At a glance
| SQL support | Snowflake SQL (ANSI-based), with QUALIFY, VARIANT path syntax and stored procedures |
| Storage and compute architecture | Managed columnar storage separate from virtual warehouse compute, plus a cloud services layer |
| Workload types | SQL analytics and BI, ELT, data sharing, Python data engineering and ML (Snowpark), AI functions |
| Integrations | Iceberg catalogs (Snowflake, AWS Glue, Open Catalog), Fabric OneLake interoperability, JDBC and ODBC clients |
| Deployment | SaaS on AWS, Azure or Google Cloud; no on-premises option |
| Pricing model | Credits for compute (price by edition, cloud, region), storage per TB per month, data transfer |
Snowflake suits teams whose main work is SQL analytics and BI and who want a warehouse with almost nothing to administer, on whichever of the three big clouds they use. Costs follow how long warehouses run and at what size, so auto-suspend and right-sizing matter more than how much data a query reads.
Overview
Snowflake is a data platform that runs only on public cloud infrastructure; it cannot be installed on premises. Its documentation describes three layers: compressed columnar storage organised into micro-partitions, compute in virtual warehouses that you create, size, suspend and resume, and a cloud services layer for metadata, security and query optimisation. Each account lives on one cloud and region, and an organisation can hold accounts on several clouds.
It is sold in four editions: Standard, Enterprise, Business Critical and Virtual Private Snowflake. Beyond SQL warehousing, Snowflake now offers Apache Iceberg tables, Snowpark for Python, Java and Scala code that runs inside Snowflake, the Cortex AI features and, since its 2025 acquisition of Crunchy Data, Snowflake Postgres for operational workloads.
Key features
- Virtual warehouses billed per second (60-second minimum per start) with auto-suspend and auto-resume
- Multi-cluster warehouses that add clusters when queries queue (Enterprise edition and above)
- Apache Iceberg tables with Snowflake or an external catalog, stored in Snowflake or your own cloud storage
- Semi-structured data in VARIANT, OBJECT and ARRAY columns, expanded with FLATTEN
- Secure Data Sharing and Snowflake Marketplace listings without copying data
- Snowpark DataFrames and UDFs in Python, Java and Scala, plus Snowpark-optimized warehouses
- Cortex AI Functions callable from SQL, Cortex Analyst, Cortex Search and Cortex Agents
Pros and cons
Pros
- Very little to administer: no distribution keys, indexes or files to manage for native tables
- Same service on AWS, Azure and Google Cloud, with a published per-region credit price list
- Separate warehouses per workload, so loads and dashboards do not compete for compute
- Iceberg tables reduce lock-in by keeping data in an open format in your own storage
Cons
- Masking and row access policies, multi-cluster warehouses and long Time Travel need Enterprise edition or higher
- Warehouses left running keep consuming credits, so suspend settings need care
- Third-party engines cannot write to Iceberg tables that use Snowflake as the catalog
- Credit use varies by warehouse type and generation, so estimates need the consumption table
Pricing
Snowflake bills compute in credits, storage per TB per month and data transfer out of a region or cloud. The price of a credit depends on edition, cloud and region. Snowflake's Service Consumption Table (effective 4 November 2026, USD) lists On Demand credits in AWS US East (Northern Virginia) at USD 2.00 for Standard, USD 3.00 for Enterprise, USD 4.00 for Business Critical and USD 6.00 for Virtual Private Snowflake, with On Demand standard storage there at USD 23.00 per TB per month; EU and Asia Pacific regions cost more (for example USD 2.60 per Standard credit in AWS EU Frankfurt). A first-generation X-Small warehouse uses 1 credit per hour, and each size up doubles that; Gen2 warehouses use 1.35 credits per hour at X-Small on AWS and Google Cloud and 1.25 on Azure. Capacity contracts are prepaid with negotiated discounts. A 30-day free trial is available.
Pricing checked on the vendor's official page on 7 November 2026. Prices change; confirm before buying.
Who it is best for
SQL-first analytics teams that want a managed warehouse on any of the three major clouds, with per-workload compute and minimal tuning.
Who should look elsewhere
- Teams centred on Spark data engineering and training their own ML models, where a lakehouse may fit better: see Snowflake vs Databricks.
- Small datasets that fit on one server: see DuckDB vs Snowflake and Snowflake vs PostgreSQL.
Verdict
Snowflake is a strong default for managed SQL warehousing when you want the same platform across clouds and little to operate. Its overlap with lakehouse platforms is now large, so compare it with Databricks or your cloud's native warehouse on a realistic workload, and model credits by edition and region before you commit.
Frequently asked questions
How much does Snowflake cost?
Compute is billed in credits. In AWS US East (Northern Virginia), Snowflake's Service Consumption Table effective 4 November 2026 lists On Demand credits at USD 2.00 (Standard), USD 3.00 (Enterprise) and USD 4.00 (Business Critical), and storage at USD 23.00 per TB per month. Other regions cost more, and an X-Small first-generation warehouse uses 1 credit per hour.
Can Snowflake run on premises?
No. Snowflake's documentation states that it runs only on public cloud infrastructure: AWS, Microsoft Azure or Google Cloud.
Does Snowflake support open table formats?
Yes, through Apache Iceberg tables, which can use Snowflake as the catalog (full read and write) or an external catalog such as AWS Glue. Data can sit in your own S3, Google Cloud Storage or Azure storage through an external volume.
Sources
- Snowflake: Key concepts and architecture
- Snowflake: Overview of editions
- Snowflake: Virtual warehouses overview
- Snowflake pricing options
- Snowflake Service Consumption Table (effective 4 November 2026)
- Snowflake: Apache Iceberg tables
Checked November 2026.
How we research reviews and what star ratings mean: our editorial method.