Snowflake Inc. (SNOW) Earnings Call Transcript & Summary

Good afternoon, good evening. Thanks for joining day 2. I think this is the last presentation of the day. We saved the best for last. We have Mr. Mike Scarpelli from Snowflake here in person wrapping off day 2 for the software track. So Mike, thanks for making the journey out to New York City. Thanks for the support of our conference. Obviously, a packed house here. I think almost every seat is filled. So for investors that may be new to the story, I'm sure 90% know who Snowflake is, but maybe just give a 30-second overview on the company.

Sure. Thank you, everyone. Snowflake's 12 years old now. Snowflake really started out as going after the cloud data warehousing market, but it was really a cloud platform, and data warehousing just happened to be one of the workloads. We gradually grew into a lot of data engineering workloads and we -- one of the things that customers like about Snowflake, number one is data sharing. No one does data sharing the ways Snowflake does. People like the ease of use of Snowflake. Does not require a very sophisticated user. It just works. We do all the work in the background. And it's the exact same experience whether you're running in AWS, Microsoft or Google. We can do cross-cloud, cross-region replication very easy. We have a number of data listings in our marketplace that are available around the world and to any region if anyone wants to use it. And yes, we just reported our Q2 results a couple of weeks ago, and we're off to a good start this year.

Great. So speaking of kind of the recent changes, the CEO transition was a big one earlier this year, Frank Slootman handing off the reins to Sridhar Ramaswamy. What have been the main highlights that you've seen, obviously, working with both executives? What has sort of Sridhar been focused on in the 2 quarters since he's taken over?

Sure. So Sridhar joined at the beginning of this year, but that was something we were working for a few months in discussions with him. And we got Neeva -- Sridhar through the acquisition of Neeva in May of 2023. When he came on board, and I remember I told Frank when we were doing that deal, I'm like, Frank, you need to spend time with Sridhar. He could be your successor. Originally, Sridhar only wants to stay 6 months, agreed to stay for 6 months, and then he was going to go off. He wanted to be CEO somewhere and he had other opportunities to be CEOs of public companies as well as running all of the generative AI in, I'll just say, one of the hyperscalers. And so when the opportunity presented itself with Sridhar, we had to move on that because he had some other opportunities. And Frank, we had been talking about succession planning for 2 years. When Frank joined the company, he had told the management, he really only want to be there for the Board 3 to 5 years. He had come out of retirement to do this. So it may have been a shock to some investors, but it was definitely not a shock to me on that. And in terms of what are the big things and why we think Sridhar is the right person, Snowflake has a very large engineering organization. Very technical engineering organization. The product is going down more into in everything happening with AI. We really thought we needed a leader who had more technical chops. That is not Frank. Frank is more of a go-to-market. He's really good at product positioning, understands product management well, but he's not an engineer. And Sridhar is. And also, Frank was getting, I don't want to say this, he's not listening. But Frank was getting old. And the energy level, I joke around with him, that he's 65, 66, he'll be this year. He was ready to retire. I hope I'm not working when I'm 65, 66. But Sridhar has really, really had an impact on the engineering organization and really driving the delivery of releasing new product features and capabilities faster, holding the engineering organization more accountable to what they're building with a go-to-market fit. He's really been driving a lot better alignment between engineering, product management and the sales organization. And I'm seeing that in -- that we have various, what we call, war room weekly meetings on things. And some meetings are once a month, some are weekly, where there are the initiatives in pulling together the stakeholders. And Sridhar is involved in so much of that detail. And then he's very focused on the go-to-market side, on really holding sales reps more accountable to a better process, better reporting, not just on bookings, but meetings and who you're meeting with in detail. Now pulling people in, sales enablement, that's another big piece. We have so many new product feature capabilities that we need to ensure the salespeople are equipped to be able to sell that. And a lot of that is sales enablement and training, and he's really driving a lot of that. And it's very positive what I'm seeing coming out of there. The guy is a workaholic.

I've heard that people think of Frank as kind of this hard-charger. Sridhar might be an even harder charger than I've worked with.

I've worked with Frank for 17 years. Sridhar is much more hard-charging. Yes. I used to joke when, for some reason, Jim Cramer used to like Frank. And I remember years ago, he'd say, oh, he has a cot in the office. And I'm like, I don't think I've seen Frank in the office [ after 5 ].

Sridhar just doesn't sleep. There's not even a cot.

Sridhar actually shares his full calendar with me, not just sharing that he's busy. I know exactly what he's doing. I can unequivocally tell you, that guy from the moment he gets up in the morning till he goes to eat, he literally has 8 hours for himself. He likes to sleep 7 hours, 1 hour to work out. And everything else, he books with anything for work. That is on weekends too.

Yes. So as you think about the demand environment and kind of on the most recent set of results, I'd be curious just what do you think some of the biggest questions or misconceptions are coming out of last quarter. Obviously, you had a quarter where you delivered upside to your guidance range, and you also raised the full year product revenue by, I think, the biggest magnitude you'd ever done. But what are kind of the biggest issues that you think investors might be missing?

Well, I would say what I think investors are concerned about is they see our revenue growth slowing down. And people saw we had a beat of 5.5% in Q1 and expected that we would be beating by a similar magnitude every quarter. And I've been saying this consistently since the time we went public that the way we guide the business is a 3% to 5% beat is a big beat, just like I don't get excited when we're 5.5%. I actually got upset once when we were 9.5% because we did a bad job of forecasting, that tells me. This is a consumption model. You're going to have variability on literally a daily basis by customers. And just like we beat Q2 by 2.5%, but what I would say is we gave a good guide for the full year with the raise. And that's really because we look at most recent consumption trends, and the consumption trends are very stable with our customers. And so I feel good about that. Now obviously, that could accelerate or could decelerate under the consumption model.

Yes, yes. One of the themes you also talked about on the GPU side, and we'll go into kind of the more gen AI-specific questions in a moment, but you did talk about some capacity constraints in certain regions in terms of GPU availability. Can you talk a little bit about those dynamics? And anything else beyond just the consumption volatility that investors should be aware of?

Sure. Well, first of all, we can get the small GPUs, the A10s, A100s, H10s and the H100s that are a little bit more difficult to get in certain regions. Where the bulk of our customers are, we have coverage. There is some demand from some of the smaller deployments, where salespeople are asking for them. And I'm telling them, let's see the customer demand before, not just you asking, before we commit more dollars to that because the problem with the GPUs is you're making a 3-year fixed commitment. It's under a reservation model and whether you're using them or not, you're paying for them. And that's hitting our margin. I personally -- and by the way, those customers that are in those regions, you can still -- you can -- we have the ability through your existing region to call into one of the regions where we have them, if you want. Now some customers don't want to do that because they want to make sure their data stays in their region. But I haven't seen a huge demand. When I see the demand from customers, we will deploy more, but I'm not expecting a lot of dollars in that. I think we have more than enough. And until we see revenue, we're not going to spend more on the COGS side. And we have a fixed amount in R&D, and that will not change that fixed amount in R&D.

Right. And some of the other moving pieces that I think are more applicable in the second half of the year versus the first half are things like some of the performance optimizations that you put forth into your product every year as well as some larger customers looking at Iceberg tables and open table formats. Can you just walk through those dynamics for the audience?

Sure. So first of all, ever since we've gone public too, we've said we estimate long term, it's 5% a year revenue headwinds associated with performance improvements. Most of them, we feel, come from our software. Some do come from the hardware, but not all the hardware improvements impact our software the same. And so this year, entering the year, we said 6%. We still see that happening. And a big piece of that is associated with what we're doing around what we're calling our warehouse optimization, where we can give customers visibility into their warehouses and utilization of that. So they can choose to pick smaller warehouses. A lot of customers still over-provision their warehouse now. Some do it on purpose to get better performance, but most don't need to do what they're doing. And that will have an impact on revenue. Now they have better visibility into that. There are other things like Iceberg. We do expect and we've always said we expect it to be more back-end loaded, more Q4, where some of our customers, now that we have the support for open file formats in Iceberg, where they're going to choose to move some of their storage out of Snowflake directly into open-format Iceberg tables. We've yet to see that happen yet. The only thing we've seen to date is we have about 400 customers that are buying with net new workloads in Iceberg format. They are smaller workloads. They're still just trying to make sure it has the same performance of Snowflake. But I do expect some of our larger, more sophisticated customers will move their storage to Snowflake. But the vast majority, I don't see that happening. And I think there's going to be a group of people, too, that will have -- be in a hybrid format. You may say, why would you do that? Even some of the ones that have been fine with Iceberg, they're like, there's a lot more work for us when it's done in Iceberg, and they didn't realize that. And then they have to take care of it. They own -- they have the data. They have to control the security and governance around that data.

Yes. One question we've recently got around Iceberg tables is if a customer decides to sort of move the storage layer away from Snowflake and go to Iceberg, is there -- what's the magnitude of that sacrifice on the query performance? Is it...

So in managed Iceberg, they're virtually the same performance. And the reason that is, is because through managed Iceberg, we're controlling the rights to Iceberg. And we put them in the right format so that they have the best performance. If it's the case of unmanaged customers controlling the rights to Iceberg, and a lot of times, there's more work to get it into the right format, and they just don't have the expertise that we have in doing that in our system in doing that. And a lot of customers who had originally said they want an unmanaged are realizing now, I want to do managed instead. And whether it's managed or unmanaged, that customer can still choose to use other query engines on that data, too.

Right, right. Okay.

And that's why we're obsessed with price performance because the cheaper we are from our price performance, they'll use our engine versus someone else's.

Right. I guess do you have a view on the customers that do go to Iceberg? Is it going to be sort of a mix of managed and unmanaged? Or how do you think that plays out?

It's too early. I do think it's going to be a mix, but I think the majority will go with managed when they realize how much more overhead there is for them to do it in the unmanaged way.

Right. It opens for them.

We've heard from some customers too, in big customers, the top 10 said, I am not interested in Iceberg. Our data lake is Snowflake and it works well. We are very efficient how we use it, and we're going to continue to do it that way. Why? So I don't have to worry about it.

Right, right. Moving to the topic of gen AI, maybe we can start high level and then get into the specifics. But what is Snowflake's gen AI strategy? We get a lot of questions of just how does a traditional data warehouse look in the world of LLMs and JSON documents? How do you sort of see yourself playing in this evolving gen AI landscape?

Well, there is an example. We have a logistics company that is using Cortex right now, and what they've done is they've developed a chat bot, replaced a legacy BI tool for their people that interact with the data warehouse. And that's a prime use case, and we see there will be more of that. We have an insurance company who's using Cortex to take all of their support cases, notes and everything, they get insights into what's going on in their customer base from a support standpoint. Those are the types of things that we're seeing being used in Snowflake right now with gen AI. There's a lot of things we're also doing on our own. This is not a customer thing, but it will help customers. It's using AI to help speed up migrations and make them more efficient using AI to help maintain pipelines for customers.

Right. So maybe going back to the products, and we're obviously a few months post the Snowflake Summit where there was a lot of product announcements, Sridhar's been hard at work. As we think about the AI-enabled products at Snowflake, what's sort of the vision and road map? You talked about Cortex. I think that's one of kind of multiple ways that you're sort of addressing this market. But just sort of give us the -- give us a vision and how investors should think about that product.

Well, Cortex is really just -- you have Cortex Search, you have Cortex Analyst. We think ultimately, Cortex Analyst is going to be the big thing, where an average person in the company can just ask questions in plain English, simple text of the data, and generate a SQL query on the back end and get the results back. That is the biggest thing that we see happening with AI and Snowflake.

Right. Right. And then on the LLM side...

We're using LLMs on that to do this. We're not looking to train LLMs and stuff. Things like that's just not interesting to us.

Right, right. But for customers that may have a preference on using something from Hugging Face or some of the -- I think you support a number of them.

We support a number, whether you want to use Mistral, Reka or Llama 3. We have our own Arctic model. There's other models as well too you can bring in to Snowflake. You use the open source. It's free. You use the Mistral or Reka, there's an upcharge for that because we have to pay them. And by the way, the AI stuff has much, much lower margins on that.

Right, right. As you think about these new products, Cortex, we talked a lot about that. How is adoption going? I know it's early, but how are you thinking about the adoption curve? And how those -- the timing of when those become a meaningful part of the business?

Yes. We have over 2,500 customers today actually using Cortex. A lot of it's not in full-blown production. A lot of them are just playing with it. We have a number of other noncustomers that are trialing Cortex as well too. I'm not including those. And so we're pleased with the uptick we're seeing in customers engaging with Cortex on our platform. I'm super excited about Notebooks and what that's doing. That's going to be a big catalyst for Snowpark, and we already have -- that will be GA later on this quarter. We already have 1,600 customers using that, and the feedback has been very positive. Sure, it has more -- there's still some feature gaps that we need to deliver on that. But our version one of Notebooks, we're very pleased with that.

Got it. Got it. And I think Sridhar has talked about seeing significant contribution from some of these emerging products, maybe as early as next year. I mean, how should we sort of think about the pace in which these ramp up? Because on one hand, I mean, 2,500 customers is -- it's, I think, 40% of your installed base or somewhat, which it's a pretty good start, considering these products are only out for a couple of quarters.

Yes. So obviously, we wouldn't have built these products if we didn't think that could have a meaningful impact, and it's -- I'm not going to guide to it now. It's still too early to tell. But we'll definitely be talking about that in Q4 when we finish the year. I think that's when it will have more of an impact. Remember, a lot of these things with Cortex, there's 2 components to it. There's the GPU component to generate things. But then actually, running the query is also driving a lot of data warehousing revenue as well, too. That has the higher margin.

Right, right, right. And I think, no contribution expected from these new products embedded in the guide?

We don't have anything embedded in our guide for this year.

Right. Even though you're seeing something?

The only thing we have in there is Snowpark, where we said at the beginning of the year, it would be roughly 3% revenue. And it is on track for the first 6 months to be that $100 million trailing 12-month revenue at the end of the year.

Right, right. As you think about Snowpark, we talked a little bit about Notebooks, which sounds pretty exciting. Are there additional products or feature improvements that you're looking to that can potentially grow that business? Obviously, I think $100 million is great, but it could be orders of magnitude larger.

I do think that. And we heard this from many of our customers is Snowpark is really going into the data science persona. And data scientists want a notebook out of the gate. They just want to have a notebook available with the product. We missed the boat on doing a notebook. We said we would rely on partners. Well, people don't want to go and procure a separate notebook. That's how Databricks got into a lot of the data scientists. By having a notebook, people started playing with it and then they started doing that. And Databricks has a very good notebook, by the way. Hex has a very good notebook, who's our partner that we partnered with. But it was clear from customers they want to have a notebook just available for free to them, and that will drive more Snowpark consumption, we feel. So I do think the faster we can get that into GA, it's going to really benefit Snowpark for us.

Yes. You brought up your friends at Databricks. So I'd be curious how you're just thinking about the competitive landscape. Obviously, they're private, so we only get selective disclosure from them, but they seem to be on par to adding similar amounts of new business this year as you are. We'll see obviously, but how often are you running into them? Do you view them as your #1 competitor? Just give us a sense on the competitive landscape.

So our biggest competitor today is Google, I've been saying that all along. From a technology standpoint, BigQuery is probably the closest to Snowflake, and I would say number 2 is Microsoft. Probably 3 would be Databricks and AWS. And what I would say is we coexist in many of our large accounts with Databricks, and they have done very well with the data science persona. I know they're talking a lot about their SQL warehouse, but I'm just going to -- I'm not disputing their numbers they give because I don't know them. I just don't see them in the SQL data warehouse space because they even quoted a bunch of customers that are saying, well, every one of those people you quoted is a Snowflake customer and they're growing.

Right, right. I guess I'm a little surprised to hear you say Microsoft is a bigger competitor than Databricks.

Microsoft, if they can get Fabric right now similar to they said Synapse, and they did not deliver on Synapse, but they still have a lot of stuff they need to do on Fabric. But that they -- if they can get that right, from what they're saying from a marketing standpoint, that will ultimately be a bigger competitive threat.

Yes, yes.

We still partner with them. And I know it's not ready today because they want to partner with us on certain accounts too. More, in particular, when Google is the competition.

Yes. I guess on the partnership front with the hyperscalers, how are those relationships going? I think Snowflake is still sort of a Tier 1 partner. In terms of the Azure Marketplace and AWS, you have a good relationship. But any changes just as they're rolling out Fabric in arguably more competitive products? Or...

Well, they've been rolling out Fabric for a while now, and most of their Fabric revenue is really their Power BI revenue. They call it Fabric now. So I would say our relationships with Microsoft are stronger now than they ever have been. We have a very good relationship with the entire sales organization there. And Sridhar has a good relationship with Satya. He's known him for years. And -- but it's going to be co-opetition with them. Data -- obviously, they still bring Databricks into a number of accounts. I think that may be is souring a little bit from everything I hear, but they have a big investment in Databricks. So it's a very competitive market. It's not like when I was at ServiceNow, there was no competition. There is real competition here with the hyperscalers and Databricks. And who knows who else is building a new mouse trap to come compete with us.

Yes. Yes. Definitely.

But it's in market. It's a massive market opportunity, and there's going to be many, many winners in this market. It's not a winner-take-all.

Right. Right.

That's why we're focused on the better price performance we deliver to customers, some more workloads will move to us.

Right, right. And I guess just going back to the last couple of quarters, we talked a lot about the consumption volatility. I think one thing that's been much stronger and less volatile, frankly, has been the bookings and commitments. You've seen very strong growth there. Can you talk about the strength? What are -- and a lot of these are multiyear, right? What are those conversations like with those customers? What are they sort of buying into from the platform vision that's causing them, in some cases, to commit 9 figures of business your way?

Yes. When people are choosing Snowflake, it's not a 1-year decision. This is why these are long sales cycles. These are -- we're replacing systems that have been in place for 20, 15 years, and so these are long-term decisions. And most customers and most global -- a Global 2000, the average deal size they start with is $150,000, and then they get that up and running. They usually buy a lot more next contract, and then they then typically enter into a multiyear contract with us because large organizations don't want to go to procurement every year. And I'll give you an example. There was a big bank, I remember. We signed a contract with them right when I joined the company. Got nothing out of them. A year later, we got a contract out of them. They barely consumed, but then we started doing a big migration. We did a 3-year contract with them, which is a pretty good-sized contract. And now they're running out of capacity, and they have the potential. They're a top 10 customer now. They grew 401% year-over-year last quarter, and they will continue to grow like that for some time because they have so much data. We're talking to them now about a potential 5-year contract, and they're open to that because they know they're going to be on Snowflake for a long time.

And on the migration point, I mean, there's still a lot of legacy business out there, whether it's IBM or Teradata. You got even some of the gen 1 cloud data warehouses like Redshift. You talked earlier about AI potentially accelerating some of those legacy migrations. How are you approaching that? Is there some relational migrator tool that you guys are introducing? How do you think about that?

No, no. Very good question. Before I answer that first one, Snowflake really started our -- what we were going after at the beginning was not the big Teradata migration stuff. It was really the first-gen cloud data of Redshift. We were replacing Redshift left and right. Then we got into Teradata and at [ Tiza ] and now we do a lot of Oracle, but then we also did a lot of Hadoop, a lot of SQL server. This one bank I was talking about, where we started with the Teradata, now we're doing a big Oracle migration with them. We have -- we actually acquired one of our partners called Mobilize, who was an early partner, who had built a lot of tooling around migrations. And we called it SnowConvert, where now we're letting -- our big customers, we actually just give it to them to use for them to do their own migrations of workloads. We're letting partners use it as well too. It's been very good at doing the [ Tiza ] and Teradata migrations, and we have another partner that we use a lot too or sub work out too that is doing a lot of the Oracle migrations and still doing Teradata migrations and stuff too for us. They have all kinds of tooling, and they're building this tooling with AI to make it more efficient.

Okay. So a lot of partner-led stuff on the AI front.

Correct. And by the way, we're giving the partners our tooling to SnowConvert.

Right. Right. Right. Okay. And if you could just sort of think about the biggest opportunities on the legacy migration side and maybe throw in AWS in there too, how would you sort of stack rank it? It sounds like you're seeing maybe more Oracle now than you had in years past. But what does that mix look like?

Well, there's still a lot of SQL server out there. There's still -- which is running on Hadoop. There's a lot of Hadoop out there. There's still a lot of Teradata out there. And it's a mix of everything. It varies by customer. I know there's one customer. We literally signed them five years ago, when I joined the company. I think it was in the first quarter I joined the company, and we're supposed to do a Teradata migration. That flipped completely just to a big Hadoop migration. And now we're talking to them about going back to, let's do the Teradata migration now. So there's still a lot of on-prem out there to do, and it's going to be many, many years.

Right. Okay. As you think about your go-to-market, I think you made some changes over the last few quarters just to incentivize consumption more than pre-committed bookings. Obviously, the bookings have continued to be strong. Several companies in like, call it, cloud consumption software have talked about some sales executions this year. How has -- how have those changes been executed at the sales force? Or how are you seeing kind of productivity levels and just overall how, considering some of the CEO-level changes and kind of the dynamics in the market?

Yes. So what I would say is the biggest change to our sales compensation plan this year was actually not the consumption. We started migrating more and more of their comp to consumption. Five years ago, that was the first thing I did. It was really more just saying, there's 35% of our reps, you are only paid on new logos, period. Because what was happening in the past up until this year is reps had a consumption quota and then they had a growth quota. And that growth quota could come from an existing customer or it could come from a new customer. And you got paid, each dollar you did, it was the exact same amount. Well, the reality is it's a hell of a lot harder to land a new customer than it is to sell more, get a new booking out of an existing customer. And so reps naturally gravitated to what was the easiest and just focused on our existing customers and ignored the landing new customers. Not to say they didn't do it, but there's a lot of opportunity. They never focused on closing. And so that is a big change that we think will have a big impact on our customer count going forward. On the consumption quotas for reps -- and then roughly 55% of our reps moved to predominantly just paid on revenue by customer. And we pivoted away from them calling out bookings on growth bookings, and what we want them to focus on are identifying new workload opportunities and the size of those workload opportunities. I will say, one of the things that we did make a tweak in the second half to also pay those reps more on the bookings as well too, we're always paying them on the bookings but it was a very, very small amount. I upped the amount we're paying to incentivize reps to also get bookings as well too. But I will tell you, bookings generally follow consumption.

Right, right. Okay. So sort of prioritizing the new logos, which got lost and -- because there's still a lot of Fortune 500 and Global 2000s that aren't Snowflake customers.

Correct.

Yes. Internationally, how is execution over there? I think you made some international sales leader changes in the last year or 2.

Yes. We -- so in EMEA, we hired a new sales leader. But then, I think it's about 1.5 years now. I would say, he's doing very well in the high end of the market. We're building out more of a commercial sales function in EMEA right now, we've been doing this year. APJ is doing very well. I would say Japan is doing well, Korea, Australia. We're actually doing well in Indonesia, and then Singapore is starting to take off. We have some really nice accounts. We actually did a -- I was shocked they actually did a $9 million cap one. It was a 5-year deal but with a telco in Indonesia, which I was shocked that they did. Usually, why I'm shocked -- it wasn't a shock we got them as a customer. I was shocked, typically from a new customer, they don't start that big.

Right. Or that long of a deal. Got it. Got it. I guess, as we think about data sharing, you mentioned that earlier as kind of one of the key drivers and key use cases. Any way of sizing that opportunity or, I guess, that contribution across your base? And do you feel like we're still sort of in early innings of that adoption? I remember that was a huge kind of killer app or killer use case around the time of the IPO. But is that still the predominant driver that you're having in a lot of these customer conversations?

Well, I would say in financial services, that is a key thing. It really leads to stickiness, and it also leads to new customers. It leads to new customers through a few ways. We have some of our financial services customers that are on Snowflake that are actually demanding their vendors deliver their data through a Snowflake data feed, which make data sharing -- which makes them, forces them to be on the Snowflake platform. We have customers that are building and financial service customers building applications on Snowflake for servicing their customers, which forces them to do stuff on Snowflake. Like Pfizer has built something that they're rolling out. What's Pfizer have? 3,000 financial institutions. Eventually, if they want to get the full benefit of it, they will become Snowflake customers. So -- and there's people like DTCC and BlackRock and State Street and BNY Mellon that are all doing things on Snowflake that want to do data sharing with their customer base.

Right. And in the disclosure, I think you give every quarter around stable edges. Is that sort of the best way for investors?

That's the best way. And the majority of those, the biggest users having the biggest number of stable edges on average is within the financial services. We think the health care space, there's a huge opportunity for data sharing, and health care is starting to take off. One of our 9-figure deals last quarter was with a health care provider.

Got it. Got it. I guess turning to profitability. Since the IPO, the margins have come up significantly, particularly on free cash flow. This year, you are reinvesting. We talked a little bit about the pressures on the gross margin side with GPUs. Also, Sridhar acquiring -- not acquiring necessarily, but hiring great talent, right, to build out these emerging products. How are you thinking just about the pace of profitability investors should expect here? And what are sort of the guardrails that you have in terms of ensuring that these AI contributions are ultimately going to be high margin long term?

What I would say, first of all, I don't think the AI is ever going to be high margin. I really don't, from everything I'm seeing. And it will be margin positive, but I don't think they're going to have the same margins as what you're seeing in the core Snowflake in terms of what -- it's a fixed amount. Unless there's revenue, there's no more that is going in. I have Sridhar's commitment on that and others. And I recognize the fact that our operating margins have come down from where they were last year, and I'm working right now on fixing that for -- now it doesn't fix overnight. But we are going to take steps that you will see expansion in our margin next year.

Okay. Okay.

Now unfortunately, I get people like Databricks that are just spending like frankly -- well, remember, actually, they have more salespeople than we have, and we're still a much bigger company than them. And I know they're burning cash like you wouldn't believe. I'm still very focused on free cash flow, focused on operating margin leverage. And I realize, too, that a lot of people are unhappy with where revenue growth has gone. And first, I want to stabilize that, and then we are working on plans on how do we reaccelerate revenue. And we have some thoughts, and we're doing some things about guiding to that. But that is what Sridhar is very focused on.

Right. Do you think on the revenue accelerate -- is that more just the contribution from new products? Or are there other things investors should be thinking about?

It's new customers. And that's where we're very focused on, getting new customers. And ultimately, it's those new customers that are going to drive future growth. New products are going to expand the opportunity within existing customers. So -- and the one thing we've been spending a lot of time on the federal space, that is a real upside for us because it's pretty small today.

Yes. Yes. Got it. And then on the efficiency side, I mean, obviously, the faster you grow revenue, the easier it is to expand margins. But what are some of the ways that you would look to optimize efficiency?

I'll give you one example. We made a conscious decision and it started last year to improve developer productivity. Well, to do that, there's a lot of different tooling that's required, and the switchover of that tooling will happen at the end of this year. Right now, we're running parallel systems that consume a lot of internal clouds. And my internal cloud spend, I'll just tell you, it's over $200 million this year that's going through OpEx. It was like $117 million last year. That is going to come down because of a lot of duplicate costs we're running there. And I've been in meetings on a regular basis on how we take that down.

Yes. Some optimizations of [ tiers ].

A lot of optimization. But we knew going into that, to switch over the tooling, you still have to run things in parallel.

Right, right. Yes, that makes sense. Probably a lot of transition costs this year with the new engineering team. We got about a minute left. I thought I'd turn it over to you just if there was anything else that you wanted to hit on or stuff that maybe we didn't ask you about that you wanted to leave with investors before we wrap it up.

Yes. No, I just want to say that it's going to take a few quarters to see the impact. And Sridhar is a new CEO. He is having an impact internally in the company. And I think it's going to take a few quarters before you're going to see in external results the impact of those things that he's doing. A lot of people think it just happens. It doesn't happen overnight. It takes time.

Right, right. Well, we look forward to that. Thanks for coming out to the conference, and appreciate everyone staying late with us. Thank you very much.

Thank you.