Three data-warehouse experts who spent years at Oracle have walked away to build the thing they believe Oracle never will: a database designed from scratch for the cloud. Benoit Dageville, Thierry Cruanes, and Marcin Zukowski have founded Snowflake in San Mateo, California, and for now, almost nobody is hearing about it, because the company is spending its early life in stealth mode, heads down, building. The bet is bold enough that, if it works, Snowflake could end up mentioned in the same breath as the other emerging cloud-data contenders. This is where that story starts.
The Problem They Left Oracle to Solve
To understand why three senior engineers would leave comfortable careers at the dominant database company on earth, you have to understand what is wrong with databases right now. The traditional data warehouse, the kind Oracle and Teradata sell, is built for a world of physical servers sitting in a company’s own data center. In that world, storage, the place your data lives, and compute, the processing power that runs queries against it, are bolted together in the same box. If you need more of one, you have to buy more of both, whether you need it or not.
Picture a restaurant where the only way to add another oven is to rent an entire additional dining room, even if your dining room is half empty. That is roughly how the old warehouses work. If your data grows but your query load stays flat, you still pay for more processing power than you need, because the two are tightly coupled. Worse, if ten analysts all run heavy reports at once, they compete for the same fixed pool of processing power and everything slows to a crawl. The system cannot flex.
The Idea: Separate Storage From Compute
Snowflake’s founding insight is deceptively simple and genuinely radical for the moment: pull storage and compute apart, and let each scale independently. Store all the data cheaply in one central place, in the cloud, and then spin up separate, independent pools of processing power on demand to query it, as many as you need, sized however you like, shut down the moment you are done.
Back to the restaurant analogy: instead of renting a whole new dining room every time you need another oven, you keep all your ingredients in one big shared pantry, and you summon exactly as many independent kitchens as you need for tonight’s orders, each working from the same pantry without getting in each other’s way. The finance team can run its month-end reports in one kitchen while the marketing team analyzes a campaign in another, and neither slows the other down because they have separate computers but share a single copy of the data. When the reports are done, the kitchens vanish, and the bill stops. You pay for storage cheaply and continuously, and for computing only in bursts when you actually use it.
This architecture is only possible because of the cloud. In a company’s own data center, you cannot conjure ten servers for an hour and then make them disappear, but on Amazon Web Services, you can, renting capacity by the minute. Snowflake is built natively on that reality from day one, which is exactly why three Oracle veterans had to leave Oracle to build it. You cannot bolt this design onto a decades-old product made for physical servers. It has to be born in the cloud.
The Quiet Years
Snowflake is staying in stealth, building the engine before making any noise. It has raised an early 26 million dollar round, and the plan is to bring in seasoned leadership to steer it out of stealth when the product is ready. That patience matters. Building a database that genuinely separates storage and compute while staying fast and reliable is enormously hard engineering, and rushing a half-baked version to market would kill the company’s credibility before it even begins. The founders are choosing to get it right first.
What It Means for the Market
Right now, Snowflake’s founding looks like a long shot against impossible odds. The data-warehouse market is owned by entrenched giants, Oracle, Teradata, and IBM, with decades of customer lock-in and enormous sales armies. A stealthy startup promising a cloud-native rebuild of the warehouse is the kind of bet most people ignore. But the timing may be ideal. The same year Snowflake is founded, Amazon launches its own cloud warehouse, Redshift, and Google is rolling out BigQuery, signaling that the entire industry is about to shift toward cloud-native data warehousing. Snowflake is not crazy; it is early, and aimed at what could be exactly the right wave.
The deeper competitive significance is that Snowflake and Databricks are attacking the same future from opposite directions. Databricks comes from the data lake and machine learning world, built around processing all kinds of messy data with Apache Spark. Snowflake comes from the structured data-warehouse world, built around clean, reliable, fast business analytics. For now, they serve different needs, but as each expands toward the other, they could collide head-on, and that collision has every chance of becoming the central rivalry of the data industry. The benefit to customers, even at this early stage, is the promise of the separation-of-storage-and-compute model: pay only for what you use, scale instantly, and never again buy a whole dining room just to add an oven. Right now, that is a radical pitch. If it works, it could become the industry standard, and Snowflake could be one of the companies that makes it so.