Skip to main content

Snowflake is a cloud platform where companies store and analyze data without managing servers.

Companies that need many teams to analyze the same data at the same time, without one team's heavy query slowing everyone else down.

Storage and compute scale and bill separately, so many teams can query the same stored data at once without contending for the same resources.

Decoupled storage and compute

Snowflake keeps two things separate, where your data is stored, and the processing power used to query it. You can add more processing power without touching storage, and vice versa, so different teams don't fight over the same resources.

A product team and a marketing team both want to run heavy queries on the same customer data at the same time. Instead of sharing one processing cluster and slowing each other down, each team gets its own, and both queries finish at normal speed.

The technical details

Snowflake calls each independent cluster of compute resources a virtual warehouse. Each virtual warehouse is an independent compute cluster that doesn't share compute resources with other virtual warehouses. One warehouse running a heavy job has no effect on the performance of another.

All warehouses read from the same central stored data, so nobody has to copy or move data to get their own compute.

Consumption-based billing via credits

You don't pay a flat subscription fee. You buy credits, and Snowflake deducts them per second based on how much compute and storage you use. How many credits an hour of compute costs depends on which Edition your account is on.

A team spins up a warehouse to run a report for ten minutes, then shuts it down. They're billed only for those ten minutes of compute, not for a full day or month of access.

The technical details

Per-credit rates vary by Edition. Standard costs $2.00, Enterprise costs $3.00, and Business Critical costs $4.00 (all list price, USD, AWS US East). Storage is billed separately, at $23.00 per TB per month on-demand.

Warehouses carry a 60-second minimum billing charge each time they activate. Snowflake describes its pricing calculator as "for estimation only, not for quoting." Costs depend on usage, region, cloud provider, and purchase model.

Editions as a security and compliance dial

Snowflake sells four Editions, Standard, Enterprise, Business Critical, and Virtual Private Snowflake. Moving up an Edition mostly buys stronger security and compliance certifications rather than more features.

A healthcare company that needs to store patient health data covered by HIPAA has to be on Business Critical or higher. A bank with the strictest isolation requirements might choose Virtual Private Snowflake instead.

The technical details

The Edition an account is on determines the unit costs for the credits and data storage used.

Business Critical, formerly called "Enterprise for Sensitive Data," adds support for HIPAA and HITRUST CSF-covered health data (requiring a signed business associate agreement). It also adds PCI DSS support, FedRAMP/ITAR support for U.S. government workloads, customer-managed encryption keys, and private connectivity.

Virtual Private Snowflake includes everything Business Critical does, but runs in a separate environment that shares no hardware with any other Snowflake account.

The AI Data Cloud and Snowflake Cortex

Snowflake's current pitch is that AI runs directly against the data stored in it. You don't have to move data somewhere else to use it with an AI model. Cortex is the name for Snowflake's set of AI features.

An analyst types a plain-language question about sales trends into Cortex Analyst. It converts that question into a SQL query and runs it against the company's own stored data, without the data ever leaving Snowflake.

The technical details

Cortex AI lets teams "access industry-leading LLMs at scale directly in SQL or via APIs... all within Snowflake's secure perimeter." This includes direct access to Anthropic Claude, Meta Llama, and Mistral Large 2. Named Cortex components include Cortex Analyst (natural language to SQL) and Cortex Agents, which orchestrate across structured and unstructured data.

Snowflake has also launched two newer, broader agent products. Snowflake CoWork is "the personal work agent for every knowledge worker." Snowflake CoCo, formerly Cortex Code, is a "data-native AI coding agent."

Problems it solves

Teams competing for the same compute, and slowing each other down

Before platforms like Snowflake, running several teams' queries on one shared cluster meant one team's heavy job could slow down everyone else's. Snowflake's decoupled architecture gives each team its own virtual warehouse. Multiple groups can query the same underlying data at once without contending for the same processing power.

Wanting to run AI against company data without moving it elsewhere

Running an AI model against a company's data traditionally meant exporting it to a separate AI platform first. That added cost, delay, and a second copy of sensitive data to secure. Snowflake Cortex runs large language models directly against data that stays inside Snowflake's own environment.

Meeting different compliance requirements without switching platforms

Organizations in regulated industries such as healthcare, finance, and government need progressively stricter guarantees, like HIPAA, PCI DSS, or FedRAMP support. Snowflake's four Editions let an organization step up to those guarantees within the same platform, rather than migrating to a different vendor as requirements grow.

Paying for idle infrastructure instead of what's used

Traditional data warehouses often required buying and running fixed hardware capacity whether or not it was in use. Snowflake's consumption-based credit billing charges per second of compute and storage used, so a warehouse that's shut down when idle stops costing anything.

Competitors

Databricks, Amazon Redshift, Microsoft Fabric, Salesforce Data Cloud, Apache Hadoop

Pricing

Consumption-based (credits), by edition. From $2.00/credit (Standard)

Business model

Public company. IPO Sept 2020, raised $3.4 billion

Links

Find it
Homepage, Docs

Going deeper

How is Snowflake's architecture built?

Snowflake describes its architecture as "a hybrid of traditional shared-disk and shared-nothing database architectures," and names the specific layers that make that hybrid work.

Why a hybrid design

A shared-disk architecture gives every compute node access to the same central data, which keeps data management simple. But it can bottleneck as more nodes are added.

A shared-nothing architecture spreads data across nodes that each store their own portion locally. This scales well but is more complex to manage.

Snowflake's hybrid design combines a central data repository, reachable by all compute nodes, with massively parallel processing (MPP) clusters. Each MPP cluster works on its own local portion of the data. Snowflake describes the result as combining the simplicity of shared-disk with the scale-out performance of shared-nothing.

The four layers Snowflake names

  • Database storage. Data is reorganized into a compressed, columnar internal format and split into micro-partitions, contiguous chunks of storage that Snowflake manages automatically. A third-party technical breakdown (AltexSoft) puts typical micro-partition size at roughly 50-500MB before compression.
  • Compute (virtual warehouses). Independent compute clusters. Each virtual warehouse has no effect on the performance of other virtual warehouses.
  • Cloud services. Coordinates everything from sign-in to query dispatch, and is where security, metadata, and query optimization live.
  • Snowgrid. Snowflake's cross-region, cross-cloud technology layer. Snowflake describes it as delivering "a single, connected experience across regions and clouds globally." It enables cross-cloud features like the Snowflake Marketplace and Secure Data Sharing.

What this makes possible

Because storage and compute are physically separate but logically connected, each layer scales independently.

Snowflake supports several table types on top of this base. These include its own optimized Snowflake tables and Apache Iceberg tables for data kept in external cloud storage. Hybrid tables are built for transactional workloads alongside analytical ones.

What does Snowflake cost?

Snowflake publishes per-credit and per-storage figures, but several caveats separate a headline number from a bill.

The four tiers

  • Standard Edition, $2.00 per credit
  • Enterprise Edition (Snowflake's own "Most Popular" label), $3.00 per credit
  • Business Critical Edition, $4.00 per credit
  • Virtual Private Snowflake, no listed price, "Talk to Sales"

On-demand storage is billed separately, at $23.00 per TB per month. All figures are USD, shown by default for one specific platform and region combination, AWS in US East (Northern Virginia).

Caveats worth reading before assuming this is your bill

These are list prices, not necessarily what a given customer pays. Rates vary by region and cloud provider.

Snowflake names two purchase models. On Demand is described as "usage-based pricing with no long-term licensing requirements." Capacity is a prepaid commitment that carries a discount off the On-Demand rate. A downloadable Credit Consumption Table PDF covers the full region-by-region and discount schedule.

Because billing is per-credit rather than a flat subscription, monthly spend depends entirely on usage. Snowflake's pricing calculator is "for estimation only, not for quoting," and "your billing statement will reflect your actual usage."

AltexSoft, a third-party technical review, treats this as a real risk rather than a footnote. Its warning is blunt. "One sloppy script, and before you know it, the entire year's budget can vanish within days," since on-demand pricing has no spending cap by default.

What happened in the 2024 Snowflake security incident?

In 2024, a hacking group called ShinyHunters ran a mass data theft and extortion campaign that affected hundreds of Snowflake customers. The investigation into it drew a specific, narrower conclusion about where the fault sat.

What was affected

Companies whose data was exposed included Ticketmaster, Advance Auto Parts, Santander Bank, Neiman Marcus, LendingTree, AT&T, Pure Storage, and Bausch Health. All were customers running on Snowflake at the time.

What the investigation found, and didn't find

In May 2024, Snowflake worked with the incident response firm Mandiant on an investigation. That investigation found no evidence that Snowflake's own environment had been breached.

Instead, it found that compromised customer credentials had been stolen through separate infostealing malware. Attackers used those credentials to log into customer accounts as if they were a legitimate user.

Two men connected to the ShinyHunters group were later indicted on charges including conspiracy, computer fraud and abuse, extortion, and aggravated identity theft. One was arrested on October 30, 2024.

The distinction matters for anyone evaluating Snowflake's security. The platform's infrastructure was not shown to have a vulnerability that let attackers in. The exposure came from how individual customer accounts were, or weren't, protected on the customer side.

What does each Edition unlock, in practice?

Snowflake's feature-and-edition matrix shows that most day-to-day SQL, data loading, and AI features are identical across all four Editions. The differences cluster around security, governance, and business continuity.

What every Edition already includes

Standard Edition already includes full SQL support, semi-structured and unstructured data support, and virtual warehouses. It also includes Snowpipe, Cortex AI Functions, Cortex Analyst, and SOC 2 Type II certification. Multi-factor authentication, encryption of all data, and object-level access control are also included at every tier.

What Enterprise adds

Enterprise Edition adds multi-cluster virtual warehouses, letting a single warehouse scale out automatically under heavy concurrent load. It also adds extended Time Travel of up to 90 days instead of Standard's 1 day, plus column-level and row-level security.

Enterprise Edition adds the ability to create and manage Data Clean Rooms rather than just use them.

What Business Critical and Virtual Private Snowflake add

Business Critical adds customer-managed encryption keys (Tri-Secret Secure) and private connectivity to the Snowflake service itself. It also adds support for PHI data under HIPAA and HITRUST CSF, PCI DSS support, and FedRAMP/ITAR support for U.S. government workloads.

It adds account failover and failback for disaster recovery too.

Virtual Private Snowflake includes everything Business Critical does, plus a dedicated metadata store and pool of compute resources. It adds full encryption of internal data transmissions within the VPS environment. A VPS account shares no hardware of any kind with accounts outside the VPS.

One tradeoff is that VPS accounts lose access to the public Snowflake Marketplace and public Listings, which remain available on the three lower Editions.

Sources

Last checked August 2026