Calculator
What does a warehouse that never suspends cost you?
Auto-suspend only saves money in the gaps that are longer than the suspend window. Set a warehouse to sleep after 10 minutes, query it every 4, and it never sleeps at all — while the settings page shows nothing wrong. This is the most common expensive misconfiguration I find in an audit, and it is invisible until you compare the setting against the traffic. No email required.
Paying for an idle warehouse
$5,280 /month
$63,360 a year · 1,760 credits · 100% of this warehouse's billed time
At 15 queries an hour, a query arrives every 4.0 min — sooner than the 10 min auto-suspend. The warehouse never gets a gap long enough to shut down, so you are billed as if it ran continuously. It looks correctly configured on the settings page.
What other settings would cost
| Auto-suspend | Idle cost/month | Saved vs now |
|---|---|---|
| 1 min | $1,320 | $3,960 |
| 2 min | $2,640 | $2,640 |
| 5 min | $5,280 | — |
| 10 min · current | $5,280 | — |
| 15 min | $5,280 | — |
| 60 min | $5,280 | — |
Don't guess the query rate — warehouse-teardown reads the real median gap straight out of your QUERY_HISTORY and tells you which warehouses never suspend. Read-only, open source, and it runs on fake data first so you can see the output before pointing it at your account.
This is one of the 18 questions in the Data Platform Teardown. The rest are published in full too.
Take the number with you.
I'll email this figure plus the free 18-check teardown — the first section is entirely about where warehouse money actually goes.
One email, sent immediately. Nothing follows it unless you reply.
Why this one hides so well
Every other cost problem announces itself. A huge query shows up in query history. A giant table shows up in storage. An idle warehouse shows up as a warehouse that is configured exactly the way the docs suggest, running a workload that looks entirely reasonable, on a bill that is quietly twice what it should be.
The tell is not in the settings. It is the relationship between two numbers that live in different places: the auto-suspend value on the warehouse, and the median gap between queries in QUERY_HISTORY. Nobody puts those side by side, which is why this survives for years.
The fix is usually a single ALTER WAREHOUSE and takes about ten seconds. Dropping auto-suspend to 60 seconds is safe for most analytical workloads — the resume cost is a second or two, and Snowflake keeps the result cache regardless.
Measure it, don't estimate it
The number on this page is only as good as the query rate you typed in. warehouse-teardown reads the real one from your account — median gap per warehouse, compared against each warehouse's own auto-suspend setting — and names the ones that never sleep. It is read-only, open source, and has a --demo mode so you can see the output before pointing it at anything real.
It answers five of the eighteen questions in the Data Platform Teardown. The other thirteen need a person — no tool is going to tell you whether a new engineer can run your pipeline locally on day one. If you would rather not run either yourself, that is what the audit is.