Snowflake's pay-as-you-go model is a double-edged sword. The same elasticity that lets you spin up compute on demand also lets your bill climb quietly while nobody's watching. Most Snowflake overspend isn't caused by genuine workload growth - it's caused by a handful of avoidable habits. Here are the cost traps teams fall into most often, and the settings and practices that close them.

Warehouses left running

The single most common leak: a virtual warehouse running with no queries to serve. Snowflake bills compute per second while a warehouse is on, so an idle warehouse is money burning for nothing. The fix is built in - set auto-suspend to a short interval (often a minute or two) so a warehouse stops when idle, and auto-resume so it restarts on the next query. Many surprise bills are nothing more than a warehouse someone forgot to suspend.

Oversized warehouses

Bigger warehouses cost proportionally more per second, and teams routinely run a Large where a Small would do. Size is not a dial you set once and forget: match the warehouse to the workload, and remember that a faster, larger warehouse that finishes in a quarter of the time can sometimes cost the same or less - but only if the workload actually parallelizes. Test rather than assume, and right-size down when queries don't need the horsepower.

Scanning more data than you need

Snowflake reads less when your data and queries cooperate. Selecting only the columns you use, filtering on clustered or naturally ordered columns, and pruning partitions all cut the bytes scanned - which cuts both time and cost. A SELECT * on a wide table, or a query with no selective filter, makes Snowflake do work you're paying for and didn't need.

Unmonitored consumption

You can't control what you don't measure. Without resource monitors and visibility into which warehouses and queries drive spend, costs drift until the invoice arrives. Set resource monitors with credit quotas and alerts, and review the account usage views to see where credits actually go. The teams that keep Snowflake cheap are the ones who look at consumption weekly, not quarterly.

Inefficient data loading and storage

Loading many tiny files, holding excessive Time Travel and Fail-safe history on transient data, or keeping unused tables around all add cost at the edges. Load in reasonably sized batches, set retention to what each table actually needs, and clean up what you don't use. These are small individually and meaningful in aggregate.

The takeaway

Snowflake cost control isn't a special skill - it's a few defaults and a habit. Turn on aggressive auto-suspend, right-size warehouses to their workload, scan only the data you need, and put resource monitors and a weekly review in place so spend can't surprise you. The platform's flexibility is genuinely valuable; the trap is treating "elastic" as "free." Spend a little attention on the settings above and most avoidable Snowflake overspend simply disappears. For the architecture behind these levers, see the Snowflake overview.