Snowflake Built a Moat Out of Data Gravity. Its Own Numbers Show the Water Receding.

A retailer and a logistics firm want to combine their data. The old way: someone exports a file, ships it across a wire, the other side ingests a copy, and within a week both companies are arguing about which version is current. Snowflake's pitch removed the file. Inside its platform, a provider grants access and the consumer simply queries the same data in place - no export, no copy, no stale duplicate. Multiply that across more than 2,400 ready-to-use data products from over 540 providers in its Marketplace⁴, and you get something that looks less like a database and more like gravity: the more data lands inside Snowflake, the harder it is for any of it to leave.

The official story is that this makes Snowflake an unbreachable fortress - a data network effect that compounds with every new customer and locks the whole ecosystem in. Snowflake's own 10-K says as much. The real story is narrower and more interesting: the moat is genuine, it is measurable, and it is leaking - and Snowflake's own numbers are where you watch the water go down.

The one number that proves the moat is real

Forget the architecture diagrams. The cleanest evidence that data gravity exists sits in a single line of the financials: net revenue retention. It measures what a cohort of existing customers spends this year versus last - new logos excluded entirely. A figure above 100% means customers, on average, dig in and spend more without anyone selling them. Snowflake's was 131% for fiscal 2024 and 126% for fiscal 2025.¹² In a year when its overall revenue still grew 29% to $3.63 billion², the customers it already had kept expanding by a quarter. That is not marketing. That is the sound of data accumulating and refusing to move - because once a company's analytics, its partners' shared tables, and its Marketplace feeds all live in one place, ripping them out costs more than tolerating the bill.

And the bills are sticky in their own right. Snowflake counted 745 of the Forbes Global 2000 as customers in fiscal 2025, and that cohort drove roughly 42% of revenue.² Remaining performance obligations - contracted revenue not yet recognized - climbed to $6.9 billion.² The gravity is real, contracted, and concentrated in exactly the large enterprises that move slowest.

Watch the water recede

Here is what the bull case skips. That 126% is not a flat plateau or an improving trend - it is the low point of a steady, quarter-by-quarter slide. As recently as April 2023, net revenue retention stood at 151%. By October 2023 it was 135%; by January 2024, 131%; through the fiscal-2025 quarters it stepped down to 128%, then 127%, then 126%.³ Nine straight quarters of compression. A moat is supposed to make expansion automatic. When the rate of automatic expansion falls 25 points in under two years, the moat is not gone - but the water is visibly draining out the bottom.

Where the gravity actually comes from - and where it doesn't

For years the standard moat story credited Snowflake's founding trick: separating compute from storage, so customers could scale and pay for each independently. That edge is gone. As Contrary Research put it, 'Snowflake's original moat was its ability to separate compute from storage, but that competitive advantage has been eroded as most modern data platforms provide similar capabilities.'⁸ The architecture that built the company no longer defends it. Whatever moat remains has to rest on the harder-to-copy thing: the network of data sharing and the Marketplace ecosystem, where value comes not from the engine but from the other companies already inside.

But the central selling point of that network - zero-copy sharing - is a simplification that breaks precisely where the biggest customers live. In a single region, the no-copy story is true: the consumer queries the provider's data in place. Cross-region or cross-cloud, it isn't. Snowflake's own documentation states that when a consumer sits in a remote region, Snowflake 'enables auto-fulfillment to replicate data to the remote region.'⁶ Replicate is copy. So the multi-cloud, multi-region enterprise deployments cited as the strongest moat evidence are the same ones where the frictionless-sharing promise quietly becomes ordinary replication - with the cost and the copies that come with it.

But Iceberg proves Snowflake is opening up - doesn't it?

The strongest objection to the leaking-moat thesis is that Snowflake is widening it, not narrowing it. Marketplace scale is real - over 2,400 data products from more than 540 providers⁴ - and the network effect is exactly what Snowflake's filings stake their growth strategy on.⁵ The fair version of the bull case is that gravity at this scale is self-reinforcing, and the slide in retention is mostly a macro cost-optimization cycle, not a structural crack. That's a serious argument, and the contracted $6.9 billion backlog supports it.²

The honest counter is that the most credible outside scorecard disagrees. Morningstar rates Snowflake 'no-moat' as of early 2026 - even as it raised its fair value estimate to $223 - because the switching-cost evidence still falls short of a formal moat threshold. Worse, Morningstar notes the AI tailwind everyone is excited about benefits Databricks and Google's Vertex AI just as much.⁷ A rising tide that lifts your rivals equally is not a moat; it's weather. The deeper problem is that the openness Snowflake now advertises cuts both ways: the easier data is to share out, the easier it is to leave - and the gravity that held the network together depends on it being hard to leave.

Snowflake built one of the cleverest pieces of gravity in enterprise software: put the data in one place, let everyone query it in place, and let the accumulation do the locking. It worked - the retention numbers prove the gravity is real. But a moat is not a level; it's a trend. The founding architecture has been matched, the zero-copy promise leaks at the multi-cloud edges, and the one metric that proved the gravity has slid 25 points while the company kept insisting the fortress was getting stronger. The water in a moat doesn't have to be drained to be dangerous. It only has to be visibly going down faster than anyone's filling it back up.