQs

How much does Snowflake cost per month?

There is no fixed monthly fee. You pay per credit for compute, per compressed terabyte for storage, and per terabyte for egress. A Medium warehouse running 8 hours a day on the Standard edition costs roughly $1,920 a month at about $2 per credit.

How much is a Snowflake credit?

A credit costs about $2 on Standard, $3 on Enterprise, and $4 on Business Critical for on-demand AWS US East. Rates rise in non-US regions and fall under capacity commitments, so the real range runs from roughly $2 to $6.

How are Snowflake warehouse costs calculated?

By size and runtime. An X-Small burns 1 credit per hour, and each size up roughly doubles it, reaching 128 credits per hour at 4X-Large. Credits bill per second with a 60-second minimum, and a suspended warehouse costs nothing.

What is the difference between on-demand and capacity pricing?

On-demand charges the published per-credit rate with no commitment. Capacity pricing pre-purchases usage upfront in exchange for a lower per-credit rate on a sliding scale. Capacity saves money only if you reliably consume what you buy.

Does Snowflake charge for data transfer?

Loading data in is free. You pay a per-terabyte egress fee when data moves to a different region on the same cloud or to another cloud provider. Cross-cloud replication for disaster recovery is a frequent source of unexpected transfer charges.

What are Snowflake serverless and Cortex costs?

Serverless features like Snowpipe, automatic clustering, and search optimization consume credits outside your warehouse allocation. Cortex AI functions bill per token, similar to LLM API providers, so they need to be tracked as their own line items.