Snowflake made “separate storage from compute” the default expectation for a warehouse, turned SQL people into a procurement force, and built the largest ecosystem in analytics along the way. It’s the platform every other warehouse compares itself against. It’s also the bill every data leader has had to explain to a CFO at least once, which is the other half of this review.
My position
Technically, Snowflake needs no defending: it’s an excellent warehouse on all three clouds, and if you need the same platform on AWS, Azure and Google Cloud there’s really one answer. The evaluation is financial and organisational. Snowflake rewards companies that assign an owner to cost from day one, and quietly punishes the ones that don’t.
How the money works, without the fog
Two meters. Storage is billed monthly on the compressed average, and it’s cheap. Compute is billed in credits, consumed per second while a virtual warehouse runs, and it’s where the money goes.
Here’s the part I verified on their pricing page so you don’t have to: there are no dollar figures on it. Four editions are listed (Standard, Enterprise, Business Critical, and Virtual Private Snowflake for the truly regulated), and the per-credit rate depends on edition, cloud and region. You get a number from the calculator or from sales. That’s a deliberate design, and it means every public “Snowflake costs X” claim you read, including in vendor comparisons, is somebody’s contract, not yours.
The good news is that the cost levers are real and yours to pull. Warehouses suspend when idle if you configure them to. Sizing down is one click. The teams that get burned are the ones where nobody owns those settings, a BI tool hammers an oversized warehouse all day, and the invoice becomes a quarterly surprise. Give one person the monthly number and the authority to change auto-suspend settings, and Snowflake behaves.
The credit ladder, from the docs
The dollar rate is contractual, but the consumption side is published and worth internalising, because it’s the half of the bill you control. A warehouse burns credits per hour by size, doubling at every step: X-Small is 1 credit an hour, Small 2, Medium 4, Large 8, X-Large 16, and on up to 6X-Large at 512 (these are the Gen1 numbers; the newer Gen2 warehouses run on faster hardware, meter at their own rates in the official consumption table, top out at 4X-Large, and in supported regions are now what you get by default, so check which generation you’re actually creating before trusting any burn estimate).
Now the arithmetic that makes the playbook concrete. An X-Small running four hours a day is about 120 credits a month. A Medium kept awake eight hours a day by dashboard refreshes is about 960. Multiply each by your contracted rate and you can price a decision before making it, which is precisely the discipline the platform rewards. Notice too what the doubling ladder implies: every size up is twice the burn, so “one size smaller than feels comfortable” from the playbook below is a 50% cost lever, and Snowsight’s default of X-Large for new warehouses is a default worth changing.
The cost playbook, specifically
Every unhappy Snowflake story I’ve read traces back to the same four settings, so here they are as a checklist. Auto-suspend on every warehouse, set to a minute or two, not the lazy default of ten. Separate warehouses per workload, so the BI tool’s chatter doesn’t keep the ETL warehouse awake. Start warehouses one size smaller than feels comfortable, because doubling is one click and most queries won’t notice. And a resource monitor with a hard monthly cap while you’re learning your usage curve, because the first quarter is where the surprises live.
None of this is advanced. It’s just work that has an owner or doesn’t, and the invoice reports which.
Editions, and which one you actually need
Standard covers most companies for longer than sales will suggest. Enterprise is the realistic step up when you want multi-cluster warehouses for concurrency or the longer time-travel window. Business Critical is a compliance decision rather than a performance one, and Virtual Private Snowflake is a full isolated environment for the handful of institutions that genuinely need it. My advice is to enter on Standard and let a measured constraint, not a feature list, justify the upgrade, since the per-credit rate steps up with each edition.
What’s genuinely hard to replicate
The ecosystem. Every ingestion tool, every BI tool, every consultant and every job applicant knows Snowflake. Data sharing between Snowflake accounts is still the smoothest in the industry. Iceberg table support means your data can live in open formats, which softens the lock-in that used to be my biggest reservation. And the higher editions carry governance features (longer time travel, multi-cluster warehouses, the compliance stack) that enterprises actually use rather than just tick.
The chatter worth knowing about
Review scores are the strongest in our directory’s warehouse category: 4.6 from 762 reviews on G2, with 75% five-star and effectively nothing below three, and 4.7 from 459 on Gartner Peer Insights. Practitioners like this product; the invoices they write about are a separate relationship.
Hacker News, as usual, is spikier. The two biggest Snowflake threads of the past year were security stories (an AI coding tool compromising Snowflake’s own Jira, and a sandbox-escape write-up), and a third thread titled “Picking the Lock-In You Want” about its Postgres and Lakebase moves. That title is the debate in miniature: the quality is settled, and the argument is about the price of being inside it. Worth reading before a multi-year commit.
Snowflake or BigQuery or Databricks?
The honest short version. BigQuery is simpler to reason about for small teams, has a real free tier, and wins by default if you’re on Google Cloud. Databricks wins when the same team also does heavy engineering and machine learning in Python. Snowflake wins on multi-cloud, on governed SQL analytics at scale, and on being the platform your next hire already knows. If your team is under ten people and your cloud is already chosen, I’d probably not start here; if you’re consolidating a company onto one analytics platform, this is the shortlist favourite for a reason.
Getting in without regret
The trial is where this is won. Load a real slice of your data, not the tutorial dataset, and replay a genuine week of queries: your BI tool’s dashboard refreshes, your heaviest join, your worst analyst habit. Watch the credit burn per warehouse in the account usage views, because that number times your negotiated rate is your future invoice, and it’s knowable before you sign anything.
When you do talk to sales, the two things worth negotiating beyond the rate are rollover terms on committed capacity (unused credits expiring is the quiet cost of overcommitting) and the renewal escalator. Consumption platforms price year one to win you; year two is where discipline pays.
Verdict
A superb product wrapped in pricing you have to manage like a live system. Take the trial, run your actual workload, set auto-suspend aggressively, and appoint a cost owner before the first contract, not after the first invoice. Do that, and most of the horror stories don’t apply to you.
