SRIDHAR RAMASWAMY: Just separating out what is real from what is hype, I think, is very hard, and I don’t think the AI industry helps itself with things like not talking about hallucination rates or not talking about, what does it take for something to truly be enterprise-grade? There’s a little bit of a, “Look, ma. It’s so cool,” kind of attitude. I think there is a maturity process that is going to happen. Figuring out what is real from what is hype is the biggest challenge that business leaders face today.

BOB SAFIAN: That’s Sridhar Ramaswamy, CEO of the cloud storage platform Snowflake. It’s been a wild few years for Snowflake, from a record-breaking IPO to a plummeting stock price, to a data breach scandal. Sridhar took over as CEO in the heat of the turmoil and has helped steady the ship, in part by leaning into AI in a very specific way. Today, he shares lessons from the company’s in-process turnaround, including insights behind high profile partnership with OpenAI and Anthropic, why Snowflake embraced China’s DeepSeek early, and how the buzz around agentic AI is being broadly misunderstood. There’s a lot to cover, so let’s get to it. I’m Bob Safian, and this is Rapid Response.

[THEME MUSIC]

I’m Bob Safian. I’m here with Sridhar Ramaswamy, CEO of cloud data platform Snowflake. Sridhar, it’s great to see you.

RAMASWAMY: It’s wonderful to see you, Bob.

Leading in a time of rapid change

SAFIAN: Snowflake has been on quite a roller coaster. It had the biggest software IPO ever in 2020, followed by a somewhat dramatic stock fall. You took over as CEO a year ago and almost immediately faced scrutiny related to a high-tech data breach. Yet you’ve also been in the heart of the hottest business arena ever, AI. Snowflake was one of the first American companies to tap into DeepSeek, the China-based open-source AI. It’s all kind of head-spinning. How’s it going? Has your first year as CEO been what you expected?

RAMASWAMY: The first year has been amazing, a little bit more than what I expected in terms of drama. But this is a time of rapid change, and I could not be more excited.

SAFIAN: I had a guest on the show recently who confided that a lot of CEOs are kind of paralyzed right now by sort of external uncertainties in the world, shifting tariffs, and regulations, and executive orders. How do you deal with, and think about, the environment and all the changes relative to the things that you can control yourself?

RAMASWAMY: One of my firm beliefs in life is that you need to focus on the things that you are going to have an impact on. There are many things that, let’s face it, we are simply not going to have any impact on. Obsessing about unchangeable things in the short term is the recipe for being uncertain about life. There is a lot of macro uncertainty. Businesses will react, and we will have to worry. For example, if the stock market keeps going down, or if the business climate gets worse, it’ll have an impact on Snowflake, but so far, it’s been heads down, get great product work done, get great customer deployments done.

The importance of strong relationships in a B2B business

SAFIAN: We first met when you were at Google. You were leading advertising and commerce, and then you started a search engine company, Neeva, that was ultimately acquired by Snowflake. Both of those businesses were more content-focused and ultimately consumer-oriented than Snowflake. What’s different about a full-on B2B business?

RAMASWAMY: There are many things that are similar in the sense that you make money off of deep relationships with your customers, absolutely. Commerce search was a consumer business, but Google has an amazing enterprise business. It’s called Ads. It predates the Google Cloud, and what was unique about Google, of course, was that a massive amount of revenue on Google Ads is completely self-serve. It’s quite magical. 50% of this absurd amount of revenue that Google makes is without ever talking to people. It’s all done via web interface.

It was quite magical, but the core of what I did at Google, which is build highly-reliable systems that never went down, and for us, that was all about money. If search ads went down, you were literally losing thousands of dollars every single second. That’s enough motivation to not bring stuff down. Then, work the relationships on the enterprise side, there is a lot of commonality to it, but obviously, it is a new set of stakeholders. Google mostly dealt with CMOs. Here, we mostly deal with CIOs, and chief data officers, and, thanks to AI, quite a bit of CEOs, as well. And so it’s building up a new network of relationships, but that’s good, honest work, and learning to cram 20 people’s names into your head is a good exercise now, as it was 10 years ago. It’s good fun, man.

“The most consequential data & AI company in the world”

SAFIAN: You recently said that Snowflake is the most consequential data and AI company in the world. That is an ambitious assertion, especially for a business that, at least previously, was known as a data storage company. How do you back up that claim?

RAMASWAMY: The most important data for the most important enterprises in the world is already stored on Snowflake. Snowflake is the gold standard for analytics. We have something like 700-odd Global 2000 companies that are on Snowflake, and if you exclude the folks from China that we are not even going after, that is 700-something out of 1,600. They all put their most important prized information on top of Snowflake. Large public companies close their books every month on top of Snowflake. Financial institutions share data with each other. Snowflake is the beating heart of at least the U.S. financial system in terms of how data moves from place to place.

What we have done over the past year is make AI a natural addition to how Snowflake operates. I think we are positioned incredibly well to continue what we did for data, which is make data now available through AI interfaces, through conversational interfaces, for these things to be tied, strung together into workflows that are increasingly going to serve higher-level business function, and that’s the reason for asserting that we are the most consequential enterprise AI and data company. Is there a bit of Benioff there? There are good qualities to Marc.

Leveraging partnerships with OpenAI, Anthropic, and DeepSeek

SAFIAN: I mentioned at the beginning that Snowflake was one of the first U.S. companies to adopt DeepSeek. You’re also the only data platform, big one, to offer models from both OpenAI and Anthropic. What did you see in DeepSeek, and second, why have you leaned into having multiple models available?

RAMASWAMY: Our strength is as a data platform. We are not a foundation model company, and honestly, most companies have no business of pretending that they are foundation model companies. It takes very specialized expertise, incredible talent density, and a very, very big wallet. And so for this, we decided to go the way of partnerships. We collaborate with a lot of folks. We focus on developing data products, which, in my mind, is the place where value is going to be realized.

When people think about OpenAI, they think, “Ah. These are the people that make the foundation models.” No, no, no. OpenAI is an amazing product company. ChatGPT is a legitimate product. It is going to approach the pantheon of the greats, the products that have a billion-plus users, and so helping people get value from models and the data that Snowflake has is what we are about. Hence the leaning into heavy partnerships. Things like hosting DeepSeek quickly, that’s just a little bit of making sure that you can still run the hundred-meter sprint in 10 seconds. It was a challenge. It was an amazing model. We had it out in two days flat.

SAFIAN: There was a lot of anxiety about DeepSeek. You don’t necessarily feel that same kind of anxiety, or even if you do, you feel like you have to have it available.

RAMASWAMY: Let’s break that anxiety down. There are many parts of DeepSeek. One is the open-source model. DeepSeek also offers services on servers that are hosted in China, where if you use their app, for example, everything that you are typing in is getting sent to China. Now, without getting too much into geopolitics, people will rightfully say that sending business data to China is a bad idea. It’s the same kind of fear that we have about TikTok. Hosting the DeepSeek model does not introduce any kind of security compromise. We host it. We take security and risk management very seriously. Us hosting DeepSeek did not cause issues like that.

SAFIAN: Any anxiety about, “Oh, DeepSeek can do things so much more cheaply than OpenAI. They’re cheaper, faster ways to build these models”?

RAMASWAMY: See, that’s the part of it that I actually like. That’s not anxiety. The reason I like that is because if there are highly capable-models that are freely available, the value of the data that is in Snowflake goes up. It doesn’t go down. The value of the model companies go down, and they have to innovate even harder. But innovation is a good thing for all of us. The cheaper that models get, the more broadly adoption there is, the more benefit that we, as society, are going to get, and certainly, Snowflake as a business. Hosting these models and running it ourselves without paying a toll, let’s face it, it can feel like that sometimes, to a bunch of other parties is, honestly, a good thing. Competition is absolutely a good thing.

Why agentic AI is misunderstood

SAFIAN: The big buzz in Silicon Valley today is around so-called age agentic AI. Now, there’s also some skepticism that the buzz doesn’t match use cases yet, the customers aren’t using AI agents as much as other simpler AI tools and chatbots. Can you quickly define, for folks who may not know, what age agentic AI is, and then explain why the buzz and the reality aren’t matching up, or at least not yet?

RAMASWAMY: Agentic AI is a vastly misused, misunderstood kind of word. It can honestly mean anything. I talked to a bunch of search companies, friends of mine that have started companies, and they make, honest to goodness, a search index that you can use to power a chatbot. They pretty much will look at me in the face and say, “Sridhar, we are powering agentic AI. That’s what we do for a living.” I go, “Wait. You make a search index.” But be that as may be, I think the promise is that language models are very good at coming up with plans for tasks. If you ask them a question, “I’m a bank. I have access to portfolios of stock that my customer owns. I want to create a report for Bob about his portfolio. What should I do?” it’ll come up with a very good plan of, “Okay. Go talk to this table. Get a list of all the tickers that Bob owns. Get additional information about these. Blend the two. Produce a great report for him.”

At its core, agentic AI is about stringing together different systems, also AI-based, to do something larger than what any individual system can do. Of course, you can imagine, in the same example that I gave you, that there is also access to a news corpus, some API that can get news about the different stocks that are in your folder. The same agent can go search for news that has happened over the past three days, and if something is deemed worthy enough, there’s a little bit of coding involved in defining worthy enough. It can say, “Hey, Bob. This is the current status of your portfolio. These are the movements in the fundamental metrics you care about, and here are top 10 articles about the companies that you have in your portfolio in case you’re interested in reading.”

What I’ve described here is a very simple agentic system that is able to talk to multiple sources, make a little bit of decisions about what is worthwhile news to show to you versus not, but you can imagine this getting more complicated. I tell people I dream of the day when I can have an agent talk to both United.com and Southwest to figure out what flight I should take. Right now, I have to do stuff like that manually, so I think those are coming. These are also pretty hard problems, Bob.

SAFIAN: It all feels like it’s just moving so fast, sort of like, what’s the next thing? I mean, because you have to plan further out to build these products, and yet it’s hard to know what’s going to come next.

RAMASWAMY: That’s life in AI. It is pretty hard to keep up, but on the other hand, the value that some of these tools can create is truly extraordinary. I’ve been using Gemini Deep Research. I use ChatGPT Deep Research, and the amount of value that I can get out of them is astounding. I think tools like that are here to stay. We are driving the adoption of those tools all across our sales teams. Because right now, before you meet a customer, for example, it is literally two minutes of work to be able to say, “Give me recent news about this customer. If there is any update to their financial performance, tell me about it, and if this person I am meeting has been in the news, give that to me, as well,” and like a minute later, you have all of that in front of you, the briefing ready, I think that’s pretty magical. Embracing change, absolutely, is very hard. I struggle with it. My team struggles with it. It’s one of those things that you just have to accept that it is uncomfortable, and that becomes a way of life.

SAFIAN: Sridhar is definitely living life in AI. I appreciate his acknowledgement that he struggles with the pace of change, that it makes him uncomfortable. Of course, he’s had plenty of experience being uncomfortable, including around last year’s big data breach. We’ll talk about that and more after the break. Stay with us.

[AD BREAK]

Before the break, Snowflake CEO Sridhar Ramaswamy made the case that Snowflake is the most consequential company in AI and data in the world. Now, he shares what he learned from last year’s big data breach of Snowflake customers, his use of weekly war room meetings, and why AI hype is the biggest challenge facing today’s business leaders, plus how he looks at the immigration debate in the U.S. as an immigrant himself. Let’s jump back in.

Navigating cybersecurity with shared responsibility

I wanted to ask you about the data breach controversy that hit last year soon after you came on as CEO. 165 companies who use Snowflake were impacted. The breach came via customers’ accounts, not through your software vulnerability, but still, I’m sure the public fracas wasn’t welcome. I’m curious how you faced that episode, whether it was an advantage or a disadvantage that you were early in your tenure, and whether there were any lessons you take away from it.

RAMASWAMY: Security. We are very clear with our customers. It’s a shared issue. Snowflake has offered multifactor authentication for over a decade. We offered something called network policies, where you can restrict who can connect to your Snowflake instance. Having said that, we are in it together with our customers. Me going out and saying that so-and-so is at fault is simply not helpful. We were very clear, after doing a lot of studies with external partners, that there were no breaches or vulnerabilities in Snowflake systems. We also started putting in place a series of schemes to make sure that we would act on our customers’ behalf a lot more quickly.

We now have systems that can detect access from surprising places and then warn our customers or turn off accounts. We now have dark web monitoring. So if there looks to be a credential that is compromised, then that credential is promptly turned off. We used this opportunity also to do other things like make sure that we had direct connections to the security folks in all our customers, so there was a lot of learnings in terms of, how do we make sure that we are in true collaboration with our customers? How do we make sure they understand that this is a shared responsibility? Hopefully, things like the incident that we experienced last year is a thing of the past.

SAFIAN: When the crisis first hit, I think a lot of folks’ impressions were, “Oh. This is not going to be good for Snowflake,” and yet, in some ways — obviously, you wouldn’t want it to happen again, but in some ways — it was good for your business?

RAMASWAMY: Look, many of our customers went through very unpleasant experiences as a result of this. This is not anything that I would wish on anybody, but having had it happen, you use it as an opportunity to both make yourself, the relationships, and the overall posture much, much better.

Inside Snowflake’s cross-functional “war rooms”

SAFIAN: You’ve attributed a lot of Snowflake’s recent progress to weekly war room meetings. I’m curious what those meetings entail and whether we should all be doing war rooms.

RAMASWAMY: War rooms, as you know, they have particular connotations, and one of my Google colleagues famously, I think it was 10, 15 years ago, objected to war rooms. So I renamed one war room to be a basket weaving room, but the idea is, how do you bring together people quickly in effective forums to help get past a stumbling block that we have? Snowflake’s war room was in the context of the product, the engineering, marketing, and the sales teams needing to work together, especially on new product offerings.

This is because Snowflake had come of age as a cloud data warehouse. We knew how to sell the cloud data warehouse. We didn’t need to bring these different functions together in order to figure out how to sell these things. On the other hand, something like AI, we didn’t know how to position it. We didn’t know what our customers were exactly looking for. We didn’t know the kind of problems that they were running into. This is where the close collaboration between the different teams that were responsible for taking new products forward were really helpful. The war rooms is much more in the context of, how do you do something new that you know you are going to struggle and get that to a point of maturity? So it’s for a very, very specific purpose.

Funding the next generation of AI companies

SAFIAN: Snowflake has invested in a lot of AI start-ups. You recently announced an expansion of your Start-up Accelerator, $200 million in new commitments. You also announced plans to build a new big AI hub at your Menlo Park campus, a $20 million AI upskilling program. How does this fit together? What does this ladder up to?

RAMASWAMY: Let’s start with the enablement investments. Our aspiration is to train a million people on using AI and data products, and we are doing this in a number of different countries, including in places like India, where an increasing number of Snowflake’s customers, often based in the U.S., are moving their technology operations to. On the start-up side, we have a very healthy balance sheet. It’s over $5 billion, and there are lots of start-ups that want to build on top of Snowflake, use Snowflake as a data platform.

So the $200 million fund is in combination with many venture partners to fund the next generation of companies. The AI hub is more a physical space. We want to host a set of companies that want to experiment with AI, and it’s a continuation of how we think about working with a technology ecosystem to help power the next generation of companies. We are just happy to continue that, because we see that as being mutually beneficial.

SAFIAN: You mentioned the trend of businesses moving to India. You are an immigrant to the U.S. from India. You came from India with just a few suitcases and a couple hundred dollars, as I recall. There’s so much angst in the U.S. around immigration right now. How much do you think about it, given your personal experience?

RAMASWAMY: Look, I’m incredibly blessed. I came with a bachelor’s degree, yes, I think it was $700. Neither of my parents went to college. I got a doctorate from Brown that Brown entirely paid for. I got a monthly stipend and a free PhD, and I think I’ve contributed in meaningful ways to the country, helping create great, amazing businesses. I think the larger issue is that our population feeling like there is enough prosperity to go around. People in our country need to feel like they have a prosperous future before they’re willing to lean in and say, “We want more immigrants to share in that prosperous future.” But I think those are the core issues that our government needs to address, where all of us feel like they have the opportunity like I got the opportunity. My take is there’s no generosity without prosperity.

SAFIAN: As a technologist engaged in the creation and the advance of technology, it sure sounds like or feels like it’s enhancing opportunity and prosperity, but that’s not translating to the broader public, necessarily.

RAMASWAMY: It’s more than perception. I think the honest question that all of us need to ask ourselves, not just the immigrants, everybody, the government, the population, is: is the prosperity truly broad? Are there truly opportunities? Me arguing that people like me are going to create more companies and more jobs, call it prosperity, in the Midwest, is just not going to resonate.

Deciphering reality from AI industry hype

SAFIAN: You’ve said this, that you can only unlock opportunity by embracing change, and for you as a business leader trying to embrace change, there’s also risk in moving too early when things are changing so much. How do you know what change is sticky, especially when headlines seem to blare about new convulsions every day?

RAMASWAMY: Build on strength. There are many things that are very real about Snowflake. There is nothing unreal about the 3.5 billion dollars that we made as revenue, about the 10,000-plus customers that we have, about the mission-critical role that we play. So with AI, for example, we didn’t chase the hype. We didn’t say, “Oh. We’ll fine-tune and host models for you. Here is a brand new way of making money that has nothing to do with what Snowflake used to do before.” We said, “Let’s offer AI as a natural companion, as a natural enhancer of what you’re already doing with it.”

And so you need to be very deliberate and thoughtful about how you create value, what your place in the world is, and finally, you also have to be a little bit of a portfolio manager. There are five or six things that we are going to try. Some of them are not going to work out, and you have to have unpleasant conversations, hard conversations internally about what you need to stop. We used to develop foundation models, but we backed away from it, because we just said, “We just cannot afford to spend the amount of money and talent that is required to train foundation models.” Was it an easy conversation? Absolutely not, but I think those are also important conversations, where you accept that you’re not going to know everything, and when the outside world speaks, you actually listen, and adjust, and move along.

SAFIAN: What do people and business leaders most misunderstand about the state of technology right now?

RAMASWAMY: I think they are feeling both pressure about things like AI, but are also flooded with options for what to do. I think there’s just so much just noise coming in terms of partnerships between X and Y or this new agent, take this or the other. I think that just separating out what is real from what is hype, I think, is very hard. I would say this is less a misunderstanding than an amount of confusion, and I don’t think the AI industry helps itself with things like not talking about hallucination rates or not talking about, what does it take for something to truly be enterprise-grade? There’s a little bit of a, “Look, ma. It’s so cool,” kind of attitude to some of the things that happen in AI. I think there is a maturity process that is going to happen, but I think figuring out what is real from what is hype is the biggest challenge that business leaders, enterprise leaders face today.

SAFIAN: Well, Sridhar, this was great. Thanks so much for doing it.

RAMASWAMY: Thank you, Bob. Really appreciate chatting.

SAFIAN: Sridhar may be soft-spoken, but he doesn’t mince his words, especially when it comes to AI. I agree that separating what’s real from the AI hype is critical for today’s business leaders, and if you’re a non-technical business leader, that challenge is even harder. As for Sridhar’s own hype, is Snowflake actually the world’s most consequential company in AI? Might it become that one day? I don’t know. Still, I feel smarter after listening to him. There’s a thread that runs through Sridhar’s comments that resonates for me beyond AI about confusion in the face of a lot of noise.

Whether we’re talking technology, tariffs, immigration, or whatever else, we strive to anchor on what’s solid. Some of our confusion is because we want clear answers when there just aren’t any, and even the most advanced AI won’t solve that problem. That’s just something we have to learn to live with. I’m Bob Safian. Thanks for listening.