SNOWFLAKE

Snowflake was born from frustration with Oracle. Benoit Dageville and Thierry Cruanes were senior engineers at Oracle for over a decade, and they watched traditional data warehouses struggle with the cloud era.

The old approach — giant on-premise appliances that cost millions and took months to set up — was clearly dying. But nobody had built a data warehouse from scratch specifically for the cloud.

In 2012, Dageville and Cruanes teamed up with Marcin Żukowski, a Dutch computer scientist who'd built a high-performance analytical database engine called VectorWise. The three of them started building in San Mateo, California, with a radical idea: completely separate storage from compute.

In traditional databases, storage and compute are locked together — if you need more processing power, you have to buy more storage too, and vice versa. Snowflake said that was insane and decoupled them entirely.

They spent two years in stealth mode before launching in 2014. The product was immediately different from anything on the market.

You could spin up compute clusters in seconds, run queries across massive datasets without managing any infrastructure, and only pay for what you used. Companies that had been spending six figures a year on Teradata appliances could suddenly do the same work on Snowflake for a fraction of the cost.

The product-market fit was almost violent.

Snowflake charges based on consumption — you pay for the compute time and data storage you actually use. Compute is measured in "credits" consumed by virtual warehouses (their term for compute clusters), and storage is billed per terabyte per month.

This model is beautiful for Snowflake because customers rarely shrink their data — they only ever accumulate more.

The key insight was separating compute from storage. Customers can scale compute up or down independently, spin up multiple compute clusters against the same data simultaneously, and auto-suspend when not in use.

This means a company can run a massive analytics query during the day, shut down the warehouse at night, and pay nothing until tomorrow. Try doing that with Oracle.

Snowflake also makes money from data sharing. Their Data Marketplace lets companies buy and sell datasets directly through the platform — weather data, financial data, demographic data — without any copying or ETL.

Snowflake takes a cut of marketplace transactions and benefits from the network effects: the more data on the platform, the more valuable it becomes for everyone.

Snowflake Data Cloud — the core platform that stores, processes, and shares data across organizations with near-infinite scalability. Snowpark — a developer framework that lets engineers build data pipelines and ML models in Python, Java, or Scala directly inside Snowflake, without moving data out.

Cortex AI — their generative AI and machine learning layer that lets users build AI applications directly on their Snowflake data using LLMs and vector search. Snowflake Marketplace — a data exchange where over 2,000 providers list datasets that customers can access instantly without copying.

Streamlit — the open-source Python framework for building data apps, acquired in 2022 for $800 million, now deeply integrated as Snowflake's app-building layer.

Snowflake grew through a relentless enterprise sales motion combined with a product that genuinely sold itself. Early on, they offered free trials that let data engineers experience the speed difference firsthand.

Once someone ran a query in 10 seconds that took 20 minutes on their old system, the sale was basically done.

They also invested heavily in a world-class sales organization. Frank Slootman, who became CEO in 2019 after running ServiceNow, brought an aggressive operational playbook that dramatically accelerated growth.

Under Slootman, Snowflake went from $265 million to over $2.8 billion in annual revenue in four years. He was famous for saying "growth is oxygen" and running the company with military precision.

The Data Cloud strategy was the long game. By encouraging data sharing between organizations on the platform, Snowflake created network effects — the more companies use Snowflake, the more valuable it becomes for everyone.

Over 9,000 customers now share data through the platform, creating a gravity well that makes leaving increasingly painful.

Databricks is the existential threat. What started as a data engineering company has built a competitive SQL analytics product (Databricks SQL) that goes directly after Snowflake's core business.

The two companies are converging fast — Snowflake is pushing into data engineering and AI, Databricks is pushing into analytics. Both are spending billions to win.

Margins are also a constant battle. Snowflake runs on top of AWS, Azure, and Google Cloud, which means a significant chunk of revenue goes right back to the cloud providers as infrastructure costs.

Gross margins hover around 70% — good for most companies, but the cloud providers themselves operate at higher margins on the same underlying infrastructure. There's always the risk that AWS or Google could build "good enough" alternatives and undercut Snowflake on price.

So far, Snowflake's ease of use and ecosystem have kept customers loyal, but it's a fight they can never stop fighting.

Sutter Hill Ventures was the earliest and most consequential investor — managing director Mike Speiser actually served as founding CEO and incubated the company. Altimeter Capital, Dragoneer Investment Group, and Salesforce Ventures participated in growth rounds.

Warren Buffett's Berkshire Hathaway famously bought $250 million in shares at the IPO price — Buffett's first IPO participation in decades. Sequoia Capital and ICONIQ Capital also invested pre-IPO.

The September 2020 IPO raised $3.4 billion at a $33 billion valuation.