Cloud Parts

Snowflake

A fully managed cloud data warehouse that separates storage from compute, with virtual warehouses billed per second.

What it is
Data warehouses
Used for
Lakehouses
Type
Commercial product
Made by
Snowflake Inc.
License
Proprietary
Facts checked
5 October 2026
Website

What it is

Snowflake stores tables in its own compressed, columnar format, split into micro-partitions in cloud object storage. Compute comes from virtual warehouses: clusters you size from X-Small (1 credit per hour) to 6X-Large (512 credits per hour), each size double the one before. Warehouses don’t share compute, so a loading job never slows a dashboard. They suspend when idle and resume on the next query, billed per second with a 60-second minimum each time one starts.

A cloud services layer handles metadata, query parsing and optimisation, and security, and Snowflake runs on AWS, Azure and Google Cloud. It also manages Apache Iceberg tables: data and metadata sit in your own bucket (an external volume), Snowflake either acts as the catalog or connects to an external Iceberg REST catalog, and it supports Iceberg spec versions 1, 2 and 3. Snowflake Open Catalog is a managed service for Apache Polaris, an open-source Iceberg REST catalog.

Why teams pick it

  • Very little to tune: no physical design, storage grooming or nodes to manage
  • A separate warehouse per workload, each sized, scaled and suspended on its own
  • Zero-copy cloning and data sharing between accounts
  • Iceberg tables let other engines that speak the Iceberg REST protocol work on the same data without exports

Watch out for

  • Proprietary and billed in credits; costs grow with warehouse size and running time
  • Snowflake manages how native tables are stored, in its own internal format; for files other engines can read, use Iceberg tables in your own storage
  • Each warehouse start is billed for at least 60 seconds, so frequent suspends and resumes for short queries add up
  • Standard tables are built for analytics; fast single-row reads and writes need hybrid tables, a separate row-based table type

Alternatives

What all data warehouses have in common →
ClickHouseOpen sourceAn open-source column-oriented database for real-time analytics, fast enough to serve dashboards over billions of rows.DatabendOpen sourceA cloud data warehouse written in Rust that keeps every table as Parquet blocks and Git-like snapshots in object storage, so stateless compute clusters can start, scale and stop without moving data. Open core: Apache 2.0, with paid features under the Elastic License 2.0.Google BigQueryGoogle CloudGoogle Cloud’s serverless data warehouse: no clusters, and you pay per TiB scanned or per slot-hour.MotherDuckCommercialA serverless cloud data warehouse built on DuckDB, where one query can run partly on your laptop and partly in the cloud.

How it fits with other technologies

What uses Snowflake

A Snowflake-managed Iceberg table in your own S3 bucket, and a small warehouse · sql

CREATE EXTERNAL VOLUME lake_vol
  STORAGE_LOCATIONS = ((
    NAME = 'acme-s3'
    STORAGE_PROVIDER = 'S3'
    STORAGE_BASE_URL = 's3://acme-lake/iceberg/'
    STORAGE_AWS_ROLE_ARN = 'arn:aws:iam::123456789012:role/snowflake-lake'));

CREATE ICEBERG TABLE orders (order_id NUMBER(19,0), amount NUMBER(10,2), ts TIMESTAMP_NTZ(6))
  CATALOG = 'SNOWFLAKE'
  EXTERNAL_VOLUME = 'lake_vol'
  BASE_LOCATION = 'orders/';

CREATE WAREHOUSE reporting WAREHOUSE_SIZE = 'XSMALL' AUTO_SUSPEND = 60 AUTO_RESUME = TRUE;

See it in a real stack

No architecture uses Snowflake yet. Drop it into the playground and wire it up yourself.Open playground →

Benchmarks

Benchmark · ClickHouse Inc.ClickBenchHow fast analytical databases run 43 SQL queries over one 100-million-row table of web analytics events, on the same cloud machines.Includes Snowflake · 7 caveats to read first
How to read a benchmark →
Where it came from2012–2025

By the early 2010s, traditional data warehouses ran on small, static clusters of machines. They depended on deep ETL pipelines and careful physical tuning, and they struggled with the schema-less JSON arriving from logs, web and mobile apps. Hadoop and Spark could process anything, but lacked much of a warehouse’s efficiency and features and took significant engineering effort to roll out. Meanwhile, public clouds offered almost unlimited compute and storage on demand, which warehouses built for fixed resources couldn’t use.

In July 2012, Benoît Dageville and Thierry Cruanes, database architects at Oracle, and Marcin Żukowski, a co-founder of the Vectorwise analytical database, founded Snowflake Computing, with Mike Speiser of Sutter Hill Ventures as its first CEO. Implementation began in late 2012. Rather than build on Hadoop or PostgreSQL, they wrote the engine from scratch for the cloud. Snowflake unveiled its service in October 2014 and became generally available on AWS in June 2015.

Snowflake made the warehouse a pure service with no tuning knobs and no hardware to buy, with storage and compute scaling separately. At its September 2020 IPO it sold 28 million shares at $120 each, raising about $3.4 billion.

Why it’s built this way

Multi-cluster, shared-data architecture
In shared-nothing warehouses each node owns a slice of the data, so resizing or losing a node means reshuffling data, and loading and querying compete for the same hardware. Snowflake keeps one copy of the data and runs separate compute clusters (virtual warehouses) against it.
Store tables in S3 instead of building a storage layer
The team experimented and found S3’s availability and durability hard to beat, even with variable performance. They put their effort into local caching in the compute layer instead of writing their own HDFS-like service.
A pure service with no knobs
Users don’t buy machines, hire DBAs or install software, and there is no physical design or storage grooming to do. That was meant to bring a warehouse to teams that couldn’t afford the cost and complexity of traditional ones.
Semi-structured data as a first-class type
More and more data came from logs, web and mobile apps as JSON and similar formats. Built-in SQL extensions let users load and query it without first designing a schema and ETL pipeline.

Timeline

  1. 2012

    Founded in July by Dageville, Cruanes and Żukowski; work begins

  2. 2014

    Unveils the Snowflake Elastic Data Warehouse in October

  3. 2015

    Generally available on AWS in June

  4. 2016

    SIGMOD paper describes the multi-cluster, shared-data design

  5. 2018

    Preview on Microsoft Azure in July

  6. 2019

    Preview on Google Cloud in November

  7. 2020

    IPO in September raises about $3.4 billion

  8. 2025

    Acquires Crunchy Data to add PostgreSQL to the platform

Sources11 · facts checked 5 October 2026

Improve this page on GitHub