Mike Scarpelli, CFO of Snowflake, needs no introduction. I think this is your first Goldman Sachs Technology Conference, right? Isn't it in person? Is this?
As for Snowflake, this is my first Goldman Sachs Tech Conference.
Thank you. Welcome to-
I did a few at ServiceNow, and even Data Domain, so.
There you go. Welcome to the first Goldman Sachs Communacopia + Technology Conference, CFO of Snowflake.
Thank you.
Thank you, thank you. Thanks for coming. I know you had your big user conference a couple of months back, lot of announcements, et cetera. So what, what is the management team's wildest aspirations? What are the aspirations for the company five years from now?
You know, we believe that we will be a meaningful player in being the data cloud. Really, we want to be a place where people can develop applications directly in Snowflake.
Mm-hmm.
We think, and a lot of the stuff that we have coming out in our pipeline over the-
Mm
... next twelve months, it will be either public preview or GA, things like Unistore or containerized services-
Mm-hmm
... streaming Dynamic Tables-
Mm-hmm
... Streamlit. We're pretty excited about what's happening with Snowflake.
Mm-hmm. Mm-hmm.
By the way, these are all things we've been working on for many years.
Mm-hmm.
Like Unistore has been a 4+ year-
Mm-hmm
... project for us.
Mm-hmm.
Iceberg Tables is another one that we've been working on for a number of years.
Mm-hmm.
You know, a lot of people ask us: Why can't you do these things quicker? Well, you know, there's a whole... When you're working at the infrastructure layer-
Mm-hmm
... you just have to get everything right. You can't be wrong.
Mm-hmm.
At the end of the day, we're one product-
Mm-hmm
... and everything works together, and everything you do at the infrastructure layer benefits everything.
Mm-hmm.
And so it just takes time. And this whole private preview to public preview, why do we do that? We learn a lot.
Mm-hmm.
Like Iceberg Tables, we came out in private preview with that. We learned a lot from customers-
Mm-hmm
... and we completely re-architected it.
Mm-hmm.
It just takes years to do, and so, you need to be patient in the world we live in-
Mm-hmm
... because you want to do it right. It's really important, too, because one of the benefits of Snowflake is it just works.
Mm-hmm.
It's very easy for our customers to use-
Mm-hmm
... and we don't want to lose that. We're not going to rush to get things to market-
No
... until we're ready.
Which of the, you know, you can guess what question I'm going to ask you. Which of these features do you think is the most exciting to you?
Well, I-
You probably tested these things, private preview. As a CFO, you probably have teams.
Well, I'm very pleased with the uptick we're seeing in Snowpark. You know, we really have only been-
Oh, okay
... in GA for Snowpark for Python-
Mm-hmm
... for about a year now. It still takes time. We've done a lot of these technical wins-
Mm-hmm
... with POC to show that we are cheaper-
Mm-hmm
... and it takes time to do the migrations. We have a lot of migrations that are happening right now-
Mm-hmm
... that will go into production, and I think will be meaningful contributors to the growth of what we're seeing already within Snowpark. But I'm really excited about containerized services that will be coming out middle of next year, and that really enables two things.
Mm-hmm.
It enables our customers to be able to bring their own large language models into Snowflake to fine-tune on their data, where they know the data is going to be secure-
Mm-hmm
... and stay in Snowflake. That's not going to get outside-
Mm-hmm
... of Snowflake. That's one of the biggest concerns we hear from enterprises-
Mm-hmm
... is training their models on their data. They're afraid of it getting into the public domain.
Sure. Mm-hmm.
I know highly regulated industries, think banking-
Mm-hmm
... like you, and healthcare and others, are super concerned about, so they're not rushing to do things. They're going to be very methodical of how they-
Mm-hmm
... think about the different use cases for generative AI in their businesses. The other thing, containerized services makes it easier for people to build applications-
Mm-hmm
... on top of the data in Snowflake, and then it enables, too, that independent software developers can build applications that can monetize it through our marketplace to make it easy for people to procure those applications in our marketplace.
A lot.
So I'm pretty excited about it.
That's just a couple of things, yeah.
And by the way, we've been working on this for a while, but more importantly, what's going to continue is data sharing. Data sharing is a key differentiator for Snowflake.
Mm-hmm.
That is a great lead generation for us-
Mm-hmm
... and we really see that taking off in financial services in particular.
Mm-hmm.
But, we're already starting to see a lot of customers coming to us because of what other customers-
Mm-hmm
... are doing on data sharing.
Mm-hmm.
We're seeing a lot of our customers who are actually telling their suppliers of data, they want them on Snowflake, or the people who buy their data, "We want you on Snowflake to do it through data sharing.
Network effect.
It's huge network effect, data sharing.
Yeah. Well, why would-
A lot of people talk about data sharing, but we're really the only ones who do it, where the data actually never moves. It always resides in the custodian of that data, always retains that data.
Yeah. Now, Snowpark, you said, for a product that's been GA for less than a year, it's doing quite well. What use cases are you seeing the most with Snowpark? What are people using it for?
A lot of data engineering use cases.
Mm-hmm.
Starting to see some data science use cases.
Mm-hmm. Mm-hmm.
We're replacing a lot of open-source like EMR-
Mm-hmm
... starting to see some Databricks as well, too, and stuff.
Mm-hmm. You, you said something that intrigued me.
What's that?
You mentioned a company by name there.
Well, I'm just...
So, since you mentioned them, you brought it up. Curious to get so they had a conference recently, and they talked about performance, price performance advantage for the SQL warehousing product versus versus another very large company.
You know, it's funny you say that. We never see their SQL product-
Yeah
... and they undercut the pricing on that to try to win stuff, but we just don't see it. You know, first of all, I think Databricks is a very good company.
Mm-hmm.
I'm not gonna say anything bad about them. What they do very well is in the data science.
Mm-hmm.
They still do a lot of ETL.
Mm-hmm.
You know, they coexist in so many of our accounts.
Mm-hmm.
We've brought them in.
Mm-hmm.
You go talk to—I know, and your bank uses both Snowflake and Databricks, and will tell you that this is where, how we use Databricks, this is how we use Snowflake. You go talk to most of our accounts-
Absolutely
... and they really see them as very, very different products.
Yeah.
We do, too.
I had dinner with our engineering organization yesterday, some of our senior leaders-
Yeah
... and that's exactly right.
I guarantee you talk to Nima, and he'll tell you that.
Yep.
But you go talk to AT&T or wherever, and they'll tell you-
Yeah
... that they're very-- they don't see them as competing products.
Yep. Yep, exactly. Now-
By the way, it's a massive market we're playing in.
Mm-hmm, mm-hmm.
It's not a winner-take-all. There's gonna be many winners in this space.
Exactly.
Yeah.
That's, that's good to know that. Good to know that. With those announcements, there are the stream of things that came out at your user conference, the Unistore. Let's talk about that. That, that could be a big, huge opportunity. The size of the relational database market-
Mm-hmm
... is just so massive, and it felt like you had taken a little bit longer to crack the code on that, where maybe it was a little bit tougher. Where are we with beta customers, Whatnot? How close are you with that technology to unlock that market?
Yeah, we're actually very pleased. We actually have 10 customers that are using it internally-
Mm
... in production today. It will go into public preview soon.
Mm.
Don't think of. That's gonna be very different. That's gonna be new use cases, and new development work will go into Unistore. I don't think of this as going in and replacing traditional OLAP data-
Mm-hmm
...bases. This is gonna be. It's not gonna have the performance of a Oracle.
Mm-hmm.
It's gonna think about, like, 10 ms latency is where we think the performance needs to be.
Mm-hmm.
And so we're pretty and it's gonna be more data-heavy, analytic-type-
Mm-hmm
... use cases-
Mm-hmm
... that we see it being used for.
Got it. Got it. Since you have a good feel for the pulse of the customer, I have to preface why I'm asking you the question this way. Jan Hatzius, who you might know, she's our Chief Economist, has been calling for a soft landing for quite some time.
Mm-hmm.
Everybody's been calling for a recession. He's been pretty spot on so far. Inflation seems to have come down, not to the degree that people might want, but it's come down quite a bit. There's a sense that things are stabilizing. I hear the word stabilization from practically-
Mm
... every single company. The last 18 months were horrendous. Rates went up massively.
Yeah.
None of us has been through really that big of a rate increase cycle in our...
Mm
... adulthood. Next 12 months, let's say things are normal, stable. Stable, okay. What do you think budgets look like for next year? As you talk to customers, are they looking to finally come out of this funk and spend after two years of being frozen to submission by the fears, unfounded, of a recession?
Yeah. You know, I can't comment in general about the economy and what's happening. All I can tell you is what I see in our customer base.
Mm-hmm.
We talked about on the call, we definitely saw a shift in sentiment. I would say it was really kind of July, we started seeing customers ring. You know, a lot of customers, during the uncertainty in the early part of the year, were just buying monthly as they went, when they had fully consumed under their contract, and their contract enables them-
Mm
... to just buy monthly.
Mm-hmm.
We call it just doing capacity purchases. We really saw customers who could have elected to do that-
Mm-hmm
... in July, that said, "No, I'll sign a new contract now.
Mm-hmm.
I know I have one customer that will fully consume, and it's one of our big customers, in January, but I know they want to do something this quarter, right now, are willing to do a longer term commit.
Mm.
So, the sentiment seems good with customers.
Mm-hmm.
I think it's them having more certainty in their businesses, not as doom and gloom as people were thinking maybe.
Mm-hmm
... 12 months ago-
Mm-hmm
... nine months ago.
Mm-hmm.
So, the pulse from our sales force is things are improving-
Mm-hmm
... right now. You know, in terms of, I think it's still a little early to see what that's gonna mean in consumption. We talked about it on the call, that we definitely see stability-
Mm-hmm
... now, in terms of consumption. We're not forecasting in our recovery right now-
Mm-hmm
... but definitely stability.
Mm-hmm.
You know, a lot of people talk about optimizations. Probably optimizations and AI are probably the two most overused words. You know, we've seen optimization since day one in our installed base of customers.
Mm-hmm.
If you go back three, and I think three, Investor Days-
Mm-hmm
... before anyone, we were talking about optimization.
Mm-hmm. Yeah, yeah.
You know, optimization-
We have a product called Warehouse Optimizer.
Yeah. You know, optimizations happen for a number of reasons. Some we drive those optimizations, and those are performance improvements-
Mm-hmm
... in the software.
Mm-hmm.
We think those are actually the more meaningful ones.
Mm-hmm.
Customers do optimizations, and optimizations typically happen for a few reasons. When you do a migration to Snowflake, you want to tag your data-
Mm-hmm
...that goes into Snowflake. You index it, everything, and as you add more data, there's turnover in headcount. People get lazy. They don't tag the data properly, the indexes, so when things get out of indexing, your queries take longer to run.... We go back in, and some customers will call that an optimization, where they re-index everything.
Mm-hmm.
So they clean it up so that your queries run more efficiently. That's a common one.
Mm-hmm.
People look at retention policies.
Mm-hmm.
People look at, "Can I write my queries more efficiently?
Mm-hmm.
And by the way, we learn, too. We had one of our customers, they figured out this way to do what we call a delta merge-
Mm
that makes running things much more efficient. As a result of that, we're now rolling that out in our software itself.
Mm
So all of our customers get the benefit it. Those are types of-
Yeah
- optimization, but every one of these things improve the price performance of Snowflake.
Mm-hmm. Mm-hmm. And so that, that's really good to see. So when you look at net expansion rates, let's look at consumption for a second. Did things continue to show the rate of improvement as you exited and as you announced your earnings the past couple of months?
I'm very pleased with what we're seeing in the consumption patterns.
Mm
of our customers right now.
Mm-hmm.
You know, I'll tell you, this is webcast.
Yeah, yeah.
You know, if I looked this morning, our top 10 customers-
Mm
... seven out of the top 10 are above what I was forecasting to be, three are below.
Mm-hmm.
Net-net, they're above.
Mm-hmm.
I'm pleased, and when I look at, I really look at the week-over-week to get a comparable-
Mm-hmm
'Cause it continues to grow and is growing very nicely right now.
Good. And, we cannot unfortunately mention this e-commerce company by name. We cannot, we just cannot. But there has been stuff in the media, I don't know if you can-
I can. Instacart.
I didn't say that. I just cannot say that. So any-
Yeah
... anything you want to offer to our clients-
Yeah
... by way of clarity?
It, well, if you actually read the f. 64 page, it's pretty clear, don't mix payment terms up with consumption. You know, what happened with Instacart is, I think they, don't quote me on this, but there was some quote-
We cannot say anything.
... about they paid $51 million, and then they said they estimate they're gonna pay us $15 million this year. But the reality is, the reason they paid us $51 million is 'cause they made two payments in that year because their contract, the way it worked, and we went in, and we had done a big optimization for them, and we had been telling them for a number of years, 'cause I know I was very involved with Nick, the CFO, on driving that optimization, and we took a bunch of costs out, but they still consumed $28 million, $28 million of Snowflake.
Mm.
Then it dropped to $11 million . That was the... We did the optimization, but now they're tracking back up. They'll consume $25 million, roughly, this year.
Mm-hmm.
It's starting to grow again, but that was through a massive optimization we did there, and a lot of that was because of a lot of inefficiencies the way they had set up-
Mm
- Snowflake.
That seems to be a pattern across your everybody's customer base. I mean, it looks like everybody... Amy Hood was here earlier today at Microsoft, and she said that we're reaching anniversary effect of optimization, the worst of the optimization.
You know, I'm not saying we're over with that, 'cause as I said, optimizations have always existed.
Yeah, yeah
... in Snowflake. It depends upon the customer.
Yeah.
Our second biggest customer is one of the most efficient customers, and they're very technical.
Mm.
And I don't see any waste in what they're doing, but some other customers-
Mm
... they just grow so quickly.
Mm-hmm.
What I would say is, there's a lot more trained people on Snowflake today-
Mm-hmm
... than there was three years ago, and a lot of the people that people hire have lived through optimization. So these people are gonna know the pitfalls of what not to do and what's best practices, and the more people we have trained on Snowflake, it'll lessen the effect of those optimizations.
Yep. Good. So-
The customer-driven optimizations.
Mike, maybe we can pivot gears a little bit to the competitive landscape, and maybe first on the hyperscalers. You know, how would you characterize the competition from the hyperscalers in core data warehousing, particularly maybe around GCP, BigQuery? You know, what does that look like today?
You know, GCP is definitely the most competitive BigQuery. It has been since day one. You know, AWS obviously is about 78% of our business is in AWS, 20% in Microsoft Azure, and 2% in Google. Google could be a lot bigger, but they are very, the most competitive.
Mm.
You know, I've been, we actually will fully consume on our contract before it expires in end of May, June 2024. I think we'll fully consume the five-year commit we, we had made in October, November. I'm working with them on a new contract. Hopefully, we can come to terms on more... I would, I would just say we're still gonna compete, but actually have met maybe better ground rules in the field, see what we can do, something closer to Microsoft or AWS. You know, Microsoft, it was huge that we signed a new contract with them because it's much better for the field, where their reps are now getting paid on the- it's not, it's not one for one-
Mm-hmm
... but they're still getting some compensation. As you know, salespeople are very coin-operated, so-
Mm
... it's important they get paid. So hopefully, we'll be in a better place with Google in the next 3-4 months.
Makes a lot of sense. And then maybe I'll never-
Oh, by the way, they're much more expensive.
Okay.
They know it, so-
Its price-performance is the distinction.
Way more expensive, and for us, if I look at my, it's not product margin, it's contribution margin, AWS and Microsoft run about 82-83% contribution margin. When I just look at what their costs are, GCP environment for me runs at 58%.
Wow.
That's a material price difference.
Yeah.
And then egress charges for GCP are crazy ... they're almost double what others are.
Yeah. Just one, one, one question, then, we can go back to you. Have you looked at Oracle?
We are actually spending a lot of time looking at OCI, for a number of reasons. Some of our customers would like us to be there.
Mm-hmm.
We've done the first phase of technical analysis, and it looks good.
Mm-hmm.
Stay tuned.
Mm-hmm. And Mike, on—
I think that would actually, 'cause I know they've committed to be the cheapest cloud for us-
Mm-hmm
... so we'll see.
On Databricks, right? So you talked about how there's still a lot of overlap in existing customer accounts, but it does seem like a lot of the innovation coming out of Snowflake, with Snowpark Container Services, getting more into the data science use case, predictive analytics. And then on the flip side, Databricks getting into the market with DB SQL. Like, are you seeing some degree of overlap in terms of kind of RFPs? Are you seeing them more frequently? How would you characterize the competitive landscape today versus three years ago?
Well, it's definitely more competitive today. A lot of that is because of what we're doing with Snowpark. It's always BigQuery, Microsoft, and not even as much AWS. When we're going into a Global 2000, looking at doing a big on-prem migration, Databricks is not there.
Mm-hmm.
Where we compete with Databricks is within our existing accounts on a workload-by-workload basis. We do compete with them, and we have to show that we are more price-performant than them.
This is a data warehousing, right?
In data warehousing-type applications.
Yep, yep.
The reality is though, is you need to really understand all of your costs of Databricks, 'cause remember, they're selling the software, and you still have your cloud costs that you need to pay for.
Mm-hmm.
It all comes down to price performance.
I think, you know, you talked a little bit about-
I would say they're very strong on the data science side.
Right.
They're very strong. That's probably their core competency.
Mm-hmm.
Data warehousing, not so much.
You talked a little bit about, you know, migrations to the platform from legacy incumbents, right? It seems like you're gaining a lot of traction with the GSIs and that capacity. So can you just talk a little bit about the runway you still have to displace some of these legacy providers? And who are you displacing the most?
So, we still displace a lot of Teradata. You look at-
The gift that keeps on giving.
And there's still a lot of Teradata out there. It's gonna be years and years before Teradata is fully shut down. I'm not even gonna say it's gonna be fully shut down. I don't know. There's gonna be some people who just resist. You know, we're still replacing Netezza. We replace a lot of Hadoop. We replace a lot of SQL Server.
Mm-hmm.
We've replaced a lot of first-gen cloud. There's been a number of failed Synapse implementations.
Mm-hmm.
We still replace a lot of Redshift, and we replace BigQuery.
Is it, is it because these implementations were smaller in scope, because the products and work is out of date, and then you go into a place where you're getting... 'Cause if you-
Couple things.
Right?
They hit scale, and they don't have the scalability. Redshift.
Yeah.
Synapse, just there's very, very, very few successful implementations-
Mm-hmm
... of Synapse out there. Even I think Microsoft has backed away. Now they're talking Fabric, which is gonna be another 2-3 years.
Mm-hmm.
Architecturally, that looks better.
Mm-hmm.
But still, that's a long ways before it's actually gonna be a product.
Mm-hmm
... out there. BigQuery is just expensive.
Mm-hmm.
Google's actually been raising prices with people, too. They kind of give them those cheap prices coming in, and then when it comes up for renewal, and once again, we show price performance on Snowflake.
Mm-hmm.
So Mike, you know, maybe we can talk about kind of this $10 billion product revenue guide, right? So you guys came out with this target back at the time of IPO, but since then, you've announced a number of new products, Snowpark Container Services, Document AI, Native App Framework, you acquired Streamlit. So it seems like there's a lot more levers for growth than you had at the time of IPO. So, so what, what's kind of your-- How would you characterize kind of your conviction in, in your ability to hit that $10 billion relative to maybe last year or even the year subsequent, prior to that?
You know, I feel good about our ability to hit that. Obviously, we're in a consumption model, and it, who knows what new technologies could come out within two years from now?
Mm-hmm.
I don't know. And it's not about... I'm not worried about the people we see in front of us. Usually, the real competition-
Competition
... comes from a new company that's kind of in stealth mode right now. I don't know. This was one of the biggest things I always worried when I was at ServiceNow, and it still shocks me today that there's really been no competition for ServiceNow. I feel good, especially with the products that we have coming out. The acquisitions we've done, I'm super pleased with, Neeva in particular, but Streamlit was a great thing.
Mm-hmm.
As I mentioned, I think it's been a home run that Meta open-sourced Llama 2. I can't believe all of our-
I'm sorry, what is a home run?
The fact that Meta open-sourced Llama 2.
Oh, yeah, yeah, yeah. The Llama. Yeah, yeah, yeah.
That's huge.
Yeah.
Um-
Why is that, though? How are you benefiting from that?
Because it enables our customers, without having to go through any type of procurement process, to bring their Llama 2 into Snowflake, to fine-tune it on their data, where it stays in Snowflake. And by the way, we think that model is pretty much the same as ChatGPT. They're probably a few months behind right now.
Mm-hmm.
That's been a real home run.
So that's a source of demand, the training, training models, with-
On your data, absolutely.
Hmm.
That's one of the biggest-
I thought that data was unstructured data. So are you talking about Snowpark or just the regular?
The regular data.
Really?
Yes.
Tell me, I mean, so, okay, I've not heard-
I'm not the technologist, but-
No, no, no, no, no. This, this is interesting because I didn't think that they, they would be training their LLMs on your architecture. Is that a trend you're seeing?
... in talking to all of our technical people, yes, I think-
Wow!
It is going to be very common. That was one of the things why NVIDIA wanted to, the partnership with them is really their NeMo software to be running in Snowpark Container Services, that you can bring your large language models and it'll work more efficiently with the NVIDIA GPUs.
Oh, the interview between Frank and
Yes, that's what it was all about.
It was crazy. I mean, I was-
Yeah
... blown away by what they were discussing. 'Cause we all, we all are trained to think that it's unstructured data, content generation, poetry writing, image generation, video generation, code generation, but this, this one's, like, looking at structured data-
This is looking at structured data, and then what I was saying, the acquisition, so super pleased with the Neeva acquisition because of the level of talent that we got there, but also bringing the enterprise. What was interesting was the enterprise search capabilities.
Right.
They were trying to build a search engine to compete with Google, which was gonna be, though, ad-free-
Yeah
... but it would be paid.
Yeah.
And by the way, this is the guy who built the whole ad business at Google.
Amazing
... for 16 years.
Yeah.
We're gonna be their enterprise search capabilities for customers to be able to find their own data in Snowflake.
Mm-hmm
... but also understanding listings in the marketplace.
Mm-hmm.
I'm very excited about that. That team under Sridhar, who was the founder, Sridhar Ramaswamy, he's running all of our AI and ML initiatives. The acquisition we did of Applica, that people don't talk a lot about, but that really enables. That's our document AI, which enables you to be able to take PDFs and contracts, and be able to search for terms and put it into a structured format. Being able to take simple text to convert it to SQL code, that's all part of what these guys have downloaded.
You're becoming a GenAI company.
you know, I'm not gonna use—I'm trying not to use the word GenAI... because every investor is talking about it, and it's so overused.
I know. Yeah, anybody-
Until enterprises really figure out what they're gonna do, it's hard to talk about it.
That, that's true. Any questions from you guys? Yes, Nancy. Let's bring a mic over to her. Okay. We got a full room for the last time.
Thank you. Hi, Mike.
Hello.
Can you just talk about how you're thinking about the pricing of some of these ancillary tools and products? Will they all be consumption-based, or will you think about that a little differently and embed that into the offer?
Good question. Everything is consumption-based. And that, and that's the beautiful thing. As we roll out a new product, there's no procurement process we need to go through because they, people buy capacity, and they can just consume based upon that. Obviously, we, in essence, we sell a credit, in which a credit equates to one hour on one node of CPU server. GPU pricing, we're trying to figure out how many credits it's gonna be per node of GPU, because it's much more expensive. But really, that's what we sell, and anything we do is gonna drive compute. And-
It's on the menu, you use your points, and you're good.
It's on the menu, and you can use your points. So it's the beautiful, like Snowpark today, it's not a new procurement cycle.
Thank you.
You had a question? Oh, you had a question. All the way in the front. Wrap it up in two-
Can you provide a little bit of color just on how it works with the AWS costs, and how it's bundled into the pricing, and over time, how you're thinking about the diversification between AWS, Microsoft, and Google? I know there was some announcement just on how there would be an interest from the company to have a little bit more diversification from Microsoft.
So, the way the pricing works is that's included in our cars The customer never, does not get... When you use Snowflake, you get one bill: Snowflake. We pay the egress charges if you're moving data from, say, many of our customers are multi-cloud. If you're moving data from Azure to AWS or vice versa, you just get one monthly Snowflake bill, and we pay all the costs directly. In terms of, right now, as I think I said earlier, 78% of our business is AWS, 20% is Microsoft, 2% is Google. I think AWS will always be our biggest just because of where they're at, but definitely Microsoft is growing. I would love Google to have a better go-to-market with us, because I think they could be bigger.
I think it'll always be AWS, Microsoft, and if it's not Google, it may be OCI one day. I don't know. We'll see, but that'll be a while.
Wrap it up on that note. Mike, thank you so much.