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

Okay. Good afternoon, everybody. Welcome to the Goldman Sachs Communacopia and Technology Conference. My name is James Schneider, I'm the semiconductor analyst here at Goldman Sachs, and it's my pleasure to welcome Cadence and CEO, Anirudh Devgan, with us today. Thanks so much for being here. We appreciate it.

Thank you. It's great to be here.

Before I'm going to get started with the disclosure, today's discussion will contain forward-looking statements, including Cadence's outlook on future business and operating results. Due to risks and uncertainties, actual results may differ materially from those projected or implied in today's discussion.

With that, let's get started. I'm sure we're going to be hearing a lot about artificial intelligence this week. We already heard about it yesterday and today. I think the EDA space is a place where AI is already being used. It can be even more beneficial to users. Maybe talk to us about how you use AI internally and what your AI offerings are? And when customers use a tool like Cadence's, what are the improvements they're seeing in either time to market, designer hours or other metrics?

Yes, great question. So first thing I want to emphasize is like, what we talk about is design for AI and AI for design. So versus other kind of software companies, I mean, the benefit of Cadence and EDA is that we are helping build AI also. So because most of the monetization right now is in the silicon and system build-out, right, whether it's NVIDIA or Google or all the hyperscalers, all the [ MAX 7 ]. So we are in a unique position that whenever they're building their silicon and systems, they use our products. And we are essential to the build-out of all AI systems. Now in parallel, we can apply AI to our software products to make them better, which is your question. And in that also, there are slight differences because one worry right now is, okay, is AI going to -- there could be some benefit, but does it cannibalize the software business, right? This is a question. I'm sure that's on your mind. So the reason it's different for Cadence is that the workload -- first of all, the barrier to entry is very high for EDA, and we have done this over -- we have a kind of vertical flow, which has been built over the last 20, 30 years. So whenever we work with even the big AI companies, it's normally a collaboration with them. The second thing which is unique to chip design and EDA is that the workload is exponential. It has been exponential for the last 20 years, but it will continue to be exponential for next 10 years. So if the workload is constant, I mean, like, I don't know, tax preparation or whatever, I don't want to pick on any particular area. Then if you have 10x productivity, then AI can cannibalize the -- but if the workload is exponential, so if you look at now, the chip size is 100 billion transistors or 200 billion for Blackwell, in next 5 years, it will be like 1 trillion or more. So it will be 10x bigger chips and the workload will be 20, 30x more. So we need the 10x productivity in AI just to keep up because our customers can't hire, like, 30x more engineers. So that's the other unique thing about EDA is that the workload is exponential. Now how we actually use AI, we have like 5 major kind of AI platforms. And the main thing is that we can make the -- not only the design faster, which you would expect. But more importantly, I think the monetization happens if we can make the design better, meaning the PPA or the power performance can be better. And I can give you a lot of examples in -- because AI is able to optimize over a bigger design space. So typically, what our customers do, they will run -- they're not running the software one time, right? They're running it and then they change something and they run it again, they change something. The software may run for 2 days, but the design takes 1 year or 6 months. So this -- in the past, there was no way to transfer knowledge from 1 run to the next run. But with AI, we can create all these models that can give a PPA benefit, PPA is power performance and area which are like 10% or 15%, okay? So typically, these days, technology scaling less they go from 5 to 3 or 3 to 2, the PPA benefit is 15% to 20%. So the AI tools are giving almost or half the benefit that you get from moving from one node to another node. So the benefit in terms of PPA is huge for AI. But to put it in context, the workload is exponential, and then we are also benefiting from the build-out of AI. That's why I think Cadence is in a unique position to benefit from.

Yes. Okay. Interesting. I think you have a really good and maybe it's even a unique view into your customers' road maps. And by proxy and the health of those road maps. What are you seeing where the levels of chip design activity in terms of design starts or tape-outs? And do you see it slowing down anytime soon or even accelerating?

Well, I think it's accelerating. And if you look at -- if I compare it to -- I mean, like, let's say, 1 year ago and beginning of the year, there was all this concern about DeepSeek and other things. By the way, I believe there will be multiple DeepSeek moments. Not just 1, okay? Because AI right now is like a dense multiply. It has to get much, much more efficient. And software will improve significantly to make it much more efficient. And this -- if you look at -- we do computational software for 30 years. If you look at the history of EDA software for 30 years, it has gotten much, much more efficient. So I think AI compute or the algorithms will get much more efficient. But at the same time, because of reasoning and other things, the amount of compute will still go up. When I talk to the big customers, they are saying like order of magnitude improvement in inference, but still the amount of inference is going up faster than that. So if I compare it from 1 year to now, I think I see even more commitment in the big hyperscalers to do their own chips and you're seeing that in the industry. And of course, the big companies like NVIDIA will do very well. And then with this physical AI like cars, drones and robots, companies like Tesla and other. So overall, if you step back, we see the semi -- so roughly, in terms of our business, 45% of our business is from system companies, and 55% is from semiconductor companies, even though NVIDIA, I don't know, is both semi and system, but we classify them in semi. So NVIDIA, Broadcom, all these companies, AMD, I mean, they're doing phenomenally well. And then the system companies are also doing their own chips. So I believe that the amount of silicon that is going to be designed from AI, both infrastructure and physical AI should accelerate in the next few years.

Very good. I think at a very high level, most investors understand that your revenue is generally tied to your customers' R&D levels at a very high level. But give us a sense about how much incremental wallet share you can drive within your customers, whether that's head count and your ability to price over time for the value you're creating with AI or otherwise?

Yes. I think -- I mean, we are, of course, tied to R&D, right? We are engineer -- we say engineers for -- we make software for other engineers. And that's the reason that the barrier to entry is very high. I think barrier to entry is lower in some of the other non-engineering software. And also, we invest significantly in our own R&D, right? So we -- so our goal always is to maximize revenue growth plus operating margin. So if you look at last 3 years, our CAGR revenue, CAGR about 15%. And this year, our margin is about 44% and operating margin. And it goes up incrementally. So if you look at Rule of 40, we are in the high 50s, okay? And so there are very few companies who can achieve that on a sustainable basis. But I believe Cadence has done that last 5, 10 years can do it going forward. So it's a combination of margin and revenue growth. And revenue growth should happen because of all this. Now pricing is a part of that. And -- but I think 1 -- again, good thing about our position and the industry because of Moore's Law, and we can argue Moore's Law is dead or alive, but one part of Moore's Law is true that the chips will get bigger and bigger. Whether they get faster, it's in question. But when you go from [ 7 to 5 to 3 to 2 to 1 ], the complexity of the chip will increase. If the complexity of the chip increase, then they need more software and hardware to design our chips. So the demand should go up. And pricing, we always work collaboratively with our customers. If we can deliver value to the top 50 companies, they will always pay us. They don't have any shortage of money. We just have to show our value for them.

Hopefully you can get a bigger slice of that.

Yes.

I mean, one thing that I -- that's interesting. One thing I thought I wanted to talk a little bit about is China, because that's been an area where there's been a lot of noise, both with respect to the U.S. export control restrictions. But also sort of your forecast, I think you now think that your China revenue can grow just a little bit this year, I believe. But maybe talk about sort of the impact of those export control regulations. Was there anything agreed to by -- between you and the government in terms of your restrictions on your business going forward? And what impact are you seeing from your customers in China going forward?

It's a good question. And there are so many details in that question because I don't know if people know there was like an EDA ban for like 6, 7 weeks that got lifted in early July. And then there's the general export control that -- so overall, I do think that China should be stable and improve barring this 6-, 7-week high does that happened. I mean, overall, our China percentage has come down. So right now, we are roughly 10%, 11%. A few years ago, it used to be 17%. And that's -- I mean, China, this year is not growing much, but overall has grown, but the rest of the world has grown faster. Now over time, I think this is sustainable. It may come down a little bit more. But we are -- the good thing is we are very diversified geographically and product industries, right? So -- but China, right now, the demand is good. They're also big in -- you know this anyway, they are very big in physical AI. If you look at 5, 6 big car companies, they're all designing their own chips, trying to do self-driving. They have like 100 robotic companies, and they're all the regular phone companies and data center companies. So I think the regulatory environment, I would say right now, at least what I can see is stable, barring that 7-week thing?

Yes, yes. So no long-term impact, but just sort of an overall stable profile?

Yes, right now. And China customers are also pretty -- during the 7-week ban, they were pretty measured. So right now, what I see in Q3 is back to normal in China.

Great. I want to talk about sort of physical design for a second and that sort of overlap with your business been topical. One of your competitors bought ANSYS recently in terms of physical design simulation, you've done some tuck-ins in the space over time. And last week, you bought the design engineering unit from Hexagon. Maybe talk about your overall capabilities in physical design and simulation on a competitive basis. And sort of how you expect to kind of drive that business going forward and the importance of the synergies you see between sort of EDA and the physical simulation design.

Yes, we have been doing this from -- I've been doing this from 2017. So you can blame me for all the consolidation that is happening. The thing was, I was going to be CEO, and they said, okay, what is the future of EDA? Okay, this is 2017, 2018. So my opinion -- and this is a very different time at that time. 2018 in the semiconductor industry, all the consultants would come tell us that there will be only 10 companies left because Broadcom will buy Qualcomm. There's a massive consolidation will happen. So -- and if you look at EDA, what is our core strength because everybody wants to grow, but you have to grow in your core strength. So our core strength and my background anyway is EDA and numerical analysis is computer science plus mathematics. So EDA is very numerical, mathematical software. And applied to silicon. And we are best in the world of doing this kind of software. So if you look at all these places out or similar, very numerical complex software. So if you -- so that's our core strength, what I call computational software, which is CS plus math. And then if you look at the word around us, and this is obvious now, but it was not obvious in '18 we look at the world in 3 concentric circles. So there's a silicon circle, then there's the system circle and then there's the data circle. And a perfect example is electric car. You have all the navigation data then you have physical car, which is electrical plus mechanical and the silicon that drives the car. So then you take computational software, which is our core strength, and you overlay that on these 3 concentric circles. So computational software applied to silicon, that's EDA, right? Computational software applied to system is system simulation. So thermal, electromagnetics, aerodynamics, that's why I entered all this space, and I think that will in 2018. And then computational software applied to data is, of course, AI. And a lot of the algorithms are very similar, numerical analysis and algebra, things like that. So from 2018, we are doing EDA plus what we call SDA, System Design and Analysis and AI. And that's not going to change. And we thought it's better to do it organically or mostly organically because the margin profile is better. EPS growth is better. And we do some tuck-ins but our culture always is organic first. So I don't believe that will change now. The 2025 version of that is different than 2018 version. So what is different in 2025 versus 2018? Number one thing the value of EDA is much higher because now because of AI is driven by infrastructure. So that's why we want to make sure we refocus on EDA and IP, and that's what we have done. And we have the broadest portfolio in EDA, and we are clearly well positioned in EDA. Now in systems, the most exciting thing for me, systems, like I mentioned, is physical AI. The future, right? You don't want to miss these big trends. So that's why we bought MSC from Hexagon because they have 2 great products, Adams, which is the #1 multibody dynamics, which is a robotic simulator and Nastran, which is structural. So we are pretty well positioned in systems. And the other exciting thing of systems is 3D-IC when the chip and the package and Cadence has the majority share with Allegro and 3D-IC based design. So I believe we can grow well in systems. And then, of course, AI, we are doing a lot. And the other thing that's different in 2025 versus 2018 -- So of course, I've talked about infrastructure AI, data centers, all those things, Edge AI. The second big wave, I believe, is physical AI, cars, drones, robots, so we want to be well positioned. Hopefully, in the next that will happen. The design is happening now, but the deployment will happen in the next 3 to 5 years. And then I believe the other big wave, third big wave of AI is sciences AI. So physical sciences, of course, chip design, but also biosciences and life sciences. So about 2 years ago, we bought a company to do biosimulation and all that. So we don't want to be too early, but we don't want to be too late in that. So that's what I believe these 3 big phases, and we want to be aligned with those 3 concentric circles.

Fair enough. You talked about EDA and the core growth there. So maybe you want to build on that because I think EDA growth has been quite good, quite solid. But I think you point into some things, whether that's AI or otherwise, that could at least theoretically accelerate that growth rate. So I'm kind of curious, what do you see the levers of your core EDA software growth being? And do you think that growth rate can accelerate in the next few years?

Well, we'll see. I mean we are always conservative in projections. We'd rather printed than talk about growth rate. I mean, what we have done, like I said, we have done more than -- in the mid-double digits and the margin is in the mid-40s. But the growth rate in the future, I mean, the silicon content will increase I mean, the projections are pretty bullish, right? It's $1.2 trillion by 2030, $2.5 trillion by 2035. And both system companies will do a lot of silicon. I mean there's no doubt -- I mean, you can see that in the latest numbers. So I feel we are very well -- and the Moore's Law will continue for next 10 years. So we are at 3 right now. It will go to 2, 1.4 and 1. I mean all these big foundries can see the road map till 1. So each of them is 2, 3 years. So for the next 10 years, Moore's Law is alive in terms of area scaling. And then you need to do more design to get the performance out of it. So the complexity of the chips will go up. Amount of silicon deployed both in data centers, Edge, physical AI will go up. And the customers will spend -- if there's a $2.5 trillion market, the customers will invest in R&D. And with AI, we hope to get bigger spend of that. See, the other thing I'm watching is -- so we get a certain percentage of the R&D budget, right? So it used to be 7%, 8%, now it's close to 11% of R&D is going to automation. If the workload goes up by 30x but your head count only goes up by 2, 3x because hopefully, the rest is AI. Then as a percentage of R&D going to software and automated should go up. So not only I expect R&D budgets of customers to go up if the market is going to be $2.5 trillion, but there is the opportunity for Cadence to capture more of that R&D budget with AI and automation.

Makes sense. Now competitively, you and Synopsys have had different advantages and different steps in the process flow and design across different tools. Where do you feel you're ahead -- most ahead today and sort of what are the areas you're continuing to focus on and sort of the core EDA flow to continue driving innovation?

Well, core EDA, we are very, very strong. And we have the broadest portfolio, 3D-IC has like majority share. And I think in systems, I think we are well positioned to the growth areas of systems, which is close to the -- either close to the chip or all the way to the data center level. I think the things we have to do better competitively at the highest level is we haven't done as well in Intel and Samsung. So Cadence historically is very strong with TSMC and TSMC customers, but not as strong in. And some of these problems predate even before I joined Cadence, they are like 15 years old. Intel for the longest time didn't work with us in Samsung to some extent. So we have to do better there. And now there are a lot of changes in both those companies. So we are working to -- and the second area we have to do better or I intentionally didn't invest as much as in IP. So our competitor is much, much stronger in IP now. IP is not as profitable as EDA. That's why I didn't invest as much and intentionally focused it on EDA and SDA. But I think now in the last couple of years, we invest more in IP. IP, these premade kind of design blocks. And because of AI and 3D-IC, there are opportunities in AI. So in IP. So for us, we need to keep our strength in EDA and SDA and AI, but add focus on IP and do better at Intel and Samsung.

Very fair. On the point about IP, maybe talk about how you expect to do better? Is it you're going to do more M&A to the extent it's available? Are you get a new more organic development of IP and just sort of think about kind of like how much you think the overall IP growth rate can lift long term?

Yes. IP, like last year, we grew 30% in IP. This year, I think it's -- I mean, we haven't finished the year, but I expect good growth in IP for the full year. And the growth is a combination of organic and anyway, we like organic asset. But we did some acquisitions in IP because some companies like Rambus, for example, didn't want to focus on IP. They wanted to become a product company. So this is great. And same thing with ARM, we bought Rambus HBM business, which is a great business. And recently, we bought Arm's Artisan business, which is their foundation IP, which ARM is a great partner of Cadence. So that was a good acquisition also. So we have some acquisitions. But in general because of chip-to-chip interconnect like all this UCIe from chip-to-chip, DDR, memory access is a big thing, PCIe. So those 3 areas we have done organically and they're doing very well. So if you have DDR, PCIe and UCIe organically and then HBM and Foundation IP inorganically together, I think, is a good portfolio. And we are not too big. We are just -- we are about -- roughly $700 million, $800 million in IP, which I think is a good size. And also part of it is Tensilica, which is very profitable. It's like software like margins. So I want to grow IP but at a good profit margin. So I feel right now the size and the margin is good that we can grow that.

Fair. Maybe just to sort of ask you, you talked about systems -- sorry, system simulation, physical design and simulation. Longer term, if you think about the 10-year plus time horizon for the company, do you think it's going to be more of a kind of a systems design company over the long term rather than a semiconductor design company?

No, it will be both. So because people say, like, even in the customer mix, we say 55%, 45%. And then the question is, oh, will the system guys do so well that you will become more system. It's very difficult to predict because the silicon guys do so well to look at NVIDIA, look at Broadcom, I mean they're actually top 10 market cap companies, 3 of them are semi companies, NVIDIA, Broadcom and TSMC, all great partners of Cadence. So it's very difficult to predict. I think both will do well. I think finally, the value of silicon is realized by the market and the customers. And they realize more that none of the silicon is possible without Cadence, okay? So that's my job to do a better job explaining that. But I think both of them will be there. So I think semi will be strong. And the Intel will come back, hopefully, Samsung. So -- and then on the hyperscalers, they will do more silicon. So -- and it's a combination. System in semi will be together.

Yes. I mean do you think you're actually enabling a lot of your systems companies to do more vertical integration?

Absolutely, absolutely. This would not be possible 20 years ago. And so we have a big role along with, of course, TSMC and ARM to make that happen. So you go back this a long time ago, when I was in IBM in late '90s, we were designed -- we used to do a lot of silicon design those days. Would design a CPU. It will take 400, 500 people 4, 5 years to do that. Right now, if you want to design a you go to TSMC, use Cadence tools, get some, maybe do your own CPU or get it from ARM. You can do it in 6 months with 40 people. That's 10x times 10x. There's a 100x improvement in productivity over 20 years. And then maybe 10x more with AI. So this is the reason all these companies can do this. The fact that chip design has become more scalable is the reason all these hyperscalers can do that. And I think it will only increase over time. And more companies -- I think more car companies, more data, you look at all these new customers that are announced like open AI or like all these car companies, they will more and more will do silicon to differentiate because you need to differentiate your offering otherwise, it all becomes uniform, right?

Yes. Maybe kind of just close on your System Design and Analysis business for a second. You've talked about enabling that business enabling companies like aerospace, defense OEMs to simulate entire systems. So if you think about Boeing or an automaker, how is selling to one of those customers different from selling to an NVIDIA or an AMD. Do they want the same times of models? And just sort of how is the overall business process different?

Yes, that's a good question. I mean, normally, the system has a longer tail, but the top customers are the same. So semi is more concentrated. We will have probably like 500 customers, let's say, and 50, 60 control most of the like 60% of the spending. I think systems may have sometimes tens of thousands of customers, so it's more spread out. And the top 50 may have like 30%, 40%. It's still a big number, but it's not as much as 60%. So that's one big difference. We need to build the long tail go-to-market whether it's cloud or kind of distributors because in the semi space, we are always direct. But in the systems space, we have to have distributors and cloud for the longer tail. But the top customers, we always -- our culture is always win with the winners, go to the top first. The top customer behavior is very similar. And actually, the top customer, there sometimes the same. Even these big aerospace companies are doing silicon design. So we already -- or you look at these big phone companies, they're already doing silicon design. So the top customer behavior is very different, but the middle and the long tail -- sorry, the top customer behavior is very similar, but the middle and long tail is different.

Then maybe kind of to wrap up. I mean if you think about the synergies you see between the EDA and the FDA businesses long term. Can you sort of flow chip design into thermal electromagnetic models directly? And sort of like how do you think about how a customer would use both those in concert to sort of develop the broader system from chip all the way up to the system level?

No, absolutely. I mean you can see that -- I mean there's 2 perfect examples. One is the phone. So there is a big -- like one of these big phone companies without getting too much -- I mean the chip design is central, but it sits in a very tight confinement. So the thermal and then the drop test, all that is -- it's almost coming together. And then same thing on the AI side, and there's no perfect example than NVIDIA or Broadcom. NVIDIA is full -- I mean, they have such a good job of optimizing chip and the system and the term. And we have all kinds of collaboration with them to stimulate data centers, simulate and of course, design the chip with Palladium and our software. So I think this is inevitable. This merger of system and silicon is going to happen, and it's only going to accelerate because of AI, either because of like the scale of it or because of the form factor. And same thing is true in the car, right? You have to customize your chip separately because it's much more power constrained environment. So one is thermal constrained or even power contained data center, car is battery constrained, phone is size constrained. For all these reasons, the silicon and system -- there is good logic to 2018. I think the EDA and SDA is invariably. And then, of course, you add AI on top of it. And I don't see -- I only see that accelerating in the next 10 years.

Fair enough. And maybe I have time for one last quick question, which is you meet with a lot of investors who ask you questions about Cadence. I'm kind of curious, what do you think is the one thing that is most overlooked by investors about your company and the story? And then if we get up on stage 5 years from now and we look back, what do you think investors will be most surprised by?

Well, first of all, we have great investors. So thank you for that. And a lot of people do understand Cadence very well because we get a lot of -- all the top investors are working with us. I think that what we can do better is to show how critical we are in the long run, because we are like R&D software for R&D engineers. So it's very different than kind of vanilla software. So engineering software, especially EDA, especially Cadence, will be critical. The second thing is people say, oh, you have done well last 5, 10 years, will it continue in the future, 5, 10 years? So we are a compounder of value, right? We are a compounder of what Warren Buffett call the 8 wonder or whatever. So I think if you look back 10 years from now, you will see that our EPS and growth rate, and we will have a good financial model. So that's our goal to keep delivering EPS growth that we have delivered last 5, 10 years in the next 5, 10 years. So you can buy Cadence and sleep well at night.

Sounds great. With that, a great place to wrap. Thanks very much, Anirudh, for being with us today. We appreciate it.

Thank you.