Snowflake Serverless Costs: 2026 Rates for Every Feature

Snowflake serverless features run on compute that Snowflake manages for you, so they bill separately from your warehouses. Each feature has its own rate: Snowflake charges most per compute-hour at a set multiplier and a few per GB. Snowpipe costs 0.0037 credits per GB loaded, classic Automatic Clustering 2 credits per compute-hour, and serverless tasks 0.9. This guide lists the current rates, explains how they work, and shows how to find what serverless features cost you. We checked every fact and figure in this guide against Snowflake’s documentation.

This article is part of our Snowflake pricing guide.

What are Snowflake serverless features?

Serverless features are Snowflake services that run without a warehouse you manage. Snowflake starts, sizes and stops the compute itself, then bills you for what it used. They include data loading with Snowpipe, keeping clustered tables organized, refreshing materialized views, running serverless tasks, as well as replicating data to other regions.

Because they run in the background, serverless costs are easy to miss. They also stay out of warehouse metering, and resource monitors do not limit them.

How does Snowflake bill serverless features?

Snowflake bills most serverless features per compute-hour, times a multiplier for that feature. One compute-hour is roughly one hour of an X-Small warehouse. For example, a feature with a multiplier of 2 uses 2 credits for every compute-hour of work. Many features also have a second multiplier for the cloud services they use. A few features, such as Snowpipe and Optima Clustering, bill per GB of data instead.

You then pay for those credits at the same price per credit as your warehouses: $2.00 on Standard, $3.00 on Enterprise and $4.00 on Business Critical, on demand in AWS US East.

Snowflake serverless feature rates (2026)

These are the rates in Snowflake’s Service Consumption Table, Table 5, effective 15 October 2026.

Materialized views, search optimization and query acceleration need Enterprise Edition or higher.

What does Snowpipe cost?

Snowpipe costs 0.0037 credits per GB loaded, with no separate charge per file. Snowflake charges text files such as CSV and JSON on their uncompressed size. Binary formats such as Parquet, Avro and ORC, compressed or not, bill on their observed size.

For example, loading 2 TB of CSV a month (2,048 GB) uses 2,048 x 0.0037 = 7.6 credits, about $23 a month on Enterprise at list price. Our Snowpipe pricing guide shows how to measure it on your own account.

What does Automatic Clustering cost?

Snowflake now has two ways of billing clustering. Tables clustered before 1 September 2026 stay on classic Automatic Clustering, billed at 2 credits per compute-hour of reclustering work. On most accounts, however, tables that get clustering keys from 1 September 2026 use Optima Clustering, billed at 0.007 credits per uncompressed GB clustered, less where the data overlaps.

Clustering cost depends on how often a table changes. For example, a large table that changes all day can recluster all day. Our article on auto-clustering costs that get out of control shows how to find the tables driving them.

What do materialized views and search optimization cost?

Both bill at 2 credits per compute-hour, plus cloud services, for the background work that keeps them up to date. They also add storage: a materialized view stores its results, and search optimization stores a search access path. Both keep billing whenever the base table changes, even if nobody queries the view or uses the search path any more.

What do serverless tasks cost?

Serverless tasks bill at 0.9 credits per compute-hour, plus cloud services, and Flex tasks at 0.5. Snowflake measures compute per second and documents no minimum for serverless features. A task that runs on your own warehouse instead bills at that warehouse’s rate, with a 60-second minimum each time the warehouse resumes.

However, Snowflake does not say which is cheaper in general. It recommends serverless tasks where a warehouse would otherwise sit under-used, with few tasks running at once or tasks that finish quickly, and user-managed warehouses for tasks that keep a warehouse fully busy.

How do you see what serverless features cost you?

Start with METERING_DAILY_HISTORY, which shows credits billed by type of service. Every service type other than warehouse metering is something other than your warehouses: serverless features, AI and other Snowflake services.

-- Credits billed by service type, last 30 days, warehouses excluded
SELECT service_type,
       SUM(credits_billed) AS credits_billed
FROM snowflake.account_usage.metering_daily_history
WHERE usage_date >= DATEADD(day, -30, CURRENT_DATE())
  AND service_type NOT IN ('WAREHOUSE_METERING', 'WAREHOUSE_METERING_READER')
GROUP BY 1
ORDER BY 2 DESC;

Then drill into the feature that costs the most. Each has its own ACCOUNT_USAGE view:

All of these views are in SNOWFLAKE.ACCOUNT_USAGE and have a CREDITS_USED column.

For example, to find the tables that cost the most to cluster:

-- Clustering credits by table, last 30 days
SELECT database_name,
       schema_name,
       table_name,
       SUM(credits_used) AS credits_used
FROM snowflake.account_usage.automatic_clustering_history
WHERE start_time >= DATEADD(day, -30, CURRENT_TIMESTAMP())
GROUP BY 1, 2, 3
ORDER BY 4 DESC
LIMIT 20;

How do you control serverless costs?

• Use budgets, not resource monitors. Resource monitors only track and suspend warehouses. A Snowflake budget covers serverless features and sends a notification when spending passes a limit.
• Turn off features nobody uses. ALTER TABLE … SUSPEND RECLUSTER stops clustering charges on a table while it is suspended; on Optima Clustering, resuming later adds a catch-up cost, so drop the clustering key to stop for good. ALTER TABLE … DROP SEARCH OPTIMIZATION removes search optimization, and adding it back later costs as much as the first time. Also drop materialized views that no query reads.
• Batch Snowpipe loads sensibly. Snowpipe has no per-file charge, but very small files load more slowly.
• Check new defaults. Query acceleration is on by default for new Gen2 warehouses on Enterprise and above, and bills whenever a query uses it.

How Seemore helps with serverless costs

Seemore’s Intelligent Auto-Clustering reviews tables that already use clustering keys to decide whether they still pay off, flags tables where clustering no longer makes sense, and recommends exact clustering keys with a forecast of cost and performance before credits burn. Cost Anomaly Detection compares spend across your accounts, warehouses and services against the last 14 days of your own history and hands you the root cause when something drifts.