Apr 10, 2025 · 1h 23m · mad
Snowflake CEO on Winning the AI Arms Race
0:00 / 0:00
▶ Watch on YouTube →
gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions
In this episode of The MAD Podcast, Snowflake CEO Sridhar Ramaswamy joins host Matt Turck to discuss Snowflake's evolution into an AI Data Cloud, the strategic adoption of open formats like Apache Iceberg, and how his background at Google and Neeva shapes the company's enterprise AI strategy.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. Matt holds 13.6% of the talking time here. How this is scored →
Matt as informed peer 3.2 Guest teaching 3.6 Guest disagreement 1.1 Matt pushing back 1.4
speaking balance: gold is Matt, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 1:03:15 Rejecting direct competition in traditional BI
Sridhar forcefully rejects the premise that Snowflake should build a traditional BI tool, dismissing direct competition in entrenched categories as a fool's errand without a 10x product insight.
Hardest push from Matt ▶ 1:01:46 Calling out the missing BI layer in the stackMatt directly challenges Sridhar on leaving out BI from his modern data stack overview, bluntly asking if Cortex Analyst is designed to kill traditional BI tools.
Biggest teaching moment ▶ 42:16 Unpacking search ad query economicsSridhar educates Matt on search monetization mechanics, revealing how single high-intent keywords like auto insurance generate $150–$200 RPMs and why conversational agents threaten this cash cow.
Matt holds his own ▶ 15:34 Highlighting deterministic BI versus stochastic MLMatt demonstrates deep domain expertise by accurately diagnosing the cultural gap between Snowflake's traditional deterministic BI mindset and the probabilistic nature of modern machine learning.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | Matt as informed peer | Guest teaching | Guest disagreement | Matt pushing back | Why |
|---|---|---|---|---|---|---|
| Welcome & Reflections on Current Public Market Volatility | 2 | 2 | 1 | 1 | Sridhar details Snowflake's core architectural origin of uncoupling compute and storage, along with early go-to-market strategies targeting progressive asset managers. Matt asks polite clarifying questions about traditional VC views on selling into financial services. | |
| Streamlit Acquisition and Native Application Ecosystem | 3 | 3 | 1 | 1 | Sridhar reflects candidly on the culture clash between infrastructure engineering and iterative ML systems at Snowflake. Matt contributes a thoughtful observation comparing deterministic BI systems with probabilistic machine learning. | |
| The AI Data Cloud & Open Formats via Apache Iceberg | 2 | 4 | 1 | 1 | Sridhar outlines how embracing Apache Iceberg shifted Snowflake from a defensive open-format stance to expanding across the entire data lifecycle. Matt asks high-level guided prompts to tour the platform capabilities. | |
| Snowflake Cortex AI Architecture: Search, Analyst, and Agents | 2 | 3 | 0 | 0 | Sridhar breaks down the Cortex AI product suite, explaining how hosted model gardens allow developers to invoke LLMs via simple SQL and Python calls. Matt prompts for clear architectural definitions. | |
| Search Infrastructure & Technical Insights from Google and Neeva | 3 | 5 | 1 | 1 | Sridhar shares deep technical insights from Google's Mustang engine and Neeva's architecture, applying them to Cortex Search and Cortex Analyst. Matt connects Sridhar's personal career background to current Snowflake capabilities. | |
| Startup Lessons from Neeva & Transitioning to Snowflake CEO | 3 | 3 | 1 | 2 | Sridhar offers an unusually transparent post-mortem of Neeva's overvaluation and market timing relative to modern AI search tools like Perplexity. Matt prompts this reflection with a well-aimed comparison. | |
| Sridhar Ramaswamy's Career Journey & The 'Epiphany Mafia' | 4 | 2 | 0 | 0 | Matt showcases strong industry knowledge by detailing Sridhar's career path and identifying the influential Epiphany alumni network. Sridhar warmly confirms and expands on his early career mentors. | |
| The State of Web Search & Google's Innovator's Dilemma | 3 | 6 | 2 | 1 | Sridhar delivers a comprehensive breakdown of search monetization economics, explaining why conversational agents threaten Google's highest-margin ad queries. Matt guides the narrative on web search disruption. | |
| Data Strategy as the Essential Foundation for AI | 3 | 5 | 1 | 1 | Sridhar details why enterprise AI projects fail without pre-existing role-level security and data governance. Matt pushes for actionable advice for companies struggling with fragmented data. | |
| Apache Iceberg Strategy and Storage Monetization Reflection | 5 | 4 | 2 | 4 | Matt challenges Snowflake's business model regarding storage monetization versus compute revenue. Sridhar delivers a remarkably candid admission that charging for storage was a strategic error that accelerated open format adoption. | |
| Reimagining Business Intelligence and the Modern Data Stack | 5 | 4 | 2 | 4 | Matt astutely notices that Sridhar omitted BI from his overview of the modern data stack and asks if Cortex Analyst is killing BI. Sridhar clarifies why Snowflake focuses on conversational precision rather than building a traditional BI tool. | |
| Real-Time Streaming, Ingestion, and AI Infrastructure | 4 | 4 | 2 | 2 | Matt brings up acquisition rumors and probes the convergence of streaming and batch architectures. Sridhar offers a pragmatic breakdown separating ultra-low-latency messaging from practical analytical needs. | |
| Scaling Customer AI Adoption from Proof-of-Concept to Production | 3 | 4 | 1 | 1 | Sridhar details Snowflake's deliberate strategy to drive low-friction, cheap AI experimentation before worrying about monetization margins. Matt guides the conversation toward enterprise ROI and production scaling. | |
| Foundation Model Strategy: Building versus Partnering | 4 | 4 | 2 | 2 | Matt probes Snowflake's pivot from training its own foundation models to partnering with Anthropic and OpenAI. Sridhar candidly explains that a company of Snowflake's size gets priced out of frontier pre-training. | |
| Community Upskilling, Silicon Valley AI Hub, and Conclusion | 2 | 1 | 0 | 0 | Matt concludes with rapid-fire questions regarding Snowflake's developer education and physical startup incubator in Menlo Park. Sridhar details their community outreach and balance-sheet startup investments. |
Statements from this episode (33)
Ramaswamy compares April 2025 market volatility to early COVID-19 pandemic
“I think tumultuous is appropriate. I feel like We are in day two of the pandemic where we all knew something was wildly different, but we didn't quite know what. So it has that sort of something's big, but we don't quite know kind of feeling to it.”
Ramaswamy: Snowflake targeted asset managers early due to lower regulatory hurdles
“We very deliberately went after asset managers as our early customers, why they're more progressive. They want to make more money. They're earlier adopters of technology and have less regulation to deal with, say, compared to banks, which just have, it's more …”
Ramaswamy: Applications built on data create more value than sharing raw data
“There is a little bit of a hierarchy of needs when it comes to collaboration. You can share raw data, it creates some value. On the other hand, if you can share a predictive machine learning model, That uses the data to predict something, you will likely get m…”
Ramaswamy: Five nines availability in Google search ads was worth $100M
“Similarly, when I ran the search ads team, getting those five nines of availability actually meant a hundred million dollars more because that thing was running at such scale.”
Ramaswamy: Snowflake's early AI efforts suffered from an infrastructure-first mentality
“And so part of the difficulty that Snowflake had with machine learning and AI was they brought a similar mentality. You're going to write a design doc. It's going to take us 18 months. It's going to be great. Opposite of what you need to succeed in iterative e…”
Ramaswamy: Infrastructure engineering leaders cannot effectively evaluate AI talent
“Somebody that's very good at infrastructure cannot tell if somebody is amazing at AI, But it is this basic understanding that there are aspects to software that are just very, very different.”
Ramaswamy: Progressive CIOs demand vendor-neutral open data formats
“Increasingly what is happening is that the most progressive of the CIOs and the chief data officers What they want is independence with respect to vendors, including Snowflake. So they want their storage, their data to be stored in vendor-neutral formats, in o…”
Ramaswamy: Snowflake initially viewed supporting Iceberg as a defensive move
“And Snowflake initially approached this as mostly a defensive move.”
Ramaswamy: Snowflake operates hosted model gardens across all cloud deployments
“We run a model garden inside each of these deployments, and we offer a set of products on top of it.”
Sridhar Ramaswamy: Snowflake Cortex Search is based on Neeva's index infrastructure
“Cortex Search is almost completely based on the Neva Search Index infrastructure.”
Ramaswamy: Enterprise AI for structured data needs high precision over high recall
“In a business context, we are much better off with a very high precision, Modest recall kind of product.”
Ramaswamy: App store search platforms are terrible and show irrelevant ads
“If you compare more recent platforms for search platforms, for example, app store searches, they're truly terrible. Even when you ask highly precise questions, they will show you completely irrelevant ads.”
Ramaswamy: Unsuccessful startups accumulate cruft due to finite shelf lives
“Companies have shelf lives. Once you spend a certain number of years going at something, if you haven't achieved enough success, you also tend to accumulate a lot of cruft that becomes very hard to discard.”
Ramaswamy: Early-stage startups need magic built by five or six people
“I think there is virtue in taking, for example, just a few million dollars and saying the first version of a company needs to be built with five, six people in a relatively short period of time, and there needs to be some magic. And if there is not that magic,…”
Ramaswamy: Neeva reached a $300M valuation on under $1M revenue
“We raised too much money. Our valuation was three hundred million dollars for a company barely making a million. That just was not okay.”
Ramaswamy originally agreed to stay at Snowflake for six months
“My original agreement with Frank, the Snowflake CEO, was that I would stay with Snowflake for six months.”
Ramaswamy says he is happy to spend 5 to 10 years at Snowflake
“That's part of what made me decide that I'd be happy spending five or 10 years at this place.”
Google earns $1 to $2 every time a user searches 'auto insurance'
“You'll be shocked to know that there are queries that have 1502 thousand dollar RPMs, which means Google makes one or two dollars every time you or I, like, a user types the word auto insurance, words auto insurance into search.”
Ramaswamy: Consumers will use AI interfaces for auto insurance in 3-5 years
“In five years, in three years, if you and I want auto insurance, we are going to go to some conversational interface.”
Ramaswamy: Neeva failed because it wasn't 10x better than Google search
“The rules that killed Neva, which is, at the end of the day, until AI search came along, simply didn't have a product that was 10 times better than Google search. That's just the reality. That's the reason we couldn't compete.”
Ramaswamy: Google Gemini won't become a hit unless 10x better than ChatGPT
“They can't make Gemini into a consumer hit unless it's 10 times better than ChatGPT, and that's a really tall order.”
Ramaswamy: Google's ad spam team had over 150 people when he left
“I had entire teams devoted. It's called an ad spam team. Like when I left, I think they had more than a 150 people.”
Ramaswamy: Open formats like Apache Iceberg lack governance and access controls
“This is also where Iceberg or just a data lake strategy or a lake house strategy is not an immediate answer because formats like Iceberg do not understand governance rules, do not understand role-level access control, they don't understand users, they don't un…”
Ramaswamy: Snowflake has zero tolerance for 18-month projects without value
“At Snowflake itself, for example, I am very flat with my team that I do not have any tolerance for 18 month projects. If you tell me that you're going to start on a rewrite, and it's going to take you 18 months before it shows any value, you know, it's not, yo…”
Ramaswamy: Snowflake monetizing storage was a long-term strategic mistake
“My take is that us deciding to monetize storage was a long-term strategic mistake. I think we simply should have said, we will pass through storage costs.”
Ramaswamy: Snowflake will gain more revenue from Iceberg than lose in storage
“I expect us, me and the team, To drive way more revenue in new things we can do with Iceberg, then we are going to lose, potentially, by some of our storage revenue going out to cloud storage”
Ramaswamy: Competing directly against free Power BI is unrealistic
“Power BI is quite good. And it's essentially part of the office offering. And so I've been skeptical of saying that we can just compete in BI with a new paid product at scale when the bar is high quality, no price.”
Ramaswamy: Sub-50ms streaming for analytics has limited practical use and high cost
“Practical uses of true sub-fifty millisecond streaming for analytics is quite limited and somewhat expensive.”
Ramaswamy: Snowflake has thousands of production AI use cases deployed
“The net of this is that we have thousands of deployed production use cases on our AI products.”
Ramaswamy: AI inference will become significantly faster and cheaper in 2025
“The world of inference, not foundation models, the world of inference is going to go through so much change this year, both in terms of GPU availability, which is easing up quite a bit, But also in terms of people like Grok, I think the one with the K or the Q…”