Informatica Inc. (INFA) Earnings Call Transcript & Summary

All right. Let's get started then. Hey, everyone. I'm Pinjalim Bora, SMidCap software analyst at JPMorgan. I'm delighted to have here Amit Walia, CEO of Informatica; and Mike McLaughlin, CFO of Informatica. Guys, welcome to the conference.

Thank you.

Welcome. Thank you.

So maybe just give a brief intro about yourself and when did you join the company, how long have you been in the company?

Sure. I joined in January, so I'm officially a full quarter in, about 4 months. I came from Fair Isaac, where I was CFO for 3.5 years. And prior to that, I was in your business, not as a research analyst, but on the banking side for 25 years, the last 12 or 15 of which have been in the tech and software sector.

For me, this is my 10th year at Informatica and fourth year as CEO. I was the President of Products and Marketing before. So as we'll go into the transformation, I kind of was involved in literally building the whole new product set and the whole new platform. So I know where all the bits and bytes are.

Yes. That's a great segue. So yes, Informatica, I used to cover it when it was public the first time around. It was a different Informatica at that time. A lot of things have changed. It went private. It actually built $1 billion business at that time period, right, went private, became public again. Help us understand kind of the evolution, right? What has -- what is the difference between today and what was?

Yes. So fundamentally different company. And I'll begin with some very simple numbers that I think folks in this room probably will appreciate. When we went private in 2015, we were a $1 billion company, $0.5 billion of license revenue, $0.5 billion of maintenance. And we had -- basically, at that point, the strategic imperative was, as I said, I ran Products at that time. Our belief was that the world of digital transformation will actually explode over the course of time. And 2015, digital transformation was in its early days. And our core belief was that there will be many, many, many more use cases for data management. Second core belief for us -- so that means a massive TAM. Second core belief was the world will be hybrid, but it will go towards cloud first. And our third belief was that a lot of stuff will need AI in the long term. And I say that because actually, we just came from a user conference and I showcased the journey. And the third one was that we started building from scratch. We -- the core thesis was, we're not going to take anything from the past and retrofit it. So we literally left our old on-prem product behind, from scratch started building a whole new set of products for these new use cases. In fact, started building out the cloud platform along with it, and that's a journey we took. And I'll give you 3 sets of numbers to anchor on. In this journey of 7 years, if I said, ignore the $1 billion business that was there, we went from $0.5 billion of license, ground it down to 0. Don't sell license anymore at all. That was the old product. From the new products that we built, we went from 0 to we've guided to $1.1 billion of subscription ARR this year. And in that, like I said, cloud was in its infancy in 2015, 2016. This year, we've guided to $600 million something of cloud ARR, growing at 35% from 0 in 2016. Now walk back. In aggregate on math, it may -- something goes down, something goes up. How many companies in the last 7 years have gone from 0 to $1 billion from net new products? Very few. Of course, the math looks out differently. And why? Because we built out this new cloud native platform called IDMC in 2016, runs many products on it, served the enterprise use cases, runs at 54 trillion transactions a month. And in that, we've grown new workloads, along by partnering with AWS, Azure, GCP, Snowflake, Databricks, whatever it is. That's the big journey we've taken. And that's what the new company is all about. It's fundamentally different. And I'll end with this. In the old days, we were only serving ETL on-prem workloads, ETL is less than 25% of the use case today. We serve broad swath of data integration, app integration, MDM, data governance, data privacy, many, many more use cases like we predicted in 2016 that the world will explode. So that's the new company, dramatically different.

That's a good data point, ETL is less than 25%, because a lot of investors still think of Informatica as an ETL company. I appreciate that. I want to go back to one phrase that you used during the conference, Informatica, which was obviously a pretty fabulous conference. I really enjoyed that one, with a lot of good AI announcements there. You use this phrase called AI-powered digital transformation. Maybe elaborate on what did you mean by AI-powered? And do you think data transformation has to be kind of a precursor for that to happen?

So -- yes. So our belief was that, look, and I said that data -- digital transformation was all about cloud and data. And in today's world, as we go forward, our belief is it's going to be AI-powered. Digital transformation, it's cloud data and AI. Now step back. There is no AI without data. There is no AI without data. And data doesn't mean just putting data in a database. There are many repositories of data. In fact, people tend to think, data is more fragmented today than ever before. What do you actually need for AI to work? Good quality of data, coming from many places that you can bring together, understanding the governance and privacy of all that stuff that's happening. That, to me, in a very nutshell is basically what data management does. So I think the question that you had asked me, and we talked about it at the conference, is that we are seeing that as tailwinds to just the raw work that we do. Simple example. If basically, customers want to make sure that they can understand a model better, you have to first bring data from many places at good quality. And then operationalizing that at scale is a massive task in itself, automating that through AI. We launched our AI CLAIRE, which is embedded in the platform, in 2017. 2017, it takes 2 years to hone an AI. So where we are? We already have tons of AI work going on under the covers with CLAIRE and we just launched our generative AI CLAIRE capabilities, CLAIRE GPT, CLAIRE Copilot at Informatica World a week ago. Tons of opportunities, we see that in the conversations with customers.

Yes, I have tons of AI questions, so I'll get to that. But on the cloud product, the IDMC journey that you are in, you're kind of leaning in on cloud now from a sales and marketing perspective. Are you leaning in from an R&D perspective as well? And is it -- I don't know if it is even fair -- but to compare kind of the product parity versus the IDMC -- self-managed versus IDMC cloud today?

So 2 different questions. But when we say leaning in, we only sell net new with cloud, barring exceptions like federal government, some far flung geographies like, let's say, Eastern Europe, there, there will be some on-prem products still to be bought. But for all practical purposes, we only sell product cloud. That does not mean that we do not support our on-prem products. Of course, anybody can say anything. We have mid-90s renewal rates. We have our dedicated teams that support our on-prem products. They run mission-critical sticky workloads, delighted with that. I think many of you, if you go around talking to your customers, or even you sitting in this room are customers of ours, where we're running operation workloads. So we have not end of lifed anything, and we will never do that. The end-of-sale things, so that's the pivot to cloud, which is where sales is this year. We obviously have a certain amount of dedicated sales and renewals on on-prem. And core engineering is all about cloud, supporting all -- some of the small amount of core work that has to happen on the on-prem. That's the premise of that one. And we see that. I mean we just came out of a Q1 where our cloud business grew 41% year-over-year. Coming out of a Q4 that we grew 40% year-over-year. So we feel pretty good about cloud.

Yes, For sure. The -- so the other part of the cloud journey, I guess, is the flexible consumption model, right? And you have transitioned that business model from perpetual to term subscriptions now, this flexible consumption model based on IPUs. Maybe -- we actually talked to one of your customers who said it massively -- he used the word massively -- reduces the friction of using the cloud platform. Maybe bring that to life a little bit. How does it reduce friction? And maybe, Mike, I can bring you into this conversation, talk about a little bit on the rev rec side of things, how does that change the model or not?

So I go with the strategic part of -- Mike, please share the details. Tremendously different and simplification. And I'll tell you why our consumption model is very different, and Mike will cover the rev rec part of it very clearly for you. Most of the companies are single product companies and their consumption is, for example, hey, any data consumption up and down, right? Ours is a platform IDMC that has data integration, application integration, data covenants, data quality, master data, [ mass ] so on and so forth. What we are telling our customers is that if you buy a consumption unit, which we call IPU, Informatica Processing Unit, you buy a dollar of it, you get everything on the platform. Let me repeat: everything on the platform. So I'll give you an example, what does it mean, which is what the customers are loading. Customers, let's say, they buy 500 IPUs. I'm making it up. You may want to begin with their data warehouse use case of ingesting data in a Snowflake, get going, sure. Suddenly you realize along the way that, you know what, now I want to add data quality to it. In the old days, you come back to Informatica. You again have a selling discussion, you cut another PO, it takes so much time that gets lost in this. You don't have to come back to us at all. You have data quality, go at it. Oh, by the way, I was doing ETL, I want to do ELT. Go at it. Oh, by the way, I am running an operational data warehouse, I forgot to do governance on top. You have it, go at it. You don't have to come back to us at all. What we are focused on is driving adoption. In fact, our customers are also learning that the world we've changed for them is so significant that simplification is dramatic for them. They are also learning in all candor. We met -- Mike and I met the CIO of a very large bank that bought thousands of IPUs from us. And it took them time to understand: so you're telling me that I have this? Yes. And I have that? Yes. And I have that? Yes. And where the conversation is going now is that, so all of those little tinkerings I were doing on the side with some experimental stuff, I can just move it here? Yes. And we invested a lot in customer success. I shared that in my Informatica World keynote, customer centricity. We have 95% renewal rates. So tons of simplification there. Now I'll let Mike also explain how it helps our revenue model as well.

Well, to start, I would highlight that when Informatica designed the Informatica Processing Unit as a measure of pricing, the company was very careful to learn the lessons from those who had gone before us, in terms of switching to a consumption model from a subscription or a license model. And some of the things that were observed was, in pure direct drive consumption models, the bill varies quarter-to-quarter, month-to-month. And the metering units for those are often pretty technical. And the user can, in surprising ways, all of a sudden run way over what they thought they were going to spend. That's bad for the customer, not bad when it's going in that direction for the company, but then they overcorrect and they cut their usage. And it creates this volatility in the bill for both the customer and the company, which really isn't good for either part. Furthermore, some companies, when they establish those pricing models with those quite technical units of metering, the connection between your bill and the business value wasn't as clear as it could be for the people who are making the buying decisions. And so the IPU does a couple of things. One, talking about the second aspect. First is, when you buy IPUs there's a rate card for what you spend in terms of IPUs, it's like spending your beads at Club Med, for data integration or data quality or whatever other service you want to consume. And the metering, the rates on that rate card are very thoughtful, to make it clear to the customer the business value that they're getting out of using those units when they consume them. And then the second, which is really important that goes to the variability, is that it's a minimum plus overage model at the end of the day. We enter into multiyear contracts with our customers, typically 3 years. Those 3-year contracts are a fixed number of IPUs for each year. And the customer can use up to that number of IPUs, and if they go over then they pay us more. But typically, they buy so that they're in the sort of 50% to 80% utilization range. So they have some surge capacity from month to month. So what that means is that we recognize -- we bill our customers a year in advance. So for the 3-year contract, we build them 3x at the beginning of the year. And we'd recognize that revenue ratably over the year without variability, unless the customer goes over their usage. So it looks from a rev rec standpoint just like a fixed SaaS contract for any other cloud-based company.

Yes, that's very clear. I almost feel like I need an IPU-based model for my TV streaming platforms. But -- okay, let's go to AI. AI is a big topic right now. I want to ask you a 2-part question. One is, how do you think generative AI or AI in general kind of changes this data analytics, data management space? In general, not just talking about Informatica, because we have seen Tableau talk about AI in the BI space, I think there are companies like ThoughtWorks and some of the other ones. And then second part is, how do you think of AI within Informatica? Like help us think through kind of the long-term vision, not just talking about what you have today?

Yes. So without doubt. I mean, while we talk about generative AI today -- that's why I actually was very open and sharing that when we started thinking about AI in 2017, we've been on that journey. And I'll give you some examples of where AI is still today and where it's going to go, because it's not like tomorrow generative AI, everybody is going to use generative AI to the core because it does a step function. But today itself, with CLAIRE within the products. So everybody has used Facebook, right? You know photo tagging? It's that same [ damn ] machine learning algorithm that is open source, that people can use it for anything. Facebook does not have ownership over that model. So we took the same thousands of machine learning algorithms and curated it for data management. So what does CLAIRE do? CLAIRE takes that algorithm and does data tagging. So for example, we can tag a data and say, "Hey, that's customer." And then CLAIRE will go and will find every time there's many, many more pieces of data and say, "Hey, these other pieces of data look like customer you don't know" and they will bring it to you. And you as a user say, yes, that's a customer, and you can tag it once and then it will immediately tag it across the whole landscape. That's data tagging, what is photo tagging? NLP, data quality. So data quality rules can be very technical. But today, let's say you decide to want a data quality rule around data governance, you can write in plain English. CLAIRE will convert it to technical language and give it to [ IT ]. That's already happening in the product today. Now with generative AI, we can go take another leap forward, and we just announced things like CLAIRE GPT and CLAIRE Copilot. And the vision there is twofold. One is intelligence, one is automation. Copilot is sitting with you. A lot of -- see, today, data has become democratized. We use that word quite liberally, but it is. Everybody wants to be a data user at varying levels of sophistication. You want to be a data user. Now you cannot do some of those tasks that easily, because they require some level of technical thought. Otherwise, we will get a wrong answer. Dangerous, sometimes all of us can be precise or incorrectly precise and that can give you very wrong results. So CLAIRE Copilot will basically sit next to you in the product and hand-hold you. I want to do this thing. It will tell you, hey, these are the 3 things we should do. It will automatically even show you what to do, it will guide you. And if you've done something, it will correct you. So that's a -- let's just say, an enabler or productivity enabler. And we know that a lot of new users are coming into data management. They're not that very technical. There's not enough time for enterprises to teach them. That significantly reduces the scale gap that is out there. That's one thing. Second is GPT, which completely changes the game. Where, for example, we showcased a demo where you can actually, through a chat command, do the same work that today IT does. You can literally say, "Can you bring my West region customers that are at a high risk of churn." Let me type that. I'm going to start bringing underlying the data sets for the West region customers, they look at what to define. It'll ask you, how do you define churn. You can say this is what I mean for churn. And through a very chat-like interface, it's going to do data-related work. And I give you those examples because we are scratching the surface of how much that will change the game in terms of what the core underlying data management is still needed. That's complicated work. But we are making it so simple that people can do that job. I would like to do it certain things that I don't [ want ] -- so those are the kind of things. And I believe that, that is going to be here to stay. We also believe that there's going to be a tons of, especially it's data, ethical things that we'll have to all think about, which is why governance will play a big role. Who's doing what? How are they doing it? You will have to have that. And we're already having this conversation with CDOs, on how do you bring data governance on top of some of these activities. So I think those are some examples, Pinjalim. I do believe that this is an area where tons will happen, has to happen. There's many, many more new ways folks are going to go do work, but these things are going to complement and do very complex technical work in an easy way. But the hard work we are going to take under the covers and make it simple for them to do, that still has to be done.

So it sounds like it definitely lowers kind of the learning curve?

Absolutely.

The friction to start using kind of Informatica, from a user's [ standpoint ].

Significantly. And by the way, I'll give you an example. I know some folks, large companies have thousands of developers running around. By the way, you all see that. After that, developers leave. You're stuck with -- for a lack of better word, half-assed code. What do -- you don't -- What is in it? What do you do with that? How do you productionize it? There is a risk in that. It reduces that risk, because you can now have with, a core set of people can run these things in a much more structured way, operational way, secure way, a governed way.

So Mike, does that potentially help on the net retention side over time? I'm not talking about next quarter. But over time, is that a kind of an upward bias?

Absolutely. That's why you have a true platform as we have. And when the platform contains all the powerful features, it has not just AI, but the ability to integrate quality and governance and access and you have the IPU model, which makes it as that customer said, massively easy to turn on and pay for without a new contracting event, without a new price negotiation, those new features, that net retention rate should take care of itself.

And have you disclosed any kind of monetization? How are you monetizing that CLAIRE Copilot, CLAIRE GPT? Or on the other side, the cost side, right, the gross margin side, is that as customers start typing in a lot of stuff, is that going to hit gross margin, maybe help us understand.

Well, for the most part, the CLAIRE functionality is built in. When you have data integration, you have the CLAIRE tools available to you. There's not a separate line on the rate card, which has 25 or 30 different services with IPU prices attached to them. There's not one that says CLAIRE, because CLAIRE is built in and everybody has access to it for essentially no additional cost versus the base price. In terms of gross margin, no, we don't expect it to have an impact. We actually have a very efficient back-end architecture that allows us to continue to realize what we believe are pretty admirable gross margins for our cloud business.

And to add to what Mike said, first of all, I mean I'll take 2 minutes. What we are doing behind the covers is there are many LLMs running around. OpenAI is not the only LLM model. There's Bard, and there's many, many more. Just like the thousands of machine learning algorithms we created for CLAIRE today, we're going to take all those LLMs and curate them for the next-gen activity. So we will actually pick the best LLM for a customer to use. Second is, you're not going to be -- if you're JPMorgan, you're not going to take your price metadata and put it in the Internet against open AI's LLM. You're going to run it in your own area, that's your differentiation. That's what we let you do. We're going to basically allow you to run that within your cloud to us. And the third one is, it helps us drive more IPUs because more customers can now do more work a lot easily. So for us, it helps us drive more IPO usage in the context of a product. Because CLAIRE GPT will be different in governance, will be different in data quality. So it allows the usage to happen. And behind the scenes, as Mike said, we've been very efficient about figuring out the right usage of the LLM, our gross margins, we've said that we know the magic number of 80 and we're not going to skip that 80 number.

Sounds good. That's clear. So coming back to another topic, which is macro, right? You had posted a really good Q1. I'm sure -- you had a conference recently, which was fabulous. You probably caught up with a lot of customers in the conference. What is your sense of kind of the macro environment at this point? How is the demand environment, the business confidence that you're seeing when you're talking to these customers?

I don't even know which macro you're talking about. Just kidding. I think it's a 2-part answer. You were there. I think the raw demand is pretty good. And what I mean by that is that, there isn't a single company or a single CEO or CIO I've talked to who said, "Oh, I'm not doing any of these data-driven digital transformation things because it's just not in my top 3 things to do. No, not heard anyone say that to me. I always give the example, and you saw American Airlines went in a war. During the middle of COVID, they bet on the platform because when they had to come out of COVID, they knew they have to still manage its customer churn and data governance and so on and so forth. So raw demand sits pretty strong. Having said that, I think everybody is facing the sense of, hey, buying cycles are elongated, everybody is paying more scrutiny to deals. And I think that's fair because nobody wakes up looking around the world and feel like the worst has passed. Every day, the discussion is soft landing, hard landing, software session is like 100 variations that you can see. But raw demand, you were there at the conference, everybody knows they have to go faster. If anything, the stuff we are doing around this allows people to do more with less. That's what they need, because they can't go around hiring more to do more. They want us to help them to do more with less. And that's what more productivity with AI, having a scale platform, having IPUs that you can use for any kind of use case, all of those things give them flexibility and productivity, that's what customers want in a time like this.

Yes. Yes. Understood. One question on migration, which we get a lot. Migration -- well, maintenance, [ parse into ] maintenance is about 1/3 of your business at this point. In the conference, you did talk about some newer kind of maintenance -- or, I should say, migration capabilities. How should we think about migration this year? Would you feel like some of these newer things that you're launching could actually accelerate the pace of migration this year?

So to give you a sense where we're coming from -- Mike, keep me honest with the numbers, if I get any of them not right. First of all, one of the things we've been blessed with, or maybe cursed with is, that we have a business in cloud that grew until 18 months before. 99.9% of the cloud business was all net new workloads. Let me repeat that, net new workloads. If somebody is spinning up Redshift, Synapse, Snowflake, whatever it is, 18 months ago, we said, look, we have this on-prem maintenance, now it is PowerCenter. And we're going to start the journey and a lot of customers wanted to move that to cloud. These are operational workloads. These are sticky workloads. And if people went on their own, they could have spent tens of millions of dollars to do that. And we didn't want our customers to do that. So we were working on the tech to help them automate that migration. So we automated a lot of it. We started that journey 18 months ago. And today, we shared that even on our earnings call, if you look at our cloud business, 90% of the cloud new business comes from new workloads, 10% comes from migrations. So from 0 it has gone to 10%, which is a good thing. We want it to be more, but it's a good thing because we're still not hostage to it. Having said that, by the way, when I say migration, people think, "Oh, it's not a lift and shift." You know, we're taking PowerCenter and migrating it to our new cloud. And at that point, it becomes a new workload. So we think the word migration, but it is a new workload. So that's the state of the land. And what we have been doing is more and more automation to help faster movement to the cloud. And what customers said, "Hey, can I move it faster? Can you automate it for me? Can you also give me the ability to risk manage the journey, so I can do it at my own pace?" Different customers had different pace. All of those things we took into account when we announced the new innovation around migration, where we said, hey, there's compatibility between the old PowerCenter and the new cloud, so on and so forth that we announced. So we're helping our customers migrate faster, it's in our interest. For every dollar of maintenance that migrates to the cloud, we get $2 in the cloud. So it's a good thing for the customers also, they're getting a brand-new better product and a platform, and for us also a good thing. So that's where we are. I mean there are more details we can go into. If I missed anything, Mike, please add.

Well, I think I would emphasize again that at the current run rate -- Amit mentioned that prior to 18 months ago that it was 99 -- I don't know if it was 99.9, but that it was essentially entirely new coming to the cloud. Now it's about $9 out of $10 come from new workloads or new customers, and only about $10 of every 100 new dollars in the cloud comes from someone who's migrating from either PowerCenter maintenance or our on-prem subscription product. That percentage may or may not grow as a percent of the total, but what we're really focused on is on the net new and what our sales force is really incented to do is to go out and find the net new workloads, to grow the pie as opposed to simply migrate. Migrating is good for us, too, because we get $2 for every $1 and that puts folks on the platform so that they can then consume more and increase our net retention rate. But don't get the impression that the growth strategy for Informatica in the years ahead is simply moving from the installed base to the cloud. We're generating a tremendous amount of net new, and we expect that to continue.

And [ barring ] the new product, the other beauty is that for the new platform. It's again, multi-cloud. So customers could be using AWS today, but you can use Azure, you can use Snowflake on AWS, Snowflake on Azure, you can use Databricks. It doesn't matter. The same IDMC can be used for any of those infrastructure or data clouds.

Yes. I have a few more questions, but I want to see if anybody else. Can we have a mic here, please?

It's a question for Amit. Actually, it may be a little left [ for you ]. How do you see competition from Kafka or the streaming data Confluent and so on?

Yes, I think they would like to be a competitor to us. I've never seen them be able to compete with us. I say that, again [ chi ], Kafka is a messaging bus. Let me repeat that. Kafka is a messaging bus. Messaging bus competes with what, ESB in the past. We actually embed Kafka under the covers from a streaming product. It's a messaging bus. It's not a data management platform. It's like saying that for, oh, for crying out loud, a storage layer becomes a data management platform. So that's the best I can answer. I've never seen, ever seen a Kafka messaging bus compete with MDM, data integration, data governance. Those are very different things.

And I would add to that, that we take streams from Kafka queues just like we take data from other sources, like transactional databases or IoT sources. And it's the data management layer that we provide. They're a source of data for our platform.

You talked a bit about sort of the role in Informatica in enabling enterprises to use sort of proprietary data in a secure way. Can you kind of touch on that a little bit more and elaborate on kind of what the role of Informatica is from a technical perspective in allowing companies like JPMorgan to run AI securely on proprietary data versus what the alternative might be?

Yes. I mean think of it as value, a simple example. Data is -- in fact, this company would have data sitting in many, many places in many, many different ways, I can't even begin to opine on how complicated that could be. You will never move your data in one place to run AI. That's the worst thing you could do. That's what everybody will tell you, move it to my platform so I can get a lot of data consumption. We basically -- IDMC is the metadata system of record. We have all the metadata that a company like JPMC could bring together. They know wherever the data is. Then you can pick whichever data you want to bring for what AI use case. And then under the covers, we basically have picked up all the LLMs that we will curate it on. And then we can run it on their version of the cloud, so it keeps it very secure for them, versus they have to take it out and put it somewhere else. So it makes it a ton more efficient for it to run versus just consistently moving around data all the time to run it somewhere. And we would have the power of all LLMs, versus this LLM or that LLM or this LLM at that point, you just don't even know which is the best LLM to use.

How much do the hyperscale cloud providers have products like yours that they would be competing with? And then the second question, is Datadog a competitor of yours?

Oh, no. Datadog is not a competitor, that's in the data [ observability ] space, different space. That's Dynatrace, New Relic, those are the ones. We don't ever see Datadog, totally different stack. Hyperscalers are partners to us, very close partners to us. Like in anything in the world of tech, there is always a shade of gray. So do I have a connector out of the box if I am Microsoft to connect to something? Sure you do. But if you look at it, we just announced, by the way, we've had long-standing partnership with Microsoft on Azure, long-standing. And we just announced all of IDMC as a native service on Azure, like native service. First of all, you draw down all credits by selling IDMC, Microsoft reps are extremely incented. And we've had many product integrations. Now we've like, literally, you whip up the Azure console, you see all of IDMC services over there. Similarly, we have the same relationship with AWS, with GCP and Snowflake and Databricks. So they've always been a great partner of ours and everybody will always have a feature or something on the corner, but that doesn't mean they're in data management. They want to drive consumption of their platform. They're more incented for us to help them drive that consumption for their data warehouse. They are more in data warehouse space. You'll have a Synapse compete with Snowflake, or a Databricks versus KNect competing on data governance or MDM where they do not have the capabilities.

I have a question on the infrastructure side. What impact to your business, do you think from the AI-driven telecom infrastructure or the hyperscale providers deploying the AI servers either distributively or centralized, because you talked about the data access earlier. So would that at least increase or decrease your data, your operating costs, or any other impact, if you could elaborate on?

I think a pretty complex question. I think Mike kind of touched it. You asked like multilevel question. For us, we don't see any cost addition to us. We -- the way it works for us is that if the way folks are going to consume our AI, and our AI is embedded in every product and every use case, means that they will be using more IPUs. The more IPUs they use, it basically drives our cost of goods sold, which is, by the way, very well set around the -- I mean we are north of 80, we're around 80 points. We don't see -- and if anything, I'd say [ tongue in cheek ], I have seen customers waste money on cloud platforms. I'll repeat it. We are very good about it. I see the waste. I see the waste. And what you see is amazing amount of waste enterprises do on cloud because I see it, we don't do it. You have folks, thousands of developers in your organizations, every day, they -- it's kind of like they'll go home, they leave the light open. They don't even shut down the servers they're running on AWS or Azure. We don't do that. We don't do that. I can just tell you a very simple thing: turn off the light, like I tell my own kids. So I can't go into too much, but we see the wastage. It doesn't come from use of technologies, 90% of consumption wastage goes by people just don't care. There are no controls. We have pretty good controls within our infrastructure.

Mike, I want to ask you one question on total ARR in the last 1 minute we have got. It seems like cloud on 35% growth will probably end up being the largest part of your 3 lines, right, subscription, maintenance and cloud. As you kind of get out of 2023, do you feel like you're taking a step down in growth in total ARR in the net new that you're adding. Do you feel like it's the trough to 2023, and we might see it stable or move up from here?

Yes. At the risk of providing 2024 guidance at this point, Pinjalim, yes, it should be the trough, that those 3 components are super important to understand. We have maintenance from PowerCenter, which is a highly profitable but legacy business is going to shrink at sort of 5% to 6% in perpetuity. We have on-prem subscription, which is the [ modern ] product, but which we're not selling any more new into, that's going to shrink as it turns naturally because we're not selling any new to it. And we have the cloud business, which is going to continue to grow at very attractive rates. And it's simple math between those 2 that are declining at slow rates, and the cloud that's growing rapidly. Just put your own spreadsheet together, and we should see '24 the growth rate inflect and start to rise as the high-growth cloud becomes a larger portion of the whole.

Awesome. That's a great note to end on. Thanks so much for the time, guys.

Thank you very much. Appreciate it.