Snowflake Cost Calculator

See what changing a Snowflake warehouse's auto-suspend setting could save, at list price or your own credit price.

Multi-cloudCalculator60 seconds

New here? See what this tool does and how to use it.

Sample workloads

Illustrative examples, not your data and not recommended settings. Each fills in the fields below; change any of them.

Current auto-suspendSHOW WAREHOUSES lists it in seconds under auto_suspend; null or 0 means never.

Compare withThe setting you are considering. Everything else stays the same.

Queries per day, per warehouse

Average query runtime

Hours a day the queries arrive in10 for a working day, 24 for around the clock. Spread evenly: about one every 90s here. No queries outside these hours does not suspend the warehouse; only the timeout does.

10 h

Warehouse size

Your credit priceUSD per credit. Defaults to $3, Enterprise list in AWS US East. Standard is $2, Business Critical $4.

USD

More settingsGen2 on AWS · 4 warehouses · 7 days a week · 1.2x clusters · no storageShow

Warehouse generationRun SHOW WAREHOUSES: resource_constraint says STANDARD_GEN_1 or STANDARD_GEN_2.

CloudGen2 costs 1.35x the standard credits on AWS and Google Cloud, 1.25x on Azure.

Warehouses on this profileEach runs the same pattern. A different workload, such as nightly ELT next to BI: price it on its own and add the two.

Days a week it runsDays with queries. With auto-suspend on, the warehouse sleeps on the others.

Concurrency factorApproximate. Mean clusters running at once under multi-cluster scaling, which needs Enterprise edition or higher; applied the same way to both settings. Leave at 1.0 on Standard.

1.2x

TB stored (optional)Compressed TB in the account, for a total with storage. The same under both settings, so it is not part of the saving.

What would changing auto-suspend save?

Sample numbers

Current: 10 minutes

$24,010.73/mo

Estimated monthly compute

Compare with: 1 minute

$17,884.11/mo

Estimated monthly compute

Estimated monthly saving

$6,126.62/mo

26% less compute

Estimated with the same workload, warehouse, and credit price. Estimated from your workload. Actual suspension timing and query performance can differ. These are sample numbers, not your savings: enter your own workload.

Side by side

Auto-suspend	Billed hours a month	Resumes a month	Monthly compute
Current: 10 minutes	308.8 h	30	$24,010.73
Compare with: 1 minute	230.0 h	12,176	$17,884.11

Hours and resumes are per warehouse; the cost covers 4 warehouses at 5.4 credits/hr and $3.00 a credit, at 1.2 clusters on average (an approximation). Queries keep each warehouse busy 27.1 hours a month under either setting; the rest of the billed time is idle: 91% now.

Shorter timeouts can mean more resumes and loss of the warehouse’s data cache. Test with your workload before changing production. Nothing here changes your warehouse.

At 10 minutes these warehouses never suspend while queries are arriving. A query starts about every 90s, so the timer never runs out until the last query of the day, and 91% of the billed time is idle. A credit report shows only the total, so this is easy to miss.

Warehouse size, at the current setting

Show

One size up

Costs $24,005.47/mo more

Double the credits per hour, roughly half the runtime. The queries still arrive when they do, so the warehouse is awake as long.

One size down

Saves $12,002.74/mo

Half the credits per hour, roughly double the runtime. Rarely saves as much as it looks.

What this model assumes

Show

One engine, two settings. Both figures come from the same model with the same queries, runtimes, warehouse, credit price, days and warehouse count. Only the timeout changes, and a shorter timeout is never assumed to make a query run faster.
Snowflake bills sessions, not queries. The first query resumes a suspended warehouse; it suspends once nothing has run for the auto-suspend time. A query that arrives sooner joins the same session. Each session bills at least 60 seconds, then per second, and queries that overlap are counted once.
Queries arrive evenly through the hours you enter. One every so many seconds from the start of the window, each waiting for the one before it if they would overlap. Real traffic clusters, so real gaps are longer and shorter than these; the result is an estimate for this pattern, not a bound on yours.
Quiet hours do not suspend a warehouse. Only the timeout does. A session still running at midnight, or on Friday night, carries into the next day: the model repeats one week, so a session is never cut off at a boundary or counted twice.
Never means never. AUTO_SUSPEND at 0 or null keeps a warehouse running once it has resumed, so it is priced around the clock, every calendar day. With no queries at all, the warehouse is assumed to be suspended to begin with, so it costs nothing under any setting.
Suspension timing is approximate. Snowflake checks for idle warehouses about every 30 seconds, so real suspensions land a little later than the timeout. No delay is added for that, and no cold-start cost is invented: the data cache a warehouse loses when it suspends shows up as slower queries, not as a line here.
List price unless you enter yours. The credit price starts at $3, Enterprise on-demand list in AWS US East, from Snowflake’s Service Consumption Table. Gen2 warehouses bill 1.35x the standard credits on AWS and Google Cloud and 1.25x on Azure, up to 4X-Large.
Multi-cluster scaling is approximate. The concurrency factor multiplies both settings’ running time by the mean number of clusters. Extra clusters really start and stop on the scaling policy, which a mean cannot reproduce.
Warehouse compute only. Storage, if you enter it, is the same under both settings and kept out of the saving. Cloud services above the 10% daily allowance, serverless features, Snowpipe and data transfer are not included.
Size changes assume runtime scales inversely with compute. That is the standard Snowflake rule of thumb and it holds for scan-bound work. It is optimistic for anything bottlenecked on a single-threaded step. Back-to-back work then finishes in a shorter window; spread-out queries keep their arrival times.

Rates used and their sourcesShow

Snowflake Service Consumption Table: credits per hour, list credit and storage prices, source, read 23 Oct 2026

Prices verified 23 Oct 2026. Public list prices in USD for AWS US East, before any discount you have negotiated.

Want this done on your real bill?

Cloud Cost Audit: We go through your real AWS or Google Cloud bill and hand you a ranked list of what to cut first. From $1,500, fixed. Nothing paid upfront.

See the service

About this tool

Enter the auto-suspend setting you have and the one you are considering: both are priced on the same queries, warehouse and credit price. The credit price starts at the Enterprise list price in AWS US East, and the credits per hour follow the warehouse generation and size in Snowflake's Service Consumption Table. Most customers pay a negotiated rate, so enter yours.

Questions

How is Snowflake compute billed?

By the second, for as long as a warehouse is running, not for query execution. A warehouse that resumes bills a 60-second minimum, then per second until it auto-suspends. This is why a warehouse serving 200 short queries an hour with a 10-minute auto-suspend bills a full hour for perhaps six minutes of work.

What should I set Snowflake auto-suspend to?

It depends on how your queries arrive, so there is no one right value. Snowflake accepts any whole number of seconds of 0 or more, and 0 means the warehouse never suspends. Each resume bills at least a minute, so going below 60 seconds rarely saves more. A shorter setting means more resumes, and a suspended warehouse loses its data cache, which can slow the next queries; that matters most for interactive BI. Enter your current setting and the one you are considering, and this calculator prices both on the same workload. Test a change on your own workload before making it in production.

How does the calculator compare two auto-suspend settings?

With one model and the same queries, runtimes, warehouse, credit price and days. A suspended warehouse resumes for the first query; it suspends once nothing has run for the auto-suspend time, and a query that arrives sooner joins the same session. Each session bills at least 60 seconds, then per second. Queries are assumed to arrive evenly through the hours you enter, so the result is an estimate for that pattern: real traffic clusters, and real suspensions land a little after the timeout.

Does a bigger Snowflake warehouse cost more?

Per hour, yes, each size up doubles credits per hour. Per query, often not, because doubling compute roughly halves runtime for scan-bound work, making the change close to cost-neutral while cutting latency. It stops being neutral when queries are bottlenecked on something that does not parallelize, or when the warehouse is idle most of the time anyway.

What is a Gen2 warehouse and does it cost more?

Gen2 is the newer standard warehouse, and Snowflake creates it by default where a region has it. It bills 1.35 times the standard credits an hour on AWS and Google Cloud and 1.25 times on Azure, so a Medium is 5.4 credits an hour on AWS rather than 4. Run SHOW WAREHOUSES: the resource_constraint column says STANDARD_GEN_1 or STANDARD_GEN_2.

What credit price does this calculator use?

It starts at $3, the Enterprise on-demand list price in AWS US East from Snowflake's Service Consumption Table. Standard is $2 and Business Critical $4 there, and other regions differ. Almost every customer of any size pays a negotiated rate, so enter yours.

What is multi-cluster warehouse scaling costing me?

Each additional cluster bills the full credits-per-hour rate for the warehouse size while it runs. A standard Medium warehouse scaling to three clusters costs 12 credits an hour, not 4. Multi-cluster warehouses need Enterprise edition or higher. The concurrency factor in this calculator is the mean number of clusters running, which is what you should be measuring rather than the maximum you configured.

More tools for: Analytics and warehouses

Want this asked continuously, against real usage?

Finitizer does this for BigQuery today. Snowflake and Redshift support is on the way.

See how Finitizer works

Snowflake Cost Calculator