Snowflake

Snowflake made one architectural bet in 2012 that reshaped the data warehouse: separate the storage from the compute so you can scale each on its own and pay only for what you run. Benoit Dageville and Thierry Cruanes came out of Oracle, Marcin Zukowski out of the Vectorwise project, and together they built a warehouse that lives on AWS, Azure, and GCP instead of your own servers.

That idea grew into what Snowflake now calls the AI Data Cloud, used by more than 12,000 organizations running billions of queries a day. It went public on the NYSE in 2020 in the largest software IPO ever, and today it carries a market cap near $88 billion. The current chapter, under CEO Sridhar Ramaswamy, is a hard pivot toward agentic AI sitting directly on top of the data.

Snowflake sells a single managed place to put all your data and then do things with it: warehousing, data lakes, engineering pipelines, analytics, data science, and now AI apps and agents. The pitch is that you stop stitching together separate databases and stop babysitting infrastructure, because Snowflake handles the scaling, tuning, and maintenance underneath.

The part that made it different is data sharing. Companies can discover, buy, and swap governed datasets across clouds and regions without copying anything, which is why so much of a customer's data ecosystem ends up living inside the same platform.

The problem Snowflake keeps hammering is fragmentation: data scattered across systems, teams that cannot share it safely, and infrastructure that eats engineering time. Its answer is a fully managed service where compute is elastic and serverless, so you are not sizing clusters or patching servers to keep queries fast.

The newer problem it is chasing is AI on enterprise data. Agents need governed, low-latency access to live data to be useful and trustworthy, and Snowflake is positioning its engine, catalog, and Cortex tools as the safe substrate for that.

This is enterprise software first. The sweet spot is a large company with complex pipelines and a real data team, the kind of org that benefits from scaling compute and storage independently and sharing governed data across business units. Mid-market and smaller analytics teams use it too, and Snowflake courts them with a self-service trial and $400 in free credits.

The buyers are data and analytics leaders, engineers, and increasingly the AI teams trying to put agents into production. The catch, and reviewers say this often, is that the consumption model rewards teams that watch their spend and punishes those who do not.

The core product is the Snowflake Platform, one managed service spanning data warehousing, lakes, engineering, science, app development, and sharing across AWS, Azure, and GCP. Everything else clips onto it.

The recent push is agentic: Cortex is the AI layer for building on your own data, CoWork is a work agent that answers questions in natural language inside Snowflake's security perimeter, and CoCo is a governed coding agent. Snowflake also now ships a native Postgres, the fruit of its Crunchy Data acquisition, so transactional and analytical workloads can sit side by side.

Snowflake makes money on consumption, not seats. You buy compute in credits and pay for storage by the terabyte, either month to month on demand or through pre-paid capacity commitments that come with discounts. There is no per-user license, which is why a data team of five and a data team of five hundred can both be customers.

The model is a double-edged sword. It aligns cost with usage, but it also means an unoptimized query or a runaway warehouse shows up on the bill, and cost surprises are the single most common complaint from customers.

Snowflake's biggest rival is Databricks, which comes at the same problem from the data-lake and machine-learning side, while Google BigQuery and the incumbent warehouses from Amazon and Microsoft round out the field. Snowflake's edge is that it runs the same experience across all three major clouds, so customers avoid lock-in, and that its sharing and marketplace create a network effect competitors have to match.

The strategic bet now is agentic AI. The OpenAI partnership, the Cortex agent stack, and the Crunchy Data Postgres move are all aimed at making Snowflake the place where enterprise AI runs, not just where the data sits.

The traction shows up in customer results, not slideware. Booking.com moved off Hadoop and unified 31 million travel listings on Cortex AI. The Massachusetts Executive Office of Education says it saves $1.5 million a year after migrating off Oracle, with analytics running 30% faster on a quarter of the compute.

AT&T reports 84% savings on estimated annual costs from results caching and answers 90% of queries in under a second. KFC cut database operational costs by 70% while processing more than 500,000 order transactions a day. Across the platform, Snowflake counts more than 12,000 customers and 3,400 marketplace listings.

The praise is consistent: it scales cleanly, it is genuinely low-maintenance, and separating storage from compute lets teams dial resources to demand without much fuss. Data sharing across business units, without duplicating storage, gets singled out by large orgs again and again.

The gripe is just as consistent, and it is money. The consumption pricing can surprise teams with bills bigger than they modeled, hidden costs like cross-region transfer add up, and reviewers warn that without a dedicated eye on usage the spend gets away from you.

Snowflake raised heavily as a private company, backed by Sequoia, Altimeter, Iconiq, and others, before its 2020 IPO became the largest software listing on record. As a public company on the NYSE, the relevant number now is market value, not a funding round: roughly $88 billion as of mid-2026.

Revenue is running near $5 billion on a trailing basis with product revenue up about 32% year over year and net revenue retention around 125%. The company is still posting GAAP net losses, so the story investors are watching is whether AI-era growth converts to profit.

Snowflake has the rare combination of scale and momentum: roughly $5 billion in revenue, growth in the low thirties, and retention that says existing customers keep spending more. The agentic-AI pivot, if it lands, gives it a second act well beyond the warehouse.

The overhang is twofold. It is still unprofitable on a GAAP basis, and its own pricing model, the thing customers complain about most, is the friction it has to manage as it pushes heavier AI workloads onto the platform. The bet is that being the trusted, cross-cloud place where enterprise AI runs is worth the premium.

Snowflake runs on eight stated values, and the ones that keep surfacing in job posts are Put Customers First, Get It Done, and Own It. The company describes itself as low-ego and high-standards, and it went distributed in 2021 with its principal executive office in Bozeman, Montana and more than 10,000 employees across 40-plus offices.

The sharper cultural shift right now is the AI-native mandate. Engineering postings openly expect people to treat AI as a collaborator, build and deploy autonomous agents in their own workflows, and dogfood Snowflake's own AI products. It is a real bet, and it is showing up in how they hire.

Pay skews high and is disclosed on a healthy share of roles. Engineering bases land roughly between $96K and $437K depending on level, with product and design clustering in the low-to-high six figures, all in USD, and Snowflake also posts bands in CAD, CHF, and AUD for its global offices.

Base is only part of the package. Roles include equity as standard, and sales roles run on OTE with commission that follows a MEDDPICC-driven enterprise motion, so total comp for closers depends heavily on quota attainment.

Security is not a footnote for Snowflake, it is part of the sell. The platform builds in end-to-end encryption, role-based access control, network policies, multi-factor auth, and data masking, with the Horizon Catalog handling unified governance, and the top Business Critical and VPS tiers add controls for regulated industries.

The legal entity is Snowflake Inc., and its privacy notice, last updated July 2026, draws a clean line: customer data uploaded to the Service is processed only on the customer's behalf and governed by that customer's own agreement, not Snowflake's marketing-site policy.

Snowflake's recent headlines all point the same direction: make the platform the home for enterprise AI. It acquired Crunchy Data for around $250 million to bring enterprise Postgres natively into the cloud, then shipped Snowflake Postgres to GA so transactional and analytical work can share one platform.

On the AI side it struck a roughly $200 million partnership with OpenAI and moved its Cortex agent tools, including Cortex Code and Semantic View Autopilot, into general availability.