Snowflake Cost Optimization: How We Cut Client Bills 15–45%

I subscribe to Reddit and a few channels related to data engineering, Snowflake Computing, and others. There are many discussions about how Snowflake is expensive and that there are better solutions out there, like Databrick, Redshift, or even MS Synapse.

At Data-Sleek, Snowflake cost optimization is critical for our clients. We monitor how each account uses its resources and design the data architecture to keep costs low.

Snowflake Data Warehouse: How does it work?

Understanding the Cloud Data Warehouse model

Snowflake uses a cloud-based data warehouse model, leveraging hardware elasticity from Cloud providers like AWS, Google, and Azure and a pay-as-you-go pricing structure that scales storage and compute independently.

Ingestion is a cost line too. Our Postgres to Snowflake CDC review documents a native connector that burned about $60 a day syncing a handful of small tables.

Unlike traditional on-premise databases, Snowflake does not need to be provisioned with hardware or instance size. Because it separates compute from storage, you can scale each separately. You could need large storage and a few queries per month, which would cost you close to the same as AWS S3 storage.

Or you could have small storage needs but lots of queries, which would cost you if you don’t manage the compute unit properly. Or you could have both (large storage and lots of requests per month).

Snowflake offers elasticity and performance based on business needs. The architecture behind that flexibility is worth understanding before optimizing against it. We cover Snowflake’s data warehouse architecture in depth, including how it compares to other cloud platforms.

Because costs follow usage patterns, understanding this model is the first step to paying only for the resources you need. Transformation jobs drive much of that usage, and you can trim compute costs with dbt Cloud by tightening how and when those jobs run.

Implementing Snowflake’s unique architecture for cost optimization

Snowflake architecture allows for dynamic scaling and auto-suspend capabilities, which means you only pay for what you use. Businesses can optimize performance by strategically configuring virtual warehouses and leveraging workload isolation, including statement timeout, without incurring unnecessary expenses.

A virtual warehouse is also referred to as a “computing unit”, a computing size that allows you to handle requests (queries). The bigger the computing unit, the faster the query will run and the larger the dataset it can handle, but the more credits it will consume per hour.

When creating a new warehouse unit, the first thing to do is to make sure the auto-suspend is set to 60 seconds.

Snowflake utilizes per-second billing (with a 60-second minimum each time the warehouse starts), so warehouses are billed only for the credits they actually consume.

The total number of credits billed depends on how long the warehouse runs continuously. For comparison purposes, the following table shows the billing totals for three different sizes of warehouses based on their running time (totals rounded to the nearest 1000th of a credit):

Leveraging Snowflake’s features for automatic scaling and resource optimization

To maximize cost efficiency, organizations should take advantage of auto-scaling and multi-cluster warehouses.

Auto-scaling dynamically adjusts the number of compute nodes based on demand, reducing waste during low-traffic periods. This can be done directly in SQL. Let’s suppose you need to run multiple SQL queries: one to load a large amount of data, another one to join the data and create a stored proc, and another one to summarize the data. Within your SQL code, you could use :
“use warehouse <warehouse_name>; copy data …. then use warehouse
use <smaller_warehouse_name> ; to run the rest.

As you can see, the usage of larger or smaller warehouse units is dynamic and instant. (You do need to create these warehouse units in advance and grant proper permissions).

Review usage patterns regularly and adjust warehouse configurations as workloads change.

Snowflake also provides query caching, which helps optimize resource utilization by retrieving previously executed query results instead of re-running compute-intensive operations. If the same queries run on multiple tables, it will use the caching layer. This is useful for reporting (when plugging dashboards such as Tableau and Power BI) that are used by many users.

Is Snowflake cost-effective?

Snowflake is cost-effective when you actively manage compute, and expensive when you don’t. You pay per second for the warehouses you run, so your bill reflects how well you size, suspend, and separate warehouses by workload.

It is a good fit when workloads are bursty or seasonal, when storage is large but queried occasionally, and when teams need to isolate loading, transformation, and reporting on separate warehouses. It gets expensive when warehouses run oversized or never suspend, when BI tools keep them awake with constant refreshes, or when always-on pipelines run all day for small data volumes.

Right-sizing Virtual Warehouses for Efficiency

Choosing the right warehouse size is one of the biggest cost levers in Snowflake. Sizes run from X-Small to 6X-Large, and each step up doubles the credits a warehouse consumes per hour.

At Data-Sleek, we properly set up the different warehouse units for the different purposes they will serve. We usually create one warehouse unit for loading data, another one to transform the data, one for reporting, and one for the data engineer. This lets us quickly see where most of the credits are spent and act accordingly.

Employing Auto-Suspend and Auto-Resume Wisely

Auto-suspend stops a warehouse after a set period of inactivity, and auto-resume restarts it when the next query arrives. Snowflake bills per second with a 60-second minimum each time a warehouse starts, so the right setting depends on how each warehouse is used.

Set auto-suspend to 60 seconds for loading, transformation, and ad hoc warehouses. Their queries run in bursts, and an idle warehouse between jobs is pure waste.

Give BI and reporting warehouses a longer window, such as 5 to 10 minutes. Dashboards often send queries a few minutes apart, and a running warehouse keeps its local disk cache, so repeat queries read from SSD instead of remote storage.

Keep AUTO_RESUME on for every warehouse. Without it, queries sent to a suspended warehouse fail instead of waking it up.

ALTER WAREHOUSE reporting_wh SET AUTO_SUSPEND = 600 AUTO_RESUME = TRUE;

Snowflake costs: how the model works

Factors that contribute to Snowflake costs

Understanding the pricing model of Snowflake is crucial for successful Snowflake cost optimization. By knowing how costs are structured, you can make informed decisions that lead to better financial outcomes.

Snowflake pricing is influenced by three primary factors: storage pricing, compute, and data transfer.

Storage is inexpensive but grows over time if data is not managed correctly across various regions. Usually, it’s about the same as the cost of AWS S3 storage. Snowflake compresses data, so some savings could be made there, too.

Snowflake’s flexibility is powerful, but without governance and clear policies, costs can spiral quickly out of control.

Valli, Data Analyst, Data-Sleek

Compute costs, determined by the virtual warehouse size and execution duration, can be optimized through efficient query execution. The best is to start with the smallest unit and scale from there. It’s essential to also look at the data architecture. Can the data be “pre-processed” with Spark, for example, instead of having Snowflake do it? These are the types of questions to ask a data architect when implementing a data warehouse solution.

Data transfer fees are minimal but should be considered for cross-region data movement. Will you be exporting data out of Snowflake? How large will the dataset be? How frequently?

Understanding the pricing model of Snowflake

Snowflake pricing is influenced by three primary factors: storage, compute, and data transfer.

The pay-as-you-go structure of Snowflake’s pricing model allows businesses to effectively manage costs by only paying for the resources they consume. Additionally, the transparent nature of credit usage gives users visibility into their expenses and enables them to optimize their usage for cost-efficiency. Snowflake’s innovative pricing approach has made it a popular choice among organizations looking for a cloud data platform that aligns with their needs and budget requirements.

How to reduce Snowflake costs

Optimizing Storage Through Data Life-cycle Policies

To manage storage costs, organizations should implement data retention policies, using time-travel and fail-safe settings efficiently.

Archiving infrequently accessed data to cheaper storage tiers (e.g., AWS S3 Glacier) and setting retention limits can prevent excessive costs. Snowflake’s automatic data compression also reduces storage consumption without additional effort.

Data Retention Policies

Data retention policies are critical to controlling the growth and cost of data storage. Scripts and other mechanisms can automate this process. Also once the raw data has been processed, it is possible to archive it in Glacier for some cases (especially for real-time analytics).

Snowflake time-travel

Snowflake time travel offers key benefits, including data recovery, historical analysis, compliance, and auditing support.

Snowflake Time Travel enables accessing historical data (i.e., data that has been changed or deleted) at any point within a defined period. It serves as a powerful tool for performing the following tasks:

• Restoring data-related objects (tables, schemas, and databases) that might have been accidentally or intentionally deleted.
• Duplicating and backing up data from key points in the past.
• Analyzing data usage/manipulation over specified periods.

Leveraging Caching to Minimize Compute Time

Snowflake caches query results at multiple levels, reducing redundant computations. By leveraging query result caching, metadata caching, and local disk caching, organizations can dramatically decrease execution times and computing expenses. Optimizing SQL queries to reuse cached results further enhances efficiency.

Snowflake Query Result Caching

Benefit: If an identical query is run again within this timeframe (by the same user and role), Snowflake retrieves the result from the cache instead of re-computing the query. Such extended durations can be beneficial for optimizing query performance.

How to Optimize:

• Avoid unnecessary variations in query syntax (e.g., trailing spaces, capitalization differences).
• Use parameterized queries instead of dynamically constructing SQL queries that change frequently.
• Encourage users to reuse previous query results instead of executing similar queries multiple times.
• Standardize query text across dashboards so the result cache actually hits. Our Snowflake SQL examples show the patterns.

Snowflake Metadata Caching

Snowflake caches metadata and query execution plans to avoid repetitive table scans.

Benefit: Faster metadata retrieval, reducing unnecessary compute costs for operations like SHOW TABLES or DESCRIBE TABLE.

How to Optimize:

• Use INFORMATION_SCHEMA wisely: Avoid excessive metadata queries on large datasets.
• Use efficient query structures: Keep queries structured in a way that allows Snowflake to recognize and reuse metadata.

Snowflake Local Disk Caching

When a virtual warehouse (compute cluster) is running, Snowflake caches data on local SSD storage for fast access.

Benefit: Reduces the need for repeated data retrieval from cloud storage, speeding up queries and lowering compute costs.

How to Optimize:

• Keep warehouses warm when running frequent queries to maintain cache efficiency.
• Group related queries together to take advantage of a warm local cache before suspending a warehouse.
• Avoid stopping and restarting warehouses frequently, as this clears the cache.

Leverage Materialized Views for Performance Boost

Materialized views store precomputed query results, allowing Snowflake to reuse them instead of recomputing.

Benefit: Faster query performance and lower compute usage, especially for frequent aggregations and joins.

How to Optimize:

• Use materialized views for frequent and complex queries with costly joins.
• Be mindful of automatic refresh costs—optimize refresh schedules for cost efficiency.

Snowflake Clustering Keys

Using Clustering Keys helps Snowflake physically organize data with table clustering, reducing the number of scanned rows for queries.

Benefit: Minimizes unnecessary data scans, lowering compute costs.

How to Optimize:

• Choose clustering keys based on frequent filtering columns (DATE, CUSTOMER_ID, etc.).
• Avoid excessive clustering on small tables—it is more beneficial for large tables.

Caching Temporary Tables for Repetitive Calculations

Store aggregated or processed data in temporary tables or transient tables for batch jobs.

Benefit: Prevents re-running expensive computations in complex queries.

How to Optimize:

• Store aggregated or processed data in temporary tables or transient tables for batch jobs.
• Drop temporary tables when no longer needed to avoid unnecessary storage costs.

Snowflake Query Optimization to Maximize Cache Reuse

Writing queries efficiently increases the chances of hitting existing caches.

Benefit: Minimizes redundant compute work, reducing Snowflake credit consumption.

How to Optimize:

• Avoid SELECT *: Specify only the columns needed.
• Use filters and partitions: Querying smaller data subsets improves performance and reduces scanning.
• Pre-aggregate data where possible: Running computations in advance can limit repetitive, expensive operations.

As you can see, there are many ways to optimize Snowflake cache and thus reduce the usage of compute. We have helped many customers reduce their Snowflake cost by 15, 25, and even 45% by many adjustments.

Streamlining ETL Processes for Lower Costs

Extract, Transform, Load (ETL) operations can quickly become expensive if not optimized. When using Snowflake, it is recommended to use the ELT (Extract, Load, Transform) approach instead. Using this method allows us to load the data in the data warehouse and transform the data later,

Loading data in batches will reduce the cost. Some data architectures might suggest storing the data in S3 first, then loading a full day of data into the data warehouse for transformation. This way, Snowflake runs fewer queries, which then limits the cost.

At Data-Sleek, we usually store data in Snowflake directly or S3, then use DBT Labs to transform the data and build fact and dimensional tables. Dimensional modeling is one of the best techniques for storing data in a data warehouse for an efficient, easy-to-use, and scalable reporting solution. If you are considering Snowflake OpenFlow for ingestion, model its runtime cost before you migrate pipelines.

Key metrics to monitor Snowflake costs

Utilizing Snowflake’s Query History for Insights

Snowflake’s Query History provides a valuable resource for understanding usage patterns and optimizing costs. By analyzing query execution times, scanned bytes, and the query frequency of repetitive queries, organizations can identify inefficiencies and tune workloads for better performance. For example, frequent full-table scans or unnecessarily large result sets can lead to excessive compute costs, potentially reaching the maximum cost of a single query. Leveraging Query Profiling and Warehouse Load Monitoring, businesses can assess query behavior, pinpoint expensive operations, and implement optimizations such as clustering, partition pruning, or materialized views. Reviewing the query billing details also helps track the compute credits consumed by specific workloads, enabling cost control and more informed decision-making. Before tuning workloads, review this Snowflake SQL tutorial for practical query patterns and examples.

Setting Up Alerts for Budget Control

To prevent cost overruns, setting up budget alerts in Snowflake is essential. Snowflake’s Resource Monitors allow administrators to define thresholds for compute credits and receive notifications before exceeding budgeted limits. These monitors can enforce usage caps by suspending virtual warehouses when consumption surpasses predefined thresholds. Additionally, integrating Snowflake’s telemetry with external monitoring tools like AWS CloudWatch, Datadog, or custom dashboards can provide real-time insights into spending trends. Proactive alerting helps finance and engineering teams maintain cost efficiency, ensuring that resource allocation aligns with business priorities while preventing unexpected spikes in cloud expenditure. Tag every warehouse by team or workload so spend can be charged back. Object tagging is part of Snowflake Horizon governance.

Query performance and optimization

Optimizing query performance in Snowflake is crucial for both cost efficiency and data quality. Poorly optimized queries not only slow down data processing but also lead to excessive warehouse usage, increasing costs. Utilizing Query Execution Plans, businesses can identify bottlenecks such as excessive table scans, missing filters, or suboptimal joins. Techniques like clustering keys, pruning unused columns, and leveraging materialized views can significantly provide performance improvements. Additionally, auto-suspend and auto-resume settings for warehouses ensure that compute resources are only used when needed. Regularly monitoring query efficiency through Snowflake’s built-in Performance Dashboard helps maintain fast response times, reduce compute consumption, and improve system reliability.

Still paying too much for Snowflake?

At Data-Sleek, we’ve helped teams reduce Snowflake costs by 15%, 25%, and even 45% through smarter architecture, warehouse right-sizing, and workload optimization.

Book a Snowflake cost review: a fixed-scope look at your warehouses, credits and query patterns.

Editions, Budgets and When to Bring in Help

Your edition and your spending controls limit how much every other fix in this post can save. If you get them wrong, even right-sized warehouses run at an inflated credit rate, and nobody notices the bill until the invoice arrives.

Choose the Edition Your Compliance Requires

Snowflake charges every credit at a rate set by your edition, so an upgrade raises the cost of all your compute, not just the workloads that need the extra features. Business Critical adds support for HIPAA and PCI DSS data, customer-managed keys through Tri-Secret Secure, private connectivity, and account failover. Virtual Private Snowflake goes further, with an environment isolated from all other Snowflake accounts and a dedicated pool of compute.

Neither edition saves money by itself. Business Critical pays for itself only when it replaces infrastructure you would otherwise build and run separately, such as an isolated environment for regulated data or your own failover setup. If you don’t need either, Enterprise or Standard is the cheaper account.

Use Resource Monitors and Budgets Together

Resource monitors and budgets solve different problems, and most accounts need both. A resource monitor is the hard stop. It counts warehouse credits against a quota and can suspend the warehouses assigned to it when the quota runs out. Resource monitors cover warehouses only, so they can’t track serverless features or AI services.

Budgets cover that gap. A budget sets a monthly spending limit for the whole account or a custom group of objects, forecasts where the month will land, and sends alerts by email, cloud queue or webhook when spending is on track to go over. A budget doesn’t stop anything by itself, but it can call a stored procedure that suspends warehouses when a threshold is reached.

Decide Whether You Need a Third-Party Cost Tool

For most teams, Snowflake’s built-in cost data is enough. The Cost Management page in Snowsight and the ACCOUNT_USAGE views, such as WAREHOUSE_METERING_HISTORY, show credit usage by warehouse, and a few SQL queries feeding a BI dashboard give finance a monthly report at no extra cost. FinOps platforms such as CloudZero and Flexera’s Data Cloud Optimization (formerly Chaos Genius) add cost allocation by team, anomaly alerts, and rightsizing recommendations. They become worth paying for when Snowflake is one of several cloud bills you have to explain.

Setting all of this up is a one-time job, but reviewing it isn’t. Workloads grow, new warehouses appear, and last quarter’s limits stop fitting. If your team does not have time to run this review every quarter, an experienced Snowflake consulting partner can set up the monitors, tags, and budgets once and hand them over.

Conclusion

Snowflake is a powerful and scalable cloud data warehouse, but expenses can quickly spiral out of control without proper cost management.

Understanding its pay-as-you-go model, leveraging auto-scaling, auto-suspend, and query optimization strategies, and implementing effective data lifecycle policies are key to maintaining a cost-efficient Snowflake environment.

Applying the same discipline to ETL cuts expenses further, and those pipelines need regular review as data volumes grow.

Using the ELT approach enhances Snowflake cost optimization by minimizing the number of queries run, thereby reducing the costs associated with data transformations.

At Data-Sleek, we help businesses reduce Snowflake costs by 15%, 25%, or even 45% through smart architecture decisions, workload optimization, and continuous monitoring. Lower credit spend only holds if queries stay fast, so we tune both sides together. Our Snowflake performance tuning guide covers the query side, from warehouse sizing to clustering keys and the Query Profile. Organizations can maximize their return on investment while ensuring high performance by right-sizing virtual warehouses, optimizing storage, leveraging caching, and streamlining ETL processes.

The key to Snowflake cost optimization is not just about reducing expenses but also aligning cloud resources with business needs. With the proper monitoring tools, alerts, and proactive management, Snowflake can remain an efficient, scalable, and cost-effective solution for modern data warehousing. Want to see where your credits actually go? Book a free consultation.

FAQ: