Latest / Startup Stories with Fexingo: Conversations About Founders, Funding, and Building Companies from Zero / How Snowflake Broke the Data Warehouse Mold
Transcript
- Lucas: Imagine you're building a data warehouse back in 2012. You've got terabytes of customer data, sales records, clickstream logs. You buy a bunch of servers from Dell, load them with a traditional database like Teradata or Oracle, and you pray you sized it right. If you underestimated, queries take hours. If you overestimated, you just burned a million dollars on idle hardware. Luna: Right, and the worst part is you can't change it. You're stuck with whatever you bought for three to five years. Lucas: Exactly. And that's the problem Snowflake set out to solve. Today's episode is about how two French database engineers, Benoit Dageville and Thierry Cruanes, looked at that problem and said: what if we just separate compute from storage? That simple architectural insight—now called 'compute-storage separation'—let them build a data warehouse that could scale up and down on demand, in the cloud, and charge by the second. It made Snowflake one of the most valuable software companies in the world. Luna: And one of the biggest IPOs ever. I think the numbers are still staggering. Lucas: Yeah, and you know, speaking of things that are worth it—if you've found these startup stories useful in your work, even just one practical takeaway you've applied, that's the whole reason we keep doing this ad-free. If today's episode felt like it was worth a coffee to you, you can find us at buy me a coffee dot com slash fexingo. Totally optional, but it helps keep the show independent. And now, back to Snowflake. Luna: Yeah, absolutely. So these two engineers—they were at Oracle for years. What made them decide to leave and build something that would directly compete with their old employer? Lucas: That's the really interesting part. Dageville and Cruanes were both senior architects at Oracle, working on the Exadata line. They saw the cloud coming—Amazon Web Services was already four years old—and they realised Oracle's whole model was built around the assumption that hardware and software were inseparable. You bought a machine, you got the database. But in the cloud, you could decouple them. And that unlocked something huge. Luna: So they left Oracle in 2012 and started building. But they didn't just write code right away, right? They spent a long time just thinking about the architecture. Lucas: Months. They actually rented a chalet in the French Alps, in a ski town called Les Gets, and just whiteboarded for weeks. No investors, no pitch deck, just two guys arguing about how to build a database from scratch for the cloud. They called the project 'Snowflake' early on because they wanted something that felt cool and fresh—nothing to do with data being cold. Luna: I love that. And then they met Mike Speiser from Sutter Hill Ventures, who basically became their co-founder in a weird way. Lucas: Yeah, Speiser was an ex-Oracle guy too. He heard their pitch—which at the time was more of a concept than a product—and he committed $25 million on the spot. That's almost unheard of in enterprise software. But Speiser had a thesis: that the data warehouse market was about to be completely reinvented by the cloud, and the incumbents like Teradata and Oracle couldn't pivot fast enough. Luna: What about the actual build? Because one thing I've always found interesting is that Snowflake didn't use an open-source foundation. They built their own proprietary engine from scratch. Lucas: That was a huge bet. Most new databases in that era—like Google's BigQuery or Amazon's Redshift—either built on top of open-source projects like PostgreSQL or used public cloud infrastructure abstracts. Snowflake wrote its own storage layer, its own query compiler, its own execution engine. The advantage was total control over performance. The disadvantage was it took them three years to get a beta out the door. Luna: So they went dark for almost three years. That's a long time when you've raised $25 million. Lucas: And they almost ran out of money. By 2014 they had maybe six months of runway left. But then they finally launched a private beta, and the response was immediate. Customers loved that they could spin up a warehouse in minutes, run a query, and spin it down. No provisioning, no capacity planning. And the pricing model—per second, with no upfront commitment—was totally new. Luna: Let's talk about that pricing. That was a big part of the disruption, right? Because traditionally data warehousing was a big capital expenditure. Lucas: Right. Teradata would sell you a $2 million appliance and then charge you 20% annual maintenance. Snowflake flipped it: pay for what you use, by the second, and if you run nothing, you pay nothing. That made it accessible to startups and mid-market companies that could never afford a traditional data warehouse. And it made CFOs very happy. Luna: But there's a catch, isn't there? The 'data cloud' network effect. Lucas: Yeah, the flip side is that once your data is in Snowflake, it's hard to leave. They built this thing called the 'data sharing' feature, where you can share live data between Snowflake accounts without copying it. That creates a network effect: the more customers use Snowflake, the more data is available inside the ecosystem, and the more value you get from staying. It's not a traditional lock-in, but it's sticky. Luna: And it's not just about storage, either. Snowflake also built a marketplace where third parties can sell data sets. So if you're a hedge fund, you can buy satellite imagery data right inside Snowflake and join it with your own analytics. Lucas: Exactly. That marketplace is a big part of their moat. But the core innovation—the thing that made them a billion-dollar revenue company—is still that architecture. The separation of compute and storage. It sounds like a technical detail, but it fundamentally changes how businesses think about data. Luna: And then came the IPO. September 2020. They priced at $120, opened at $245, and closed the first day at $253. It was the largest software IPO ever at that point, with a market cap over $70 billion. Lucas: And it happened during a pandemic, when most IPOs were being delayed. But Snowflake's revenue growth was just too strong—they were growing over 100% year-over-year. And they had this massive total addressable market story: every company in the world has data, and most of them still run it on-premises. The cloud migration was just beginning. Luna: But then came the competition. Amazon Redshift, Google BigQuery, Microsoft Azure Synapse—all the hyperscalers started offering similar services. How did Snowflake hold its ground? Lucas: A few ways. First, they stay cloud-agnostic. You can run Snowflake on AWS, Azure, or Google Cloud. That's a huge advantage if you don't want to be locked into one provider. Second, they keep innovating—they added support for unstructured data, machine learning, and real-time analytics. And third, they have that network effect from data sharing. But honestly, the competition is fierce. Redshift especially is very good and very cheap if you're already on AWS. Luna: So what's the lesson for founders here? I think it's that sometimes the biggest breakthrough isn't a new technology—it's a new architecture that unlocks a new business model. Lucas: Absolutely. Snowflake didn't invent cloud computing, or databases, or even the idea of separating compute from storage—Google had done that with BigQuery. But they were the first to package it in a way that enterprises could buy easily. They turned a technical refactoring into a product revolution. And they did it by focusing on the customer pain of unpredictable costs and rigid infrastructure. Luna: And they did it with a very small team for a long time. I think at the time of the IPO they had just over 1500 employees. That's lean for a company that's essentially building a new database layer. Lucas: It's a reminder that the best software companies are often the ones that do one thing really well. Snowflake's mission from day one was: make data warehousing easy. They didn't try to do everything. They just executed that one idea better than anyone else. Luna: So where are they now? It's mid-2026. They've had some ups and downs post ipo. What's the current state? Lucas: They're still growing, but growth has slowed to around 30-40% annually, which is natural when you're over $3 billion in revenue. The stock has been volatile—it hit a high of over $400 in 2021, then dropped below $150 in 2022, and now it's trading somewhere in the middle. The big question is whether they can maintain their premium valuation as the hyperscalers catch up. Luna: And they're betting on AI now, right? They launched Snowpark to let data scientists run Python and ML models directly inside Snowflake. Lucas: Yeah, Snowpark is their attempt to become the platform for all data workloads, not just SQL. They want to be the place where you train models, not just where you query tables. It's a smart move—it increases stickiness and expands the TAM. But it also puts them in more direct competition with Databricks, which is essentially the leader in that space. Luna: So it's a three-way battle now: Snowflake, Databricks, and the hyperscalers. Who wins? Lucas: I don't think anyone 'wins' outright. The data infrastructure market is huge enough for multiple players. But Snowflake has a head start in ease of use and enterprise trust. If they can keep innovating and stay independent, they've got a good shot at being a dominant platform for the next decade. The key is whether they can keep their culture of engineering excellence as they scale. Luna: And that's the real challenge for any startup that makes it big: how do you hold onto the original spark while becoming a giant? Lucas: Yeah, that's the question every founder should ask themselves. Because the architecture that got you to $1 billion might not be the one that gets you to $10 billion. And the team that built the first product might not be the team that builds the second. Snowflake's story is still being written.