Snowflake Inc.
| Snowflake Inc. | |
|---|---|
| Type | Public |
| Traded as | NYSE: SNOW |
| Industry | Cloud computing Data warehousing Enterprise software |
| Founded | July 2012 |
| Founder | Benoît Dageville Thierry Cruanes Marcin Żukowski |
| Headquarters | Bozeman, Montana, United States |
| Key people | Sridhar Ramaswamy (Chief executive officer) Frank Slootman (Former chairman and chief executive) |
| Products | Snowflake Data Cloud, Snowpark, Cortex, Snowflake Intelligence |
| Website | snowflake.com |
Snowflake Inc. is an American company providing a cloud-based platform for storing, processing and analysing data. It is among the most significant of the data infrastructure companies that emerged in the 2010s, and its September 2020 listing was the largest software initial public offering ever conducted.
The company was founded in July 2012 by Benoît Dageville and Thierry Cruanes, who had spent many years as database architects at Oracle, together with Marcin Żukowski, a co-founder of the Dutch database company Vectorwise. It operated without publicity for two years, emerging in 2014.
Its product is a data warehouse built specifically for cloud infrastructure rather than adapted to it, and its defining architectural decision, separating the storage of data from the computation performed on it, is the reason it displaced established competitors.
Snowflake runs on Amazon Web Services, on Microsoft Azure since 2014 and on Google Cloud since 2019, and since 2026 on StackIT, and it is used by several thousand organisations.
History
Founding and the architectural idea
Benoît Dageville and Thierry Cruanes had worked on Oracle's database for more than a decade, on query optimisation and parallel execution, and had concluded that the architecture of conventional data warehouses was fundamentally unsuited to cloud infrastructure.
A traditional data warehouse, whether from Oracle, IBM, Teradata or Microsoft, is a cluster of machines in which each holds a portion of the data and performs computation on it. This couples storage and compute: adding capacity to store more data also adds processing that may not be needed, adding processing requires redistributing the data, and the cluster must be sized for the busiest moment and paid for at all times. It was the right design when the machines were physical and had to be bought in advance.
Cloud infrastructure removes those constraints. Object storage, such as Amazon S3, is effectively unlimited, cheap and durable, and compute can be started and stopped in seconds.
Snowflake's architecture therefore separates them. Data is held once in cloud object storage in a proprietary columnar format. Computation is performed by clusters called virtual warehouses, which are started when a query runs and stopped when it finishes, and several may operate on the same data simultaneously without contending with each other. A third layer, the cloud services layer, handles metadata, transactions, security and optimisation.
The practical consequences are what sold it. A department running a heavy analysis at month end can start a large cluster for an hour and pay for an hour. The data science team and the finance team do not slow each other down, because they use separate compute against the same data. Storage costs fall to the price of object storage. And there is nothing to administer: no indexes to tune, no partitions to design, no capacity to plan, which removed an entire category of specialist labour that conventional warehouses required.
Marcin Żukowski contributed the vectorised execution engine work from Vectorwise and from academic research at the Centrum Wiskunde en Informatica in Amsterdam.
The company was founded in July 2012, operated in stealth, and made its product generally available in 2015 under Bob Muglia, formerly a senior Microsoft executive, who was chief executive from 2014 to 2019.
Slootman and the listing, 2019 to 2024
Frank Slootman, who had previously led Data Domain to a sale and ServiceNow to a listing and substantial scale, became chief executive in May 2019, brought in to build a sales organisation capable of selling to the largest enterprises and to take the company public.
Snowflake listed on the New York Stock Exchange on 16 September 2020 at US$120 a share, opened at US$245 and closed the first day at US$253.93, valuing the company at approximately US$70 billion. It raised approximately US$3.4 billion, the largest software offering ever conducted, and the extent of the first-day rise, which more than doubled the price at which the company had sold its shares, produced a wider argument about how offerings are priced.
Berkshire Hathaway invested, which was unusual given its stated avoidance of technology and of initial offerings, as did Salesforce Ventures.
Growth was extremely rapid, with revenue more than doubling in several consecutive years, and the company's net revenue retention rate, the measure of how much more existing customers spend each year, exceeded 160 per cent for a sustained period, among the highest ever reported.
Slootman's approach, set out in his book Amp It Up, emphasised urgency, high standards and the raising of ambition, and he was among the more quoted executives in enterprise software.
He retired on 28 February 2024 and was succeeded by Sridhar Ramaswamy, who had run Google's advertising business for many years, had founded the search company Neeva, and had joined Snowflake when it acquired Neeva in May 2023 for approximately US$185 million.
Artificial intelligence and recent years, 2024 to 2026
Ramaswamy's appointment signalled the company's direction. Snowflake had been a data warehouse; it intended to be the place where machine learning and generative applications are built on that data.
Cortex, launched in 2024, provides large language models, vector search, embedding generation and model deployment within the platform, so that a customer can build applications against its own data without moving it. Snowpark allows Python, Java and Scala code to run inside the platform, extending it beyond SQL. Snowflake Intelligence, introduced in July 2026, provides conversational analysis across an organisation's data.
Acquisitions have supported this: Streamlit in 2022 for approximately US$800 million, which provides a framework for building data applications in Python and which is very widely used; Neeva in 2023; Crunchy Data in June 2025, bringing enterprise PostgreSQL capability; and several smaller purchases.
Snowflake reported product revenue of US$1.23 billion for the fourth quarter of its 2026 financial year, up 30 per cent year on year, with full-year product revenue substantially higher.
The data breach of 2024
In 2024 several hundred Snowflake customers were affected by a mass data theft and extortion campaign conducted by the group known as ShinyHunters.
Organisations whose data was taken included Ticketmaster, Santander Bank, Advance Auto Parts, Neiman Marcus, LendingTree, AT&T and others, and the volume of records involved was among the largest in any single campaign.
The mechanism is important and was widely misreported. Snowflake's own systems were not breached. The attackers used credentials stolen from customers' employees and contractors, in many cases through information-stealing malware on personal computers, to log into customer accounts that had not enabled multi-factor authentication. Snowflake had made multi-factor authentication available and had not required it, and the affected accounts were those where it had not been switched on.
The episode produced a general reassessment across the industry of the responsibility a platform bears for insecure customer configuration. Snowflake subsequently made multi-factor authentication mandatory for new accounts and introduced policies allowing administrators to enforce it, and the shared responsibility model, under which the provider secures the platform and the customer secures its use of it, was widely criticised as inadequate when the customer's failure affects millions of other people.
Connor Riley Moucka of Kitchener, Ontario, and John Erin Binns were charged in connection with the conspiracy, Moucka being arrested on 30 October 2024.
Products and technology
The Data Cloud is the company's term for the platform, comprising storage, compute, governance and the marketplace.
Virtual warehouses are the compute clusters, sized from extra small to very large, started and stopped automatically, and billed by the second while running. Customers buy credits and consume them, which means the revenue model is consumption rather than subscription, so that Snowflake's revenue rises when customers use it more and falls when they optimise their queries, a dynamic that has produced volatility in reported growth.
Time travel and zero-copy cloning allow a customer to query data as it was at an earlier moment and to create a full copy of a very large dataset instantly without duplicating the storage, because the clone references the same underlying files until changed. Both are direct consequences of the storage architecture and both were difficult or impossible in conventional systems.
Data sharing allows one organisation to grant another live access to a dataset without copying it, provided both are on Snowflake. This is the basis of the Marketplace, through which data providers sell access to datasets, and of the company's network argument: the more organisations on the platform, the more valuable it becomes to be on it.
Snowgrid provides replication and failover across regions and cloud providers.
Iceberg tables, supported since 2023, allow Snowflake to operate on data held in the open Apache Iceberg format in the customer's own object storage rather than in Snowflake's proprietary format. This was a substantial concession, made because customers and competitors had made the openness of table formats a competitive issue, and it reduces the lock-in that was part of the company's commercial position.
Competition
Databricks is the principal competitor and approaches the same problem from the opposite direction. It began with Apache Spark and machine learning on unstructured data and has added warehousing; Snowflake began with warehousing and has added machine learning. The two have converged on the same proposition and compete directly, and Databricks' acquisition of Tabular, the company founded by the creators of Iceberg, in 2024 was a direct move in the argument about open formats.
Amazon Web Services competes with Redshift and with its wider analytics portfolio, Google with BigQuery, which is architecturally similar to Snowflake and in some respects predates it, and Microsoft with Fabric and Synapse. All three are simultaneously Snowflake's infrastructure providers and its competitors, and Snowflake pays them very large sums for the storage and compute it resells.
Oracle, Teradata and IBM hold the installed base of conventional warehouses that Snowflake displaces.
ClickHouse, DuckDB and open-source alternatives compete at the lower end and, increasingly, in specific workloads.
Assessment
Snowflake is the clearest example in enterprise infrastructure of a company built on one architectural insight applied rigorously.
The insight, that separating storage from compute makes a data warehouse elastic and removes the administration, was available to everyone. Google's BigQuery embodied a similar idea earlier. What Snowflake did was build a product around it that worked on all three clouds, that required no tuning, and that could be sold to conventional enterprises by a conventional sales force, which is a different achievement from having the idea.
Its difficulties are structural. It buys its infrastructure from three companies that compete with it, which caps its margins and its independence. Its consumption pricing means that a customer optimising its spending directly reduces Snowflake's revenue, which produces unpredictable growth. Its differentiation has narrowed as competitors have adopted the same architecture and as open table formats reduce the value of keeping data in a proprietary one.
Its opportunity is that the data an organisation holds has become more valuable, because generative systems are only as useful as what they can read, and a company that holds that data in a governed and queryable form is well placed to be where those systems run. That is the argument behind Cortex and behind Ramaswamy's appointment, and it is the same argument Databricks, Microsoft, Salesforce and Palantir are making. The data layer is now the most contested position in enterprise software, and Snowflake holds a good one without holding the applications above it or the infrastructure below.