Cadence Design Systems, Inc. (CDNS) Earnings Call Transcript & Summary

I'm Vivek Arya, I cover semiconductor semi-cap equipment at BofA Securities. Really delighted to have the team from Cadence join us this morning. John Wall, our CFO; and Richard Gu, VP of Investor Relations. I'll go through a list of my -- actually, I will pass it off to Richard to read the disclaimer first, but then I'll have some of my questions. [Operator Instructions] But with that, warm welcome, John and Richard...

Thanks, Vivek. Thanks for having us.

Richard...

Yes, I'll talk quickly. Just today's discussion will contain forward-looking statements and Cadence will disclaim any obligation to update that. All right. Just telling people there are some seats here.

Thanks, Richard. I appreciate it.

So maybe, John, let's start at the top. Over the last decade or so, we have seen the semiconductor industry essentially grow at the 6%, 7%, 8% kind of pace. But during that time frame, we have seen your electronic design automation, EDA industry, consistently accelerate from 7%, 8%, towards double digit, and over the last few years, mid-teen. So maybe walk us through what helped that acceleration. And then more importantly, are these kind of growth rates sustainable, right? What are the key drivers of that?

Yes. Look, I think the trend has been pretty consistent for the last kind of couple of decades. Essentially, when you're doing chip design, the companies that are doing chip design, the R&D budget is you could bifurcate it into their spend on people, the engineers that do the design; and the tools that they use for the design. And I think over time, tools are doing more and more of the heavy lifting. I mean, there's just -- you can see the pace in increase in complexity of designs in chips. And clearly, population growth is not keeping up with that. So you can't do it manually. More and more chip design has become automated. It's like 99% automated. You can't do it physically anymore. And I think if you look at our share of -- like if you take the semi customers, if you look at our share of their R&D budgets, it's probably, right now, it's maybe 88%. They're still spending 88% of the budget on people and only 12% of the budget on tools. But with AI and everything, I think AI can do more heavy lifting for customers, and we've seen that. I mean, we've been dealing with AI for a decade. But if I go back, take your trends like a decade ago, maybe was it 500 engineers taking 5 years for a chip design. And now maybe 50 engineers can do it in 0.5 years or 6 months, maybe even a year. That's huge productivity improvements that we're providing through tools, and I think we're extracting more value for that. And I think it's -- the value proposition is easy to explain to customers but -- and there really isn't huge alternatives. So I think the productivity gains are there. It's easy to sell the value that we should outpace growth in semis consistently. And I do think that 88-12 will naturally have to become in a 75-25 and maybe even 50-50 over time. But because like I said, there's not enough population growth and I expect the growth in complexity to continue.

What are you seeing in terms of design starts? What are you seeing in terms of computing complexity going up? What are those early leading indicators telling you right now by the next 2 to 5 years?

I firmly believe that history will record this period in our time is like the design engineers that use our products are probably the Renaissance artists of their time. We pretty much just provide the paintbrushes, right? But our customers will paint masterpieces. Some of them are...

A very colorful description.

Well, they do. I mean, I'm always careful with our group at Cadence that we have to always remember, we're the paintbrush. But it's like -- it's great to see Jensen in the spotlight and what they're doing. But I mean, they're the true artists. It's true they can't do what they do without our technology, though. So there is a dependence on that. But I do think, I mean, Jensen would probably paint a masterpiece with a rock, wouldn't he? But I do think we provide tremendous value. And we cover off on -- in terms of trends, we cover off on -- like we're involved in everything. I mean, Anirudh likes to tell everybody this, a lot of the technologies just wouldn't exist without -- there's always some element of Cadence Design in any technology product.

You described the different eras and the masterpieces. But every era has a lot of skeptics, also. So let me give you a skeptical point of view, which is the EDA industry definitely benefits from AI. But is it fair to say that you have already seen a lot of those benefits? Because a number of the products, whether it was Hopper or Blackwell or the product at AMD or Broadcom, the design of these products started several years ago, right? So is it possible that you have already seen that big first phase of product design and growth could mature for some time before accelerating as these product and their monetization take -- how do you look at this, no pun intended, cadence of how the semiconductor industry is developing AI versus when you see the benefits of this come through for Cadence?

We'll see multiple benefits in terms of design phases. Initially, I think Anirudh likes to compare it to the Internet. Do you want to talk to that one, just the infrastructure being the product piece?

Sure. So overall, I think to your question, we definitely foresee that it's a big tailwind, right? And it's multiple waves. The wave 1 is always the infrastructure kind of build out. Right now, you're seeing all kinds of companies try to design their own custom silicon, right? AI chips. I think the velocity is also increasing, too, right? You heard about like NVIDIA and AMD, they're all coming to the 1-year kind of cadence already. So I think the proliferation of the design activity is not only just more companies, system companies are doing their own AI chip design, but it's also semi companies are doing designs and all kinds of -- various types of chips designs. So I think overall, I think if you look at the build out, we are already working with all these companies on those, so we're a beneficiary. And the second wave is, obviously, we have AI to our own product set, right? So we have 5 different AI products riding 1 common big data platform. We're also applying large language model to really help with the sort of spec to RTL conversion, which is a great use case. And the third wave longer term will be new verticals, new opportunities like bimolecular or science simulation. So yes, we definitely see this as a big long-term tailwind for our industry.

Yes. There's new applications all the time. What Cadence does -- I know we sell some hardware and we sell some IP, but predominantly, it's computational software, engineering software. And when you look at chip design, like 99% of that is automated already. When you go to system analysis, that's like -- so computational software applied to silicon is EDA. When you apply computational software to things like cars or phones or buildings, that's system analysis. And that's probably only 20% automated. And then when you apply it to data, it's AI. And when you look at applying Cadence capability and simulation capability in like the medical space, that's probably only 1% automated. So I think there's huge opportunities for this to grow for many decades, I would say. And I think it's just a natural progression. I mean, we've all seen the progression of this. Like you go back a decade or so, phones were becoming smartphones. And now we've seen cars become smart cars, buildings becoming smart buildings. And that's the kind of transition. What our customers really are looking for, the biggest problems that they're trying to solve, is they want to have one unified platform where they can do chip design, verification, packaging and board and system analysis kind of all at once. So that if you're changing something in place in route, then you can see what it does to your electromagnetic dynamics on the system analysis side at the same time. And I do think Cadence is probably the best positioned to provide that platform.

We'll, I'm sure, discuss more about AI. The one question, John, that we always grapple from the outside is Cadence has described its growth rate as double digit. Now double digit is a very, very wide spectrum. So as CFO, what informed you whether it's going to be 10% or 15% or 20%? Like how do you decide that, over the next 3, 4, 5 years? What are the leading indicators? What is on your dashboard to say where you will be in that growth curve?

That's a great question. Yes. And so I mean, we see double digits, we mean 10%, not 99%. So when we say double digit, we're really -- I always find that we try to take pressure off the team. I mean, if you listen to what -- you follow us for any length of time, you'll find we're very conservative and prudent in our outlook because the opportunity for Cadence is long term. We want people focused on long term and designing the best technology and solving the biggest problems for humanity, essentially. I mean, that's -- I mean, when people use Cadence tools, on the EDA side, the core tools, you're using them to optimize for power, performance, and area. We're going to need Cadence tools more and more to optimize for power, otherwise with AI, everyone is going to need like a nuclear power plant in their backyard to run all these things. So when we're kind of looking at growth rates, typically, what I tell the team is -- I mean, I'd love to have 20% growth across all product lines. I know it's impossible. But if I can get the growth, I want more profitability. So if it's a more mature product, then you want more profitability. And then I take the profitability and I'll give it to the opportunities where we have more opportunities for growth. And what's important to me is the -- how it all fits together. The model is we're trying to aim for double-digit revenue growth, right? But with increasing operating leverage, the 50% incremental margin is hugely important to me in terms of the model. But we started 7 years ago when it was like mid-20s. But we've been delivering every dollar of revenue growth since -- was it since 2016?

Yes.

Yes. Every $1 of revenue growth, since 2016 more than $0.50 of that revenue growth has dropped through to the bottom line. And as a result, of course, operating leverage has increased from kind of 25% range to 42% last year. But -- so with double-digit revenue growth and improving operating leverage, that double-digit revenue growth can give you kind of mid-teens operating income. And then with using our free cash flow because we spin off a lot of free cash flow. So we use half of our free cash flow to buy back stock and reduce the share count over time, and that can turn that into high-teen EPS growth or even in the low 20% range. If I look at our average over the last 7 years, it's been 23%, 24% EPS growth. And that's the way the model works. Now like I say, some of them, there -- the other thing is, you could bifurcate Cadence tools into interactive tools and kind of multi-licensed user tools. So for example, Virtuoso is a great franchise in our analog space, but every Virtuoso license needs a driver. So if you're a 100-person analog engineer design house, but we're probably selling you 100 licenses of Virtuoso, and growth there is going to come from price increases because we're not going to sell you 110 until you hire 10 more engineers. On the simulation side, one engineer could use 100 licenses. And if one engineer can use 100 licenses for simulation, one engineer could use 1,000 licenses, we just have to figure out how to make it more efficient for them to be able to use, get them access to compute power to do that. So the growth on simulation technology is hugely profitable, hugely profitable, because revenue can go nonlinear in comparison to your expense base because one engineer can use so many licenses. So what we tend to find is, over time, pricing on simulation tools comes down per unit, but the volume more than makes up for it. And then for -- from a forecasting perspective -- or actually, the AI tools where those are really helpful. You take like Cerebrus, which is probably the more advanced one because that was the first one we launched maybe 3 years ago. But Cerebrus allows a digital design engineer who, previously 1 digital design engineer to use 1 license of Innovus. Now using a Cerebrus cockpit, they can control 10 licenses of Innovus. So it's converting what was previously an interactive tool into a multi-license tool. And when we give metrics about our AI tools earnout, there's 3x the amount of revenue this year from the 5 AI apps than we had last year. That's not including the pull-through of the extra Innovus licenses. That's just the sale of Cerebrus. But -- so when it comes to forecasting, we're blessed in that so much of our business is recurring. But I mean, even this year, what is it? We're expecting about 82.5% at the midpoint of our revenue to be recurring. It's from that huge substantial base of customers that do 3-year license contract renewals. It's coming out of backlog. You know it's in there. It's very easy to forecast. This stuff that's more difficult to forecast is the upfront revenue because it's based on timing. That's more pipeline business. Like our hardware as a pipeline business, often, you really only have visibility in the pipeline maybe 2 quarters out. For IP, you have more visibility that like IP, we're expecting a lot of IP revenue in the second half because that's when milestones -- if we're to execute properly, will naturally trigger the revenue from bookings that are already in backlog. And it's kind of easy to forecast from that perspective because you know what's there. And as long as we don't have execution issues, it will deliver that much revenue. And generally, our style is we try to derisk anything we see. So last year, we saw Oppo in, 1 of our customers in China, that 3,000 engineers just decided to cancel a design project. So when the renewal came around for that, we went from selling 3,000 licenses to 0. Now what tends to happen is that will come back to us. It will come back to us over time. But because those 3,000 engineers will end up somewhere else. And whenever they end up somewhere else, those customers that they go to will have to buy licenses for them. So that eventually comes back, but that's -- it's easy to see it. But we saw that happening last year, that's why we were guiding that, hey, we expect China to dip this year because of that phenomenon. We know it will come back, but I'm always conservative about the timing of when it comes back, just to help the team focus on the long term.

One other topic that I wanted to touch on is your competitor has announced a fairly large acquisition, right, of Ansys. And obviously, I think we all kind of conceptually understand silicon and systems or electrical and mechanical, right, these designs are coming together. How does that impact Cadence's growth opportunity? Does the fact that you have these two large companies getting together, does that foreclose your growth in any way? Does that keep you out of markets? And then how does Cadence respond? Like will you have to consider a large M&A also? Or medium M&A for that matter?

Okay. So -- let me take the last bit first because we -- because the answer to that is no. We won't have to consider large M&A. So what we typically do is tuck-in M&A. Cadence, they're innovators. But the -- like we have over 11,000 people, maybe 11,500 people right now, but over 10,000 of those are engineers. But it's always been an R&D innovation machine, they're creators. And innovators lead, imitators tend to follow. But what I see in terms of what our competitors are doing is I think they've seen the value of what we're doing. This is typically what happens when you have a high-margin business, anyway, everybody wants to come in and do the same thing. It's very hard for them to do the same thing, though. But what we built with all -- with our own organic innovation plus some tuck-in M&A. I think in terms of tuck-in M&A, including BETA CAE, which is the recent one that we announced, and that was $1.24 billion, I think we -- that's probably doubled our M&A spend in the last 2 or 3 years. But we have -- we feel we have a complete solution there. Synopsys had to pay, what was it? $35 billion. So it's a big expense on their part to do that, and we see them offloading the SIG business. I do think they want to look more like us. But Cadence are innovators. They tend to -- I mean, and like I say, we're leading, and it's following Anirudh's vision. Anirudh's vision with Intelligent System Design strategy was that he knew that core business is really important, and core design, chip design is largely automated. But system analysis, when you look at the problems our customers have, system analysis, it's 20% automated. The medical field is probably 1% automated. We're heading down that path. We did OpenEye. I think we're -- even with Synopsys acquiring Ansys, we're still ahead, that they'll probably have to buy some digital biology companies to keep up with us, and maybe they'll see that in a few years' time. But we don't have to do big M&A. I think we prefer to make...

[indiscernible] do you think equivalent to that combination? Is it different? Or is it that it's a rising tide so all benefit?

I think it's better because it's kind of like -- I mean, it's old technology versus new technology for a start. But I mean, Ansys had a HFSS product for electromagnetic solving. That was our first kind of foray into system analysis, was Anirudh wanted to make his own electromagnetic solver using our simulation capability, and he created Clarity. And the way they measure the capability of simulation tools is based on how powerful your matrix multiplier is. Matrix multiplier is the number of variables you can handle. But HFSS could handle 50 million variables by 50 million variables, which is impressive and it was the leading product at the time. But our product handles 50 billion by 50 billion. And it's so far ahead, unfortunately, the market didn't need anything like that capability. But you get to a point with some complex designs that you have to break up the design to use HFSS. You don't have to break up the design for Cadence, and it's brute force. So the market is kind of coming to us anyway in terms of capability. The other thing is when you have that strength of simulation capability, you're able to simulate moving targets. And that's why we thought, well, molecules are moving targets. That's why we thought entering into the medical space and helping with drug discovery was a good science project for us, and we bought OpenEye to test that out. And that definitely has legs. So I do think when you compare the two, it feels a bit like Netflix and Synopsys have bought Blockbuster. It's like, great. I mean, if you're looking at the capability, the potential through the rearview mirror, that's one thing...

I believe Netflix is shipping DVDs or the...

Well, I just think that -- like our stuff is new, right? But it's new, it's new innovation, and that's typically what Cadence is all about. I mean, we attract the best engineers because we're committed to making versus buying.

So if I look at the next, let's say, 5 years, right? What would -- what scenarios would make your growth diverge from your competitors? Or do you think investors should not worry too much about. Like they are both good companies in a good space, right? Because one has a different mix of IP versus EDA and now this inorganic asset that you have a largely organic focus, but a different mix of IP. One has more digital than analog. So how are investors supposed to sort which company will do better over the next 5 years? How does that landscape shake out.

No, fair point. Both companies are great, right? And you can't design this new technology. You can't design chips without one or both of us. And typically, it's both of us. I do think, though, that there's been -- our advantage in making and being early is that we've only had to spend $2 billion, $2.5 billion putting the pieces together inorganically to get the Cadence machine operating and being able to solve the customers' problems. But when you're paying as much as $35 billion, I think that will suck some dollars away from R&D and into servicing interest costs. So I would expect that this is an opportunity for us in the short term to take advantage of that. I do think that we've been more profitable. We've been focused more on profitability. There's a lot of business in some -- I mean, in some areas of IP, for example, there's a lot of companies that are happy to give you business just to take some design work off their hands. And we've avoided what we call selling $20 bills for $15. But you can print any revenue growth you want if you're prepared to do loss-making business. We're not prepared to be loss-making business. So we're very thoughtful and data-driven in terms of how we price our business. And that's why we say -- that's why we're always cautious about double-digit growth. We're not chasing revenue growth for the sake of revenue growth. What we want is profitable and sustainable revenue growth. And ideally, I'd like to see that EPS growing 20%-plus every year.

Got it. Are customers willing to pay extra for AI, let's call them copilot-like feature, that are enabling, as you mentioned, greater productivity. Are customers willing to pay incrementally for that? Or do you think they're still in a trial stage?

No, they're definitely willing to pay for this. I mean...

[ Expect year over year, ] is it like 10%? The AI incremental benefit, right, from value and pricing perspective. How much is it, roughly?

That's harder to determine. I mean, we've announced like that we have the 5 AI tools, and we're very, very specific about we're not reclassifying revenue when we give you metrics about the 5 AI tools are now 3x the revenue. It was -- and 3x what it was last year. But, it's tens of millions. No, no, it's not hundreds of millions yet from those. But those will not be the only AI tools. I mean, there will be more AI copilots and everything that will come out over time. I mean, I think that's just an inevitability. And customers are willing to -- definitely willing to pay for it. And the more value they see from it, the more they'll pay. But it's a process in terms of extracting that value. We found that when we did the whole digital transformation, so you remember that, right? So Anirudh joined us about 10 years ago, we were way behind Synopsys in digital. And Anirudh made the brave choice to scrap a whole bunch of our digital tools and redesign them from scratch, blank page, kind of using all the latest in AI and machine learning techniques. And a lot of our simulation capability came out of that. But this 50 billion variables by 50 billion variables is a result of more than a decade's experience of doing that transformation. Synopsys never really had to do that. So what we found, when we did that digital transformation, eventually customers started using our tools, they saw that, hey, we're really competitive now. And I think if you take out maybe 1 or 2 of the large customers, that we're probably already 50-50. We're certainly more than 50% at TSMC, I would say. And with that whole transition in terms of digital and getting pricing right for digital took about 6 years, about 2 contract cycles to get that right. Initially, we probably underpriced it to make sure we're proliferated and customers were using it. And then once they saw the value, it was benchmarking effectively. I mean, most customers, they'll look at benchmarking our tool against the competitor's tool for their specific design. And whichever one comes out with the best power, performance and area metrics, that's the one to use. Pricing doesn't matter a huge amount because the economics of it is that if you're designing a chip that you're going to print out 1 million times or so, it's a small fraction of improvement in PPA more than offsets whatever you're paying for the tools. But like I said, that time, it took us two contract cycles to figure out the pricing for digital. And I think it's going to take us two contract cycles to figure out the pricing for these initial AI apps. But we will launch new AI apps as we go as well.

How are you looking, John, at kind of bookings growth for the rest of the year into -- or at least here? Because last year was stuck for the semiconductor industry, especially a lot of your analog customers. So as they are just starting to emerge from the downturn with all the inventory adjustments over, do you think you would start to see that in your bookings improvement in the back half? Or is that more a '25 factor for...

Well, I think that weakness from last year is already showing up in our recurring revenue, right? So because recurring revenue dipped from like 13% growth for the last number of years down to 9% in the first half. Second half, it's already recovering. It's like it's on track for 11% year-over-year. When I look at -- again, we're spoiled in that, most of the other CFOs, I mean, on our network would be like tell me that it's very hard to forecast because you're trying to figure out, you're looking at a lot of pipeline businesses. With us, more than 80% of the business, you know when the renewal is coming around, you know that -- because the licenses expire in a particular date, you know the customers have to renew by that date. And often, you're discussing the configuration within 3, 6 months in advance of that. So it's relatively easy to forecast, but also the timing of renewals impact bookings. But like if all of your bookings happened in 1 quarter and then you didn't have like -- and it was just add-ons for the next 11 quarters, you would have a huge explosion in backlog, and then you burn backlog for 11 quarters, and then you have a huge exposure in backlog again. But they're typically spaced out. When I look at the annual value of our backlog, Q2 was the lowest, like the lowest quarter for expiring annual contract value. So there's not a huge amount of renewals in Q2. So I expect backlog will dip slightly in Q2 as a result. But no, the second half is strong, though. But I would expect to finish the year higher than we did last year. I think -- what are we, $6 billion?

$6 billion.

Yes. So I think it will be higher than $6 billion by the end of the year. But it might dip. And that was $6 billion at the end of Q1, but it might dip in Q2 just by virtue of the fact that there's less contracts expiring in Q2. So therefore, you're going to burn another quarter of that backlog until the of the contract renewal comes around.

And then just a last question in the 2 minutes we have. So you're at the start of the upgrade cycle for your Dynamic Duo, Z3 and X3 around the corner. When does that start to show up in numbers? And how much of a content or ASP lift do you think versus Z2 and X2?

That's a great question. But in terms of capability, it's gate to capacity, right? So the gate to capacity, emulation capacity that Z3 has is double. Isn't it double?

Yes.

Yes. Of what is Z2 is. So the price typically doesn't double. So the gate, price per gate will come down, but the price of a Z3 is almost double the price of a Z2. So profitability should improve. Revenue will improve. The challenge with it is that it takes time to build the Z3s. But I'm quite happy if people want to keep purchasing the 2s, but they made such a big launch, and Jensen said such nice things about it. Everyone is very interested in it.

Yes. Blackwell was designed on Z2.

That's right. I think, sell more Z2s. But of course, any customer that has the ability to wait, and everybody know the Z3 is available, some of them are willing to wait for the Z3, which kind of creates a bit of an air pocket for Q2, to be honest, on the verification side. So we factored that into the guide already and I think that got last in the noise a little bit. And it's just our style. We're prudent. We try to derisk the guide, essentially. So we derisk Q2. And with the big launch or the hardware systems, we never took the hardware revenue up for the big launch of Z3.

But does it create some about trend potential for next year?

I think so. I think so. It sets us up really, really well. I mean, the fact that we're -- we've -- it's a showcase product for us. It's Cadence on Cadence. The motor on that is tremendous. I mean, we've used all Cadence IP and all Cadence tools to create our own product, but we've done it in half the production cycle of the previous system using our AI tools, but huge productivity benefits there. And I think it sets us up with the demands for the -- I mean, we couldn't keep Z2 on the shelf. It was selling out all the time.

Any supply constraints, do you think?

Not with our guide, right? Our guide is conservative.

How about over the next several quarters?

We basically need 7% to 8% revenue growth in the verification group to hit our guidance for the year. So we're not expecting to blow the light out. Now the thing is, it's 7%, 8% because they had a tremendous year last year, that happened. I mean, if it doubles in demand, we could run into some supply constraints. And we had that with the previous system. We ended up having like 6-, 7-month backlog at one point. And then we managed to clear through that. We also have a huge underserved community. There's a whole bunch of smaller customers that want access to that emulation capacity, but they can afford it. They can't afford to -- it's like purchasing an emulation systems like buying a Ferrari. And if you only have a use case to use it for a weekend, you don't want to buy the car. So we provide it in the cloud. And I do think there's an opportunity for us to use the existing inventory in the cloud to service those customers. But what it does from an accounting perspective, boringly, is it will flip it from upfront revenue to cloud revenue is daily. So that piece of it will become ratable. But we're delighted with the capability of the product and everything that Jensen said about it. And just the feedback we're getting from customers is very consistent. But I think it sets us up for multiple years. I think we're clearly leading in that space.

Terrific. Thank you so much, John. Thank you, Richard. Really appreciate your time. Thanks, everybody.

Thank you.

Thanks, everybody.