Hi everybody, my name is Natalie Winkler. I'm a semiconductor analyst here at UBS. I'm very excited to have Anirudh Devgan with us, CEO of Cadence Design Systems. Before we start, I want to read a quick forward-looking statement.
These are big chairs, you know.
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 projections or implied in the discussion today. So maybe we can kind of start with the big picture, right? When we think about Cadence, what's Cadence's role in this very fast-developing semiconductor ecosystem?
Yeah, first of all, thanks for your interest. Basically, Cadence, we provide product, mostly software and some IP and hardware products to design chips and electronic systems. So what we like to say is almost any chip design in the world today uses some form of Cadence products. And so that's true for, of course, the traditional semiconductor companies, but also about 45% of our business is coming from system companies like car companies or phone companies or hyperscalers. So that's the mix of our customers. And of course, there's a lot of design activity now for AI and other things.
Maybe you can speak about the design activity for AI. Are you guys seeing that more on the data center side, more in the long tail of edge applications, and where the customer base is expanding for you?
I mean, definitely on the infrastructure side and data center. I mean, I've talked for a while that I see three phases of AI. The first phase which we are in is infrastructure. So that's all the data center and even some laptops and all, but mostly, of course, data center. The second phase being physical AI. So there's cars, drones, and robots. And third phase being sciences AI, which is more like biology and materials. So I think we are still in the first phase, most of it. And I think that first phase still has a long ways to go. But we are also investing in the other two phases. Yeah.
An excellent, because that was kind of my next question. If we think five and 10 years out, how should we map those phases onto kind of your vision to that?
Of course, these are difficult to predict exactly, but I think the infrastructure phase is, of course, going gangbusters right now. The projections for that are very optimistic the next three to five years that the amount of compute and AI usage will be exponential. I think we still see a lot of demand in the infrastructure phase. Then the physical AI phase, in my mind, of course, already design activity starting, but to reach critical mass is maybe three to seven years. I mean, some Waymo and Tesla are already doing a lot of self-driving, but I think within a few years it should become a lot more mainstream. We're already seeing design activity in preparation for the physical AI phase.
So if the infrastructure phase is from now to at least five years, then the physical AI phase, I think, is three-seven years from now in terms of reaching. And then sciences, even though we are doing a lot of sciences work and drug discovery and things, I think it will take some time. So that I put more five-ten years from now. So these three phases. But the investment is there. Most of it is in the first phase and then proportionately less in the second and third phase.
Excellent. And when we think about maybe a little bit more near term, so you recently increased your calendar 2025 revenue growth expectations from 12%- 14%. And you specifically talked about very strong backlog, right? And I'm wondering if you could talk a little bit more about which other segments you're seeing kind of most of the strengths in the near term end of this year and the next year as well?
Yeah, we had record backlog. We reported end of Q3, and we also indicated that I think Q4, I mean, of course, we haven't finished it yet, but all the indications are that we should end up with another record backlog end of Q4. We are always focused on, I mean, for those who are familiar with us, you already know, but who are not, that we are always focused on revenue growth plus margin. Our job is to make money for investors, so our margin this year is roughly 44%, a little more than 44 and a half. Revenue growth is 14%. Okay, that's a rule of 58, and we want to keep building on that. We have increased margin every year for the last five, 10 years, and we can keep doing that going forward.
And the revenue growth, yeah, I mean, we always want to make sure that we have profitable, sustainable revenue growth. And I think last five years our CAGR is about 14%. And then we'll see how things go in the future, so.
Excellent. And maybe we can talk about the EDA side of business. And specifically, as we think about different AI applications, how would that be changing EDA business model over time?
Yes, yes. So AI first has two implications because one question is, okay, how does AI affect software itself, what we sell? But for us, one thing to remember is we are also, there are only very few software companies that are helping build AI. So EDA, especially Cadence, whether it's partners like Nvidia or Google or all the big Mag 7 companies, they're using our software to build AI, right? Whether it's TPUs or GPUs or all the custom chips. So part of our business is growing because there's a lot of design activity for AI. Okay. And then the other part is applying AI to our own products so we can make our products much more efficient. So I think there's at least 10x productivity improvement we can deliver by applying AI to our products.
So if you look at our products, in the last 20 years, we already improved productivity by 100x by a lot of other methods, more classical mathematics. But AI can help us next five years improve by 10x. Okay. And so the question always is, okay, what does that mean in terms of usage? And so one thing to remember is for our customers, the workload is exponential. So if you look at the chip design today versus chip design in 2030, right now the chips, like the biggest chips are 100 billion or 200 billion transistors. In 2030, they will be like one trillion transistors, at least 10 times bigger. Plus we have 3D IC, all the software.
So the design complexity is going to be 30-40 times more than now, which is very different from the worry of AI disrupting software is assuming that the workload is constant or only growing up by GDP or something. But if the workload is going up by 30-40x, all our customers want to use our AI tools because there is no way they're going to hire 30 times more engineers from now to 2030. I think they will hire more engineers, maybe two-three times more. And the remaining 10x gap has to be made up by software. So all our customers, if you look at, we are part of R&D, right? We are engineering software. So if you look at the percentage of R&D spent on Cadence and EDA has gone up from 7%-8% to about 11%.
So R&D is going, and then our percentage of R&D allocated to us is going. When I talk to all the big CEOs of our customers, they want to continue to do that trend. They would rather spend on automation and compute to improve the design efficiency. So our goal is, given the workload is exponential, to provide value to our customers so we become more essential to their R&D operation. But there are a lot of ways in which AI, I mean, I can give a lot of examples of how AI can improve. The tools can get 5-10x better. A lot of times the PPA, power performance and area, can be 10%-20% better because AI is doing a much better job of optimizing the design than a human can do. And 10%-20% is huge for power and area.
I guess when we talk about EDA business, I wonder how does that really translate in pricing? I guess the concern would be in the world of agentic AI, if some of the agentic features can kind of potentially reduce the number of seats of EDA software that you need. Is that at all a threat or really the pricing per license and the number of licenses people will actually need is far offsetting that potential headway?
So if you look at our license usage, it's almost exponential. And of course, pricing improves a little bit with volume, but the number of license growth is. And the reason for that is typically when something is faster, you do more of it. And even with AI agents, so the way I always, I mean, I said this for a long time, to really do a good AI solution, you need multiple factors. So there is the AI itself, but you also need the base tools, the ground truth, like how the transistor operates or the classical kind of EDA tools, and then the compute that it runs on. And this is what you're seeing with agents now. The agents will do the AI, but they will also call a lot of tools which they're already good at doing. If you're doing placement, that's a solved problem in mathematics.
You don't need to run it with AI. Optimization, maybe you run AI and then you call these tools. So typically when we deploy our tools like Cerebrus, that gives huge benefit. We have five big AI platforms. They will use a lot more of our base tools. So the actual number of usage of the tools is only going up. I mean, one of it is with AI. The other is, of course, the chips get bigger and yeah.
Maybe on that point, if we could talk about how the hardware business is performing given the traditional kind of refresh cycle and really how we should think about the synergies of the hardware and software businesses going forward?
Yeah, so part of our business is we sell a hardware system. I mean, we call it hardware, but it's hardware-software together, which is like an emulator. So it will verify the design like 1,000 times faster than you can do on a regular silicon. And we sell like a rack system with hardware and software. And almost all the big chips, all the big AI chips use Palladium, which is a hardware platform to build these things. And so the benefit of that is that you can basically emulate the chip before it is fully done or comes back from TSMC. So like about two years ahead, you have a model in Palladium, and then you can run software and do full software bring up everything like that. So Palladium and these hardware systems became basically essential to the design of all modern chips.
And then in our verification suite, we will also sell software that goes with Palladium. So that's one of the big advantages that Cadence has and the reason we are doing well in the ecosystem, especially the AI ecosystem and a lot of the other big like mobile and communication, but especially in AI because the chips are so big, is the strength of our Palladium system. And Palladium, we build ourselves. We design the chip ourselves. It's a fully integrated. We have a 10-year lead in terms of how to build these things. And then it pulls in the software as well.
Excellent. So pivoting a little bit more into the IP side, you guys have obviously seen very strong momentum in the leading-edge IP and arguably is in contrast to some of the peers, right? I think like Synopsys, for example. So wondering how you guys are seeing the dynamics of the IP portfolio going forward, really, and where's kind of the most growth going to be coming from going forward?
Yeah, IP is performing well. So we have five segments we report. I mean, five main areas. I mean, three of them are EDA and then IP and systems. And so right now, all five are doing pretty well. Okay. And EDA has been a traditional strength of Cadence. Okay. And then a few years ago, we invested more in IP and systems. Systems because our customers were becoming more system companies. And IP, we got later into it, but we are always careful about margin, not just revenue growth. But I think now with AI and all, I think in the way the IPs, there are five key AI IPs that we are investing in, things like chip-to-chip interconnect, HBM memory, DDR, SerDes, these kind of things. And I think we are well positioned with those.
And also, the number of foundries is increasing because there are more advanced node foundries. So, the combination of our portfolio and then more demand for AI systems and foundries. I see good growth for the IP business going forward.
Should we think about the IP business from the standpoint of kind of license type of revenue or really there's increasing opportunity for royalties as well?
We already have IPs that have royalty, and that's a very profitable business. Part of our IP business is Tensilica, which is used in a lot of kind of edge applications and edge AI applications. Tensilica is, I think, the number two kind of platform after ARM, which is like a license core with royalty. That's almost like software margins, which is very good for our IP business. Then design IP, which is like these protocol-based like SerDes and DDR, those are more like usage licenses with some royalty, but mostly usage. Overall, I'm happy with the mix. I'm happy with the profitability. The profitability of IP is still lower than EDA because EDA software business is, of course, we have 90% gross margin plus, but still the growth is higher.
So I always evaluate each business on this rule of mix of revenue and margin. So even IP is lower, but I think it can grow higher than Cadence average. So at this point, we are investing in that.
I guess if I think about the growth in IP business going forward, how incremental would be that royalty business you already have to the growth rate or really the bulk of the growth rate is affected from the license kind of type of engagements?
Yeah, I think it's both, but the Tensilica part, the design IP, which is less royalty, is growing faster than the Tensilica part. But Tensilica part is still significant. But the growth, to answer your question, is more on the design IP because of all these chips being designed, which are more kind of AI HPC. Now, as it moves more to physical AI, maybe there will be more Tensilica growth in the future. But right now, it's more of these big data centers, which are design IP related. Yeah.
Excellent. I was hoping to talk about the Hexagon acquisition and how you see the synergies and specifically given the track record of acquisitions that you guys have, how you think the integration process will go there.
Right. Yeah, we are always measured in acquisitions. We always say organic is delicious. So we are an R&D-driven company. We want to, because that's the most profitable way to grow anyway. But from time to time, we will do M&A, especially if the opportunity is good. But that's always the second preference for us. So the question is, why did we do Hexagon? It's basically for physical AI. Okay. It's basically for physical AI. So we have a system business, which is growing well last five years. So half of the system business is focused on 3D IC. That's the big trend for all these AI systems, which is multiple chips in a package and all the analysis that goes with it. So in systems, that's a huge trend. There will be a trend for the next five, 10 years.
The other thing I'm always optimistic about is physical AI. Okay. And in physical AI, I think everything is going to change. So if you look at the three-layer cake again, which is AI and then physical simulation or the ground truth and the silicon. So when you go to physical AI, all three are different. Of course, the silicon is different because it's a power constraint. So you look at the Tesla car has this AI4 chip or AI5 chip. It's very different than a data center. Same thing with BYD and all the other Rivian and all these companies. So the silicon will be different. But silicon, we were already well positioned, and there will be more mixed signal and all in Cadence strength. But also the AI model is going to be different.
So I mean, all this talk recently anyway of a world model, W-O-R-L-D, like a physical model rather than an LLM model. And the thing is in LLM models, all the data is available already on the internet to train the model. But in a physical model for a robot or a drone, the data is not available. And the data is not easy to get because you have to put all kinds of sensors on people. And so in there, the simulation becomes critical. So Hexagon D&E business has the leading multibody simulator. It's a robotic simulator in the market. So that will really be critical for this physical AI model. So that's why we had a good discussion with Hexagon, and they wanted to focus actually, they are building their own robot. They wanted to focus in a different way.
And all the software businesses, which they called D&E, we acquired, and we can integrate in our flow. And so then in systems business, we'll have one half focused on 3D IC, one half focused on physical AI, and both are big growth drivers.
And the physical AI opportunity effectively opens a new customer base for you, right? Kind of the emerging physical AI.
Absolutely. And there are some traditional customers there too, like cars. That's a big business already. And of course, it's going through a lot of change with self-driving and electrification. But a lot of the business, that BETA CAE, which is the acquisition we did, and Hexagon is already in automotive. But then there could be newer things like drones and robots. So all three will be critical. Yeah.
Have you guys sized sort of that Physical AI opportunity specifically in the fields you will be playing with the total kind of total addressable markets?
Yeah, it's difficult to say. I think it can be huge, but I think it will be the main thing I want to make sure is we are already well positioned in infrastructure AI. I want to make sure that as this new thing happens, we are also well positioned in physical AI. It's difficult for me to point out exactly, but I think it will be significant. Yeah.
Excellent. So a key competitor in your space, Synopsys, they recently did an acquisition of Ansys, right? And I think one of the applications there was also the physical AI and the digital twin capabilities that Ansys had. I'm wondering how that has changed compared to landscape for you.
Yeah, we are doing this from 2018. So I don't know if you go back and look. I'm the one who, because when I was supposed to take over as CEO, one question from the board was, "Okay, EDA is a good business, but what's the future of EDA?" And this is a very different time. At that time, 2018. This was before all the AI. And I always believe that silicon and system have to merge. And you're seeing that. Of course, now it's obvious, right? Whether it's NVIDIA or Qualcomm or Broadcom and all these hyperscalers. So we have been investing from 2018 in this. And our system business has grown like, I don't know, 25% a year for the last five, six years.
So we have a pretty good portfolio, and we are focused on the high growth part of the system business, like I mentioned, 3D IC and physical AI. So it's good. We're growing well. Customers are happy with our solutions, and we go from there. We were already competing with them separately. Together is not. We just want to focus on what we can deliver to our customers.
Great. So I guess coming back to more financial side. So from the regional standpoint, I think in the recent quarters, you've seen a significant strength in China business. And just kind of curious what's driving the outperformance for Cadence versus the peers there. And how does China actually fit in kind of actually maybe this is more a long-term question in your long-term vision.
China is good business. I think China, if you step back, has come down over the years, and of course, semiconductor companies have a lot of China exposure. From an EDA or software standpoint, we used to have, I think, 16%-17% used to be China a few years ago. Now it's like 11%-12%. Okay. It was still good business for us, and China has a lot of design activity, as you know, both in infrastructure and physical. I think this year was a weird year for obvious reasons, a lot of geopolitical. So when we started the year, we were pretty conservative. That's our culture anyway. We'd rather print the numbers than project something we are not sure about. So we were pretty conservative in our China assumption because we knew that there would be a lot of uncertainty.
But we are doing better than we thought, which is good. And I think China business should grow this year. But the behavior of the customers is fairly normal to me, so it looks like. And we had some issue in Q2, Q3. I mean, there was some because of some of the restrictions, some of our business moved from Q2- Q3. But now all those things are resolved. So the shape of the curve in China is a little different for quarter by quarter. But if you step back and look at the full year, I think we will end up around 11%-12%, something like that. So I feel that the situation I mean, of course, it's very difficult to predict the geopolitical, but it seems stable for now. And the design activity in China is back to normal.
They are investing a lot anyway in silicon systems, in cars, robots. There are a lot of companies in China. Yeah.
Excellent. And so maybe a little bit on your margins. So you're known to operate your margins at 44%. How should we think about the trajectory from here? And specifically, in light of the acquisition, I think you mentioned that MSC may have been somewhat underinvested. How should we think about the impact from that?
Yeah, we manage the overall margin anyway for Cadence. I mean, MSC still will be important, but it's a small part of Cadence. So our margin is about 44%, a little more this year. And what we always look at is incremental margins. If we add $100 million of revenue, what is the margin on that? So for the last several years, I don't know, eight years running, our incremental margin is 50% or better. Okay. And that's what so there's still a lot of room for improving margin from 44%. Actually, our organic incremental margin is close to 60%. Okay. Now, if we do some M&A, then it comes down to maybe low 50s%. But M&A we will do if it makes sense, and sometimes it does make sense. So yes, there will be some effect on margin from M&A.
But overall, we'll try to make sure that the margin still improves for the company over time. And we are always shooting for 50% plus incremental margin.
And that applies to short term as well as long term, right? That commentary?
Now, sometimes there's like in a particular year, you close it. Depending on when you close, there could be some like for example, last year, our margin was slightly lower, incremental margin. But this year was really high. So if you average the two years out, our incremental margin is like 53%-54%. So it may happen in 2026, our incremental margin is a little lower. But 2027, it may accelerate. But overall, our goal is still we can drive margin improvement and also make the team more efficient, of course, use AI internally. So again, revenue growth and margin both. There's still room to go. Yeah.
And when we think about the physical AI opportunity, how should we think about the kind of margins in that potential business going forward, several years out?
That should be good.
There's no structural change to either gross margin, operating margin compared to.
No, no, no. This is mostly software. No. And also the physical AI, of course, see, the main lot of businesses is still infrastructure AI and all the AI build-out. The good thing with physical AI is that it also reinforces infrastructure AI. Just as an example, like Tesla, of course, they run the model on the car, but they train it on the data center. And same thing with other things. So it will be additive to the current trend, and it will reinforce the data center side. So no, we'll make sure margins are good anyway. And we have done this for like eight, nine years if you look at our margin trend. Yeah.
Excellent. And maybe in the last minute or two here, how should we think about the capital allocation priorities for Cadence after the deal closes with MSC and Hexagon?
Like I said, most of it is organic investment. And of course, we generate a lot of cash. There's no chain there. We will take 50% of our cash flow, and we buy back our stock. And we have done that also for seven, eight years. And the reason for that is that we are always looking at SBC. So we also track margin minus SBC. Stock-based comp is a very important thing for us. So it's about 8-9% right now, which is still better than the peers. Because to me, 44% margin doesn't mean anything if your SBC is so high. Because all our employees would rather get stock rather than so we're also very careful on SBC. Now, it's going up a little bit, but overall, still much better than everybody else.
And then the goal of buying 50% back is, we want to make sure that there is no dilution. So we're actually buying more than we issue in SBC. And then the remaining cash, we'll see if we do some opportunistic M&A. But it does not change our model, which we have done. Good thing Cadence is a predictable business. We are an integrator of value, compounder of value. So this should be the same. And this kind of M&A doesn't change our financial model.
Excellent. I think this is it. Thank you very much.
Yeah, thanks a lot.