Jun 18, 2025 · 37m · saastr
Snowflake's CEO on the AI Data Cloud, Partner Strategy, and What’s Next
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
Snowflake CEO Sridhar Ramaswamy and Observe CEO Jeremy Burton join Jason Lemkin to discuss Snowflake's evolution into an AI Data Cloud, the dynamics of consumption-based enterprise sales, and how strategic partner ecosystems scale modern cloud software.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. Jason holds 35.9% of the talking time here. How this is scored →
Jason as informed peer 4.3 Guest teaching 3.3 Guest disagreement 0.7 Jason pushing back 1.2
speaking balance: gold is Jason, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 29:07 Ramaswamy dispels superficial partner illusions
Ramaswamy delivers a blunt reality check to startups, stating that unless an internal champion's compensation and career depends on them, partnering announcements are merely superficial.
Hardest push from Jason ▶ 21:00 Lemkin challenges technical expectations for sales teamsLemkin directly questions how technical an enterprise sales rep must be to close complex data deals in the AI era.
Biggest teaching moment ▶ 17:37 Ramaswamy educates Lemkin on consumption accountingRamaswamy clarifies that Snowflake cannot recognize pre-committed bookings ratably under GAAP until actual compute credits are burned, leading Lemkin to admit he had the accounting wrong.
Jason holds their own ▶ 5:10 Lemkin shares SaaS case study on sales rep churn analysisLemkin demonstrates domain mastery by sharing how an SMB salon software company discovered their highest-performing sales reps were driving the highest customer churn using Snowflake data.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | Jason as informed peer | Guest teaching | Guest disagreement | Jason pushing back | Why |
|---|---|---|---|---|---|---|
| Introducing the Speakers, Snowflake, and Observe | 4 | 3 | 1 | 1 | Lemkin opens by introducing the guests and their enterprise pedigrees across Oracle, EMC, and startups. Ramaswamy politely corrects Lemkin when he assumes Neeva was acquired for a billion dollars. | |
| Snowflake Evolution into an AI Data Cloud | 5 | 2 | 0 | 0 | Lemkin illustrates Snowflake utility with a real-world SaaS case study on sales rep churn. Ramaswamy details Snowflake's roadmap into an AI Data Cloud leveraging agentic models and automated analysis. | |
| Enterprise Demands, AI Feasibility, and Incremental Value | 4 | 3 | 1 | 2 | Lemkin presses on what enterprise customers are demanding from Snowflake in the AI era. Ramaswamy explains the necessity of managing customer expectations and distinguishing between automated workflows and partial AI assistance. | |
| Observe Architectural Bet on Snowflake Platform | 4 | 4 | 1 | 1 | Burton walks through Observe's deliberate architectural choice to build atop Snowflake rather than creating a bespoke database. He breaks down the tradeoff of lower initial gross margins against faster time-to-value. | |
| Consumption Revenue Recognition and Land-and-Expand Economics | 3 | 6 | 1 | 2 | Ramaswamy educates Lemkin on Snowflake's strict GAAP consumption-based revenue recognition model. Lemkin admits he was unaware that pre-committed contract dollars cannot be recognized ratably without actual platform usage. | |
| Sales Specialization and Technical Knowledge in Sales | 5 | 3 | 1 | 2 | Lemkin inquires about sales specialization and shares an anecdote about a non-technical CRO being excluded from high-stakes AI deals. Ramaswamy explains that reps need business pattern matching rather than deep underlying parameter knowledge. | |
| Workload Scale, Systems Integrators, and CEO Time | 5 | 3 | 0 | 1 | Lemkin analyzes Observe's platform query volume relative to Snowflake's overall scale. Ramaswamy details how GSIs handle the bulk of implementations while he divides his executive time between road trips, product teams, and partners. | |
| Partner Strategy Discipline and Long-Term Networking Advice | 4 | 4 | 1 | 1 | Ramaswamy lays out the strict criteria for ecosystem partnerships, noting startups only have a real partnership if an internal employee's career depends on their success. Burton provides practical advice on maintaining long-term executive networks. | |
| The Evolution of Data Engineers and Analysts | 5 | 2 | 0 | 1 | Lemkin asks how data engineering and analyst roles will transform under AI workflows. Ramaswamy predicts a shift toward Cursor-style data orchestrations and semantic metadata tagging rather than manual query construction. |
Statements from this episode (20)
Ramaswamy: Snowflake is building deep research AI for all enterprise data
“That's kind of where Snowflake is headed, which is all of your enterprise data, but now the intelligence of something like ChatGPT deep research with access to every single data set that you have.”
Ramaswamy: Never commit to five-year implementation projects in the AI era
“I tell people, never sign up for five-year projects. I don't. If any of mine is doing RE, like any team comes to me and says, I have a great idea. It's going to take five years to implement. I go like, this is the age of AI. I just don't want to hear you. Talk…”
Burton: Observe is built entirely on Snowflake infrastructure
“How much of Observe is?
Jeremy Burton: 100%.”
Burton: Sub-100ms query latency is not required for enterprise SaaS apps
“There was a bunch of folks who had a theory that every query had to return in a hundred milliseconds, which I would like that as well, but that's just not required in an enterprise SaaS application, and so they wanted to build their own database.”
Burton: Building on Snowflake hurts initial gross margins before optimization
“Because in the short term, your gross margins are going to be worse than comparative companies, because obviously you've got to pay the Snowflake bill. But if you're committed, you'll figure out a way how to exploit the unique features of the platform. You'll …”
Burton: Observe has tens of millions in revenue with 60% gross margins
“And look, as we said today, amount of gross margins were, you know, tens of millions of revenue, and the gross margins are already up around 60%.”
Burton: Enterprises can burn down Snowflake commits by purchasing Observe
“Where it does play out in the larger accounts, obviously through, let's say the Snowflake marketplace, an enterprise customer may have a big commit with Snowflake. Well, they can actually burn down that commit by buying Observe.”
Ramaswamy: Snowflake's initial enterprise deals typically start at $50K to $100K
“Our typical first deals are like 50 K, a hundred K. They tend to start small.”
Ramaswamy: Snowflake customers can expand from zero to $5M in two years
“Some of our biggest customers have gone from like zero to one to two to five or a space of two, two and a half years.”
Ramaswamy: Snowflake strictly separates acquisition and expansion sales representatives
“We have completely separate reps for acquisition.
Yeah.
There's a hundred former division.
Sorry?
The teams that close the customers are off the deal when it closes?
They keep it for a year, year and a half.
We will continue to evolve that, but the set of peop…”
Ramaswamy: Snowflake targets winning 500 to 600 new customer logos weekly
“Then we have a different team that's focused on, you know, us winning five, 600 new logos every week.”
Ramaswamy: Enterprise sales reps must be technical enough to explain agentic AI
“Honestly, you have to be pretty technical today to succeed because your customer is going to ask you about Agent TKI.
You can't white code your way through answering what Agent TKI can do for the customer.
It's hard.”
Burton: Observe ingests 1PB of data and runs 190M Snowflake queries daily
“We do a hundred and ninety million Snowflake queries a day, and we ingest a petabyte of data.”
Ramaswamy: Snowflake processes approximately 5 billion queries per day
“Out of like five billion-ish queries a day.”
Ramaswamy: Partnerships are fake unless a partner employee's career depends on you
“And the thing that I always tell any partner is unless there's someone in snowflake whose career and future depends on you succeeding, you don't have a partnership. And by the way, that's true for every company.”
Ramaswamy: Data engineers will transition to AI orchestrators within two years
“I think there's going to be a lot more cursor style coding of these data engineering workflows. I think things like being able to extract metadata so that the data set that you extract is almost self describing so that AI can get to work on it. I think that is…”
Ramaswamy: Data analysts will shift from writing SQL to preparing AI-ready datasets
“I think what that data analyst, instead of writing millions of variations of basically similar queries, is now going to be describing datasets. He's going to be running these queries, but he's going to give semantic, they're going to give semantic context to w…”
Burton: Poorly documented past root causes are the major troubleshooting bottleneck
“The biggest problem we have right now is You see a problem. You think it's something you've never seen before. Actually you have, you just never root caused it. You never documented it properly. And by the way, when you need to search it, you can't find it.”