The cloud platform that split compute from storage - and turned pay-per-second analytics into a category. Now it wants to run your AI on the same governed data.
In 2012, two data architects who had spent years inside Oracle asked a question that sounded almost too simple. What if you stopped renting a database and started renting a moment of computing power instead? What if storage and processing did not have to be bought together, sized together, or paid for together?
Benoit Dageville and Thierry Cruanes, joined by Vectorwise co-founder Marcin Zukowski, built the answer into a company they named Snowflake - a nod both to their love of skiing and to the "snowflake schema" used in databases. For about two years, they worked in near-silence. The product did not surface publicly until 2014, and general availability arrived in 2015.
The core idea was architectural. Traditional data warehouses bolted compute to storage, so scaling one meant scaling the other, and idle hardware still cost money. Snowflake separated the two layers. Storage sat in cheap cloud object stores; compute spun up as independent, elastic "virtual warehouses" that could be sized on demand and billed by the second. When the query finished, the meter stopped.
That decision did two quiet but important things. It let different teams run heavy workloads at the same time without fighting over the same machine, and it aligned the bill with actual use rather than with a fixed contract. Customers paid for the seconds they consumed. It is the kind of change that looks obvious in hindsight and was contrarian at the time.
Snowflake also refused to pick a side in the cloud wars. It runs on Amazon Web Services, Microsoft Azure, and Google Cloud - the same three giants that sell competing data products. Rather than force customers to move, Snowflake met them wherever their data already lived. That neutrality became a selling point in its own right.
At its simplest, Snowflake is a place to put all of an organization's data and then do useful things with it - warehousing, analytics, data lakes, engineering pipelines, and increasingly machine learning and AI. Because it is fully managed, customers do not patch servers, tune clusters, or plan capacity in the old way. They load data, write mostly familiar SQL, and let the platform handle the rest.
Its users are broad. Data engineers build pipelines. Analysts run reports. Data scientists train models. And business teams - in finance, retail, healthcare, media, and technology - increasingly ask questions in plain language. More than 12,000 organizations use the platform, including a large share of the world's biggest public companies. Names associated with Snowflake over the years include Capital One, Pizza Hut, AdTheorent, and Instacart.
The problems it removes are the unglamorous ones that quietly drain enterprise time: infrastructure that has to be provisioned in advance, data copied into a dozen silos, teams blocked because someone else is running a big job, and analytics bills that arrive whether or not anyone did any analysis. Snowflake's answer is elasticity plus governance - scale up when you need it, scale to nothing when you do not, and keep a single controlled copy of the truth.
There is one feature that rarely makes headlines but does much of the work: data sharing. Instead of exporting files and emailing spreadsheets, an organization can grant governed access to live data that never leaves its home. Partners query the same data in place. It is friction removed so quietly that many users never notice it is there.
The core platform: elastic, pay-per-use warehousing, data lakes, and analytics across AWS, Azure, and Google Cloud.
Discover, share, and monetize datasets and native apps - without copying or moving the underlying data.
Run Python, Java, and Scala plus machine-learning workloads natively inside Snowflake.
Fully managed AI service with serverless access to large language models, built directly on governed data.
An open-source, enterprise-grade LLM released to compete with models like Llama and DBRX.
Agentic AI products (formerly Snowflake Intelligence and Cortex Code) that query and build on governed data.
The data market is crowded - Databricks, Google BigQuery, Amazon Redshift, Microsoft Fabric, Oracle, Teradata. Snowflake's separation is less about any single feature and more about a set of consistent choices.
Snowflake sells consumption, not seats. Customers pay for the compute and storage they actually use, billed largely per second, so a team that runs more workloads pays more and a team that pauses pays less. Growth is measured by how much existing customers expand their usage over time - a land-and-expand pattern that ties the company's revenue directly to customer activity.
That model has drawbacks and virtues. Revenue can be lumpy quarter to quarter because it follows usage. But it also means the vendor rarely has to defend a bill for capacity no one touched, and it removes the incentive to over-sell shelfware. CEO Sridhar Ramaswamy, who took over in 2024 after co-founding Neeva and earlier building Google's ads business, has defended the approach publicly and pushed the company deeper into AI.
In the market, Snowflake sits at the center of the enterprise data-and-AI stack, competing most directly with Databricks for the title of default platform. Its neutrality across clouds, its data-sharing network, and its managed AI services are the pillars it leans on. Additional revenue comes from the Marketplace and from AI consumption as agentic tools spread.
The company's expertise is, at root, distributed systems and query performance - the hard engineering of making enormous datasets feel instant. That foundation, laid by founders who had built these systems before at Oracle and Vectorwise, is what everything else is stacked on. The AI layer is new; the discipline underneath it is not.
Dageville, Cruanes, and Zukowski start the company and work largely in stealth.
The cloud data warehouse goes public, then reaches general availability on AWS.
The former ServiceNow chief arrives to scale the company toward a public listing.
Snowflake lists on the NYSE in the largest software IPO ever, drawing a rare Berkshire Hathaway investment.
The Neeva co-founder and former Google Ads leader becomes CEO and accelerates the AI strategy.
The company leans fully into AI and acquires PostgreSQL provider Crunchy Data for about $250M.
At Summit 2026 Snowflake ships AI agent governance, rebrands CoWork and CoCo, and lands a $6B AWS commitment.
It provides a cloud-based data platform where organizations can store, process, analyze, and share data - and now build AI applications - without managing their own infrastructure. It runs on AWS, Azure, and Google Cloud.
It was founded in 2012 by Benoit Dageville, Thierry Cruanes, and Marcin Zukowski. Dageville and Cruanes were former Oracle data architects; Zukowski co-founded Vectorwise.
Through consumption-based pricing: customers pay for the compute and storage they actually use, billed largely per second, rather than fixed per-seat licenses. Revenue expands as customers run more workloads.
Its primary rivals include Databricks, Google BigQuery, Amazon Redshift, Microsoft Azure Synapse/Fabric, Oracle, and Teradata.
Cortex is Snowflake's fully managed AI service that gives serverless access to large language models, so enterprises can build AI applications and agents directly on their governed data inside Snowflake.
Figures are drawn from public disclosures and reporting and are approximate where noted. Company profile compiled by YesPress Newsroom.