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

Good morning, everyone. Before we get started, they've asked me to read a quick safe harbor statement, so I will do so now. Before we begin today, the company is -- the discussion will contain forward-looking statements and will make use of certain non-GAAP financial measures. Please see Cadence's most recent 10-K, 10-Q and website for a discussion of risk factors and the use of non-GAAP financial measures. Well, that's out of the way. Thank you for coming, everyone. I'm Stacy Rasgon. I'm Bernstein's senior research analyst. I cover the U.S. semiconductor and semiconductor capital equipment space. And it's my honor to have our guest here today, Dr. Anirudh Devgan, the President and CEO of Cadence. Our talk today's going to last about 50 minutes. If you want to have -- ask questions, if you look on the inner cover of your program, there is a QR code you can scan that will take you to our question-and-answer forum. It's called pigeon hole and you can submit your questions there, and we'll leave time at the end for that if there are any. So Cadence, Look, they play a very, very critical part of the semiconductor value chain. It's not making chips themselves, but rather selling the critical software and infrastructure that's needed to design those chips. The space is known as EDA or electronic design automation is a fascinating space. To tell you about it, it's just a great pleasure to welcome Anirudh. So thank you so much for being here today.

Thank you, Stacy.

And maybe just to get started, maybe you can just a brief overview of like what is EDA? And what is Cadence and what do they do? Where does it fit in the supply chain? And like what are these things that you do that are so critical that nobody else can do?

Absolutely. First of all, thank you for being here, and thank you for your interest in the company. So what we make is basically software, we have some hardware products, but it's mostly software to design these chips. And this is not your vanilla kind of HTML software, it's like very mathematical software. So what we call computational software. So a lot of computer science and maths because these chips are very complicated. It's probably the most complicated things humans have done, right? So like these days, it's like 3-nanometer or 2-nanometer, you could have like 100 billion transistors on a chip. So there is no way you can design it manually. So most of the...

They used to, right? They used to design.

Like 30 years ago, 40 years ago. Or like there's other kind of software like I mean great company like your architect designs a home and 5 bedrooms or whatever. So that you can do manually. But if you have 100 million transistors and then it's not just the placement of transistors. It's like how fast does it run? Does it function correctly and because the manufacturing is so expensive. So a lot of it is done in software now and it has developed over the last 30 years. So almost any chip that is designed today uses some form of hidden software. So we are lucky to work with all the big companies across the world. So that's our base business, okay? And there is some IP to that. IP means previously designed blocks like DDR or some CPU cores. But most of our core business is EDA, and we are the most -- we are the biggest and the most diversified EDA player, okay? And then the same software, the competition software can be used for other things. So we have recently a few years ago, expanded into system design and analysis. And of course, AI. AI is mathematical software, too, right? So that's the kind of brief overview.

Got it. Maybe you can talk a little bit about -- a little more about those different pieces. So the EDA -- the IP block. So this is something -- it was, I think, several years ago, you started to get more into this, I'm always fascinated by the emulators, which are these effectively big supercomputers. Maybe you could talk a little bit more about the different pieces and the characteristics relative to your core business.

Exactly. So in the core business is EDA and IP, and then we can talk about AI and systems. But in EDA and IP, there is mostly software, IP and hardware. These are the 3 components. So software means all kinds of analog/digital verification. IP is about 12% of our revenue. We don't overinvest in IP, we have enough IP because one thing with IP, this is good business, but it's not as profitable as EDA. Just because the customer always is build versus buy. I can design it or I buy it. Whereas in EDA, there's no way even for the biggest companies in the world to do their own EDA. So EDA is always better margin. But IP, we do critical IP, like recently all this 3D-IC, like UCI interface or DDR and then we have Tensilica for embedded processing. So it's a good business, about $400 million. And then hardware is like emulator, it's like a rack, you can buy that will emulate the chip. So -- and almost all big companies use that now this is the leading product in the industry. And it will run like 1,000x faster than like a CPU from Intel or AMD. So this like essential now in the design process. So because these things are so expensive and you want to get them right the first time...

Before you send it to the fab.

Before you send it -- so what will happen is on these emulators -- and I think NVIDIA and Jensen also talked about it recently is that that's a longtime partner of ours is that they will actually emulate the chip before it is manufactured. So you can run all tests on it, make sure it works. Not only that, you can run software on top of it. So you can boot windows or Android or run your AI stack, all in emulation before it goes to manufacturing. And there is no way to design these complicated chips without emulation. So that's the third part of our core business. And I can talk more about that.

We'll get there and the whole Jensen and the whole AI thing, we'll get there. I will ask with the general like the business. How does it work? Is it a subscription? Is it like is it a site license or a seat license or for the EDA piece in particular?

Yes. The EDA piece it's a subscription. So it's almost all subscription. I think hardware piece, there is some upfront component to it. Now some customers will lease the hardware, so it's subscription, but some big customers will buy it. But the software piece is 100% subscription. And we have -- so overall, our revenue is about 85% to 90% subscription which is great these days or any days, and it's very repeatable. So almost 99% renewal rate or more. I mean unless the company goes out of business, they will typically renew, right? And then we have different business models. I mean, it's moving away from like seed-based, it's more like how much software you need to run to design the chip. So it's not -- some products are tied to a number of engineers, but more and more is tied to amount of work that is happening. And we have all kinds of business models, especially with the big companies.

Got it. In general though, is it fair to say that the growth of maybe EDA in general is not so much driven strictly by the growth of semiconductors in any given year, but more around like the number of new designs and the complexity of those designs that are being created. is that what drives it.

Exactly.

Is that what drives it, okay.

If you look at 3-nanometer, look at what some of the leading foundries said, the number of designs are double -- the design starts have doubled. And the other thing is, let's say, the customer goes from 5-nanometer to 3-nanometer. And Moore's Law is slowing. We can talk about that. But one part of Moore's law is not slowing is area scaling. Maybe power and performance is not getting better, but the area is definitely better. So what happens is it's the same size chip at 5-nanometer, you go to 3-nanometer, the number of transistors doubled. So if it's 100 billion it will go to 200 billion. The moment it goes to 200 billion, you need more software because the block size went up, you need more emulators.

There is a scale, like linearly? Or is it like 4x software, if you double the trends? I don't know if that's a metric that you guys look at or...

So like verification, for example, will go up exponentially because if you double the number of transistors, the number of state space goes up exponentially. So you need more simulators and all that. And then the emulators we sell, the hardware systems are based on gate count. So that's linear, but parts of it are definitely exponential. So that's 1 reason that EDA has grown so well. And then I -- so then people ask me, okay, is that going to continue?

Well, that's the next question. It's like as people are looking for semis to grow to $1 trillion by 2030 or 2035 or whatever. And that's -- I don't know what that is mid or maybe mid- to upper single-digit growth, is that -- should EDA grow faster than that? And historically, probably has outgrown the semi industry, right?

Yes, especially last few years. I mean if you look at our last few 3-year CAGR, our revenue is 15% growth. And our margin is 41%, 42%. I mean there are multiple reasons for that. I mean one is the growth in like licenses and all that. And it is going to continue for a while. So we are entering 3-nanometer then there is 2, there is 1.4, and there is 1 at least. So that's 2, 3 years -- so that's 10 years at least, okay? And then we have 3D-IC on top of it. So I think this area scaling will continue for a long time. Now the architectures will change. There's more and more compute rather than -- the speed up is coming from more compute like GPUs rather than performance. So that will continue. And then you add on top of that expansion into systems, that's a new TAM market for us, okay, that's unique to Cadence. And then I think AI has the opportunity of moving more automation in the design process. So right now, we are about 10% of customers' R&D is EDA or automation. But I think 5 years from now, it could be a bigger portion. So that's the third reason I think we can grow.

Yes. I mean that's always been like one of the things I've always wondered about is, it always felt to me it was like there's our R&D spend and here's the amount that we're going to spend on EDA and that's how much we're going to spend. And in some sense, sort of limited because you think about like the value that depends on what you do and what the EDA space does. And I was also surprised you didn't actually already capture more than you are. Now you can contrast it with the systems, my assumption would be the system players are buying are probably less sensitive about maybe what they're paying for similar stuff. Like are those dynamics accurate? Like how do you think about your ability to capture more R&D spend in the past, given how difficult it's been like -- or in the future, given how difficult it's been in the past?

I think both companies are doing well, right, semi and system and different characteristics. But there is a big opportunity to get our fair share. So -- and I think the customer -- we also have great relationship with all these customers. But the system companies are, of course, operating at a different level. I mean, they are much bigger companies, sometimes. And also there is a software component to it. So there is more. And then the other reason I moved into the system design and analysis is not only sell them IC tools, but system tools and palladium. So the...

Can you give an example of what that means, by the way, like system design. Like what does that mean in your kind of like -- what -- is it, I don't know, plant design or like I don't know what are you doing...

Yes, system also is like often used term, right? So a good example is your phone, right? So of course, it has a lot of chips in it and they are super critical chips. But the design of the mechanical stuff around. So system, by definition, has to be mechanical plus electronics and hardware plus software, okay? So the way we look at the word is there is silicon, system and data. So these are the 3 kind of concentric circles, okay? So a good example is like a car, like electric car. So you have all the data, all the navigation data, then you actually have the physical car, which is mechanical plus electronics and hardware plus software and then the silicon that drives the car. And so when we apply computational software to silicon, that's EDA, apply competition software to like cars and planes and phone, that's system design and analysis. And then when you apply competition software to data, that's AI. So system for us would be like a car or a plane or a phone or a data center. And that market itself is about -- so the system software market in my assessment is about $50 billion, okay? So EDA is about $10 billion and growing, okay? The system market is about $50 billion, but we don't want to participate in all of it because some of it is not really computational. So like if you're doing some gooey based tool, that's not really computational system design or PLM or -- I mean they are a good market, but it's not competition. What is computational is simulation. And so that's about $8 billion to $10 billion. So that's what we entered. So simulating the phone for thermal effects. Remember the Samsung phone was melting a few years ago or like electromagnetic interfering. I mean that's or like aerodynamics for cooling, same thing with cars, the range of the car, especially electric car, it depends on aerodynamics. So there's like computational fluid dynamics, finite element, electromagnetics. That's a very interesting market. And the R&D synergy is high between EDA and that, and it's -- we need more and more simulation at the system level. And also, it is almost as profitable or more profitable than EDA. So for a lot of reasons, it makes sense for us to expand into systems.

Got it. Got it. I'm going to jump ahead. So I've got more of this, I want to ask, but you said the word AI a whole bunch of times. And clearly, every company needs to be an AI. Let's talk about the Cadence AI story I think there's a number of different levels of it. I mean there are -- there's leveraging AI within the core EDA function itself to actually like bring out better chip designs. And there's other partnerships, for example, you mentioned working with NVIDIA and maybe even on the hardware. So I don't know yet. So maybe we could talk a little bit more broadly. And it's got cuLitho now, which is there like framework or library, I guess, for chip design and they've been partner of yours as this look like.

I mean first of all, AI is, of course, a lot of discussion now. I mean we have been doing it for more than 5 years and even generative AI for like first product we launched, which is Cadence Cerebrus is like 2 years ago. So there's a lot of ways -- and this is great for the industry. So there -- like you said, there's a lot of ways we can participate. So one is just the design of these chips. So because these are all our customers, right? So whether it's all the big GPU companies or a lot of the cloud players have their own silicon efforts now, okay? So we participate in all of them. We talked about NVIDIA -- NVIDIA has been our partner for a very long time, not just on the software side, but also on the palladium so like the emulation. And same thing with the other GPU companies, all the cloud companies, even all the ASIC companies. They were talk about recently like the custom silicon guys. So they're all -- we are working very closely with the 2 or 3 main ASIC companies because what happens is when the big system companies do silicon design, they don't do all of it in-house immediately. They do some -- they will do the front-end in-house and then do the back end with ASIC company. And then later on, they can move the whole thing in-house, which is COT. So in that case also, we will work with the system company to do hardware and software at the front end and then work with the ASIC provider in the back end. So that's one model. The other model is the system company is doing everything themselves. And third model is working with the silicon providers. So we have plugged into almost all of these kind of critical companies. And then a lot of times, the critical part of AI is also the software stack, which, of course, [indiscernible] has a huge lead. So almost all AI companies are also using palladium for software development. And a lot of times, they will give palladium to their end customer. So even before you have a chip, you have an emulator. So a lot of times, they will give palladium boxes to their end customers and the end customer can do software checking on.

That's running the chip that the customer ultimately is buying for them, so they can do all their software and optimization, everything.

Exactly. So there's a chip design part, but there's also the software bringer, which is again possible only in emulators. Because if you wait for the chip to come out and then do software development, it will take forever, right? So the best-in-class companies and NVIDIA is a great example. The chip comes out 3 months later, you can buy it from Super Micro, DGX. So that's because they overlay hardware and software development. If this is hardware, this is software. If you do it like this, it will take forever. But if you overlay and the only reason overlay is possible is through those emulators. So all the other AI companies also will provide emulated to their end customers. So that's like building out the AI infrastructure with like EDA tools, with palladium emulators, thermal is a big thing. So we have all kinds of thermal simulation tools. So that's one big thing. The other thing is applying AI to our own products. And this kind of is very interesting. So and this is -- we have done this reinforcement learning, which is now all the buzzword for Generative AI for some time. So the way we apply to our own products is -- so EDA for 30 years has done very complicated software. But one thing we never did was -- we never did like workflow automation. So EDS software is always what I would call, single run that you give it an input and it gives the output, say that I want this kind of CPU and -- so it will run for like 1 or 2 days typically and it'll give you a...

Is that how long it takes?

Yes, it depends on the -- but like the more complicated tools will run like a couple of days, okay, a few days. And they do all kinds of -- I mean, this is probably the most complicated mathematical software ever written. It does all kinds of optimization, geometry processing. But it was -- when you run it again, it has no idea of what happened before. So it was single run, okay? But the design doesn't happen in 2 days, right? Even the best companies will take 6 to 12 months to do design because you are changing -- so typically, if you go to all these big customers of ours, they typically will run, let's say, Innovus, which is one of the flagship implementation tools. They will run Innovus. They'll change something. They will run it again. Either the process changes or the design spec changes or they are exploring the design space for the best power or performance. So typically, all these customers are doing is they run the tool, they change it, they run the tool, they change it, okay? So that workflow automation we never provided. And the reason we said, like, why didn't we provide it is because there was no foundation -- mathematical foundation to transfer knowledge from one run to another run. It was always done by the human intuition. This is how I designed a GPU last time or CPU last time. And this is what if I'm designing 3-nanometer, this is what it needs, okay. So what we can do with reinforcement learning, and we have several tools, but one of the tools in implementation is called Cerebrus -- Cadence Cerebrus. It automates that running of that searching that search space. So it's huge value because then you're replacing something that was manual before.

Does that mean they have to run it less?

No, they run it more. So what happens, I'll give you example, right, with 1 big customer. So that design is CPU design, okay? It had like 17 wearables. Either some are process variables, some are design variables. This is the user level variables. I mean inside the tool, it does all kinds of stuff. So there are like 4 ways to automate that, okay, or do that. So first way is just human, right? Manually you do it, which is the current way. The second way is like design of experiments, basic statistics, okay. Design of experiment is infeasible because 17 variables will take like 4 million runs, okay, it's not possible. The third way to do it is what is called gradient based optimization, okay? Like when you train the weights in AI or...

It's like back propagation.

Yes, back. But that's too difficult because these things are -- you can't take a gradient out of them. It's important. That's why we never did it, okay. The fourth way, which is now possible is gradient free optimization, which is reinforcement learning, okay? So in that case, for that example, instead of running manually or 4 million runs, it does it in 200 runs. You still have to run it multiple times, but the Cerebrus tool is doing that for you, okay? And then we run it in that case on 10 machines in parallel. So if you use...

Each one still takes a couple of days, right?

Yes.

Okay. Yes.

But it doesn't take like 20x longer. There are some tricks we have. So in the end, if you run it on 10 machines, it takes about 3x longer. So what used to take 2 days, 1 run with reinforcement learning takes about a week 200 runs, and there are some tricks to make it more efficient. But you are replacing like months of work, if you've done manually, that's 1 thing in 1 week. But what is even more powerful is that the results can be better than can be achieved by a human. Because it's very difficult to optimize on 17 dimensions, whereas this reinforcement learning is intelligently -- so sometimes -- in a lot of cases, we have 5% to 10% better PPA than the old way of doing it.

PPA is?

Power, performance and area. So that's huge, okay? 10% better power because when you go from 1 node to another node, the gain is now reducing is like 10% to 15% in power of performance. You can get that or half of that by better algorithm to search the design space. It's invaluable. And it's not that -- it's not that there was extra juice in there. I mean there was extra juice, but it was impossible to find it just by running iteratively by human process, whereas this way with reinforcement learning, it will search that space in.

So if the number of runs are going up like 100x just say, is your revenue opportunity for something like this 100x versus the old way? Or like how does that work?

Yes, there is a lot of opportunity to capture like bigger percentage of R&D spend.

Yes. I mean that's -- like if you can save 10% of your data, that's huge, right?

That's huge. And we have lots of examples. So we have now out of 10 out of the top 20 customers using 8, 5 big hyperscaler company using it.

Why aren't the other ones using it?

It's just a matter of deployment. But the effect is huge because it is replacing something that was manual, okay? And I think what is likely to happen is like if you look at -- we have like 100 billion transistors. By end of the decade, it will be 1 trillion transistors, okay? So the work is going up -- the size is going by 10x. The amount of work is probably going up by 30, 40x, okay. So there is no way to hire 30, 40x more engineers. I mean, they're not even that many graduating. So what I expect will happen is that the customers will still hire more engineers, maybe like 2, 3x more than now. But the rest of the productivity gap -- so we need to be 10x more efficient even with like 3x more engineers. So as a result, I think the customers will still grow, but if there is an opportunity for a bigger portion go to automation than headcount.

Yes. Got it. How much more complexity is -- you mentioned a little bit on systems with UCI and chips. How much more complex it is moving to like a chiplet architecture bring into this? I'm assuming that's part of that 30 to 40x increase. But is it all driven by chiplets? Because that's where a lot of this like I can't put a trillion transistors on a single monolithic die, like the die is not big enough the area is not big enough.

Yes, it will be -- I think the single die will still go up because you have 3 to 1.5...

Where's Jensen now. He's -- I can't remember all the -- they're all very limited, right? I mean you can't make it...

But even -- but like earlier point. Even if it's the size of the chip is the same when you go from 3-nanometer, 2-nanometer number of transistors will double. So the size of the chip, but the way to increase the size of the chip is 3D-IC. See, if you look at last 5 years and NVIDIA talks about this very well, and it's a great example. The performance has not come from pure scaling. Performance has come for putting more things on the chip, right? So instead of 1 CPU, now there are 8 CPUs or GPUs is perfect example. There are so many cores. There are thousands of cores on those chips. And that's why they can do a lot of computation. So the "Moore's Law" has been driven by more things on a chip. So what is the natural extension of that is putting more chips on a package. And I think we talked about it even 20 years ago, but it was not feasible or the regular chip was growing anyway. So -- but now even like Amazon launched this graviton. It has 7 chiplets on an interposer or a package. So I think that's kind of orthogonal way to get more capacity. And that's starting in HPC first, but I think it will permeate all. And from a cadence standpoint, okay, so because we are doing very well on the chip design software. But 3D-IC, the most important part is package design, right? It's just multiple chips in a package. So we have a majority share of package design software. It's called Allegro. So then we can put a system together with chip design software, package design software and then analysis on top because one of the biggest things in 3D-IC is thermal dissipation because we have so much more. So we have this integrity platform, which is analysis, package design and chip design, and Cadence is the only company to do that. And a lot of the recent flow like TSMC launched 3D blocks in October, a lot of it is based on Cadence flow. So we are in very well positioned to do 3D-IC. And then the same design space exploration that you can do with reinforcement learning at the chip level can be done in 3D-IC level and can be done at the system level because it's the same problem, like -- yes, so the beautiful thing about reinforcement learning is doesn't matter what's in the core, it's a way to search the design space. So the core engine could be place and route could be thermal simulation, could be 3D-IC. So that's why we have like 5 major products now, all based on reinforcement learning, yes.

Okay. Got it. I guess one of those ones you talked a lot about like the capabilities at Cadence. What is the broader competitive environment? You've got Cadence, you got Synopsys, you have [ Magna ], I guess there were different some consolidation and purchases in this area. There's some upstart Chinese competition, which is -- we'll get to geopolitics maybe in a bit as well. But I mean that's certainly a concern because a lot of your most advanced products are caught up with some of the export controls and they can't get sold. But what does the competitive environment look like relative to, say, like a Synopsys, was like the other like large...

Yes. So I mean, one good thing is that we are now down to like 2 major players in like 10 years ago. And you asked me like, why didn't we capture more of the value? I mean I think we had -- there were a lot more companies like 10, 15 years ago. So now we're down to primarily 2 major players, okay? And Cadence versus -- our major competitor, we are much more diversified. So we are always strong in analog historically.

You were the analog company back in the...

When I joined Cadence 10 years ago, we were the analog company and then they were the digital company. But now we have analog, we have digital. We are in the all top 20 digital companies. We are close to all the leading foundries, and we are very strong in verification with palladium and so we have the most comprehensive portfolio across all kinds of chip design. And then the other unique thing is this move into systems. So with like 3D-IC and packaging and thermal because we -- a lot of people used to think that the 2 companies are similar, are like twins or something 10 years ago. But they are diverging now, okay? And I think this move into systems, which we started about 4, 5 years ago, I think is also unique to Cadence. So I would say we are much more diversified in terms of products and customers and more into systems. I think we are less in IP versus our competitor. But that's intentional to some extent.

What areas -- I guess if you could just picking and choosing an IP in terms of where you want to serve?

Yes. That's why we are more profitable.

Okay.

So in the end, it's not your -- it's like take-home pay is more important than your -- so we are like a lot more profitable. Our margin is like -- this year's guidance is 41.5% non-GAAP operating margin. And also our SBC is much lower than our competitors. So one key metric I track is non-GAAP operating margin minus SBC. So our SBC is about 8%, and non-GAAP operating margin is 41.5%. So the effective -- because we are giving out stock. So we have to take that into account. So that's 41% minus 8%, right? So that's roughly like 33%, 34% profit. And that's probably 9%, 10% better than our competitors because typically they give higher SBC and lower margins.

What about the risk of like sort of Chinese upstarts, especially in the wake of like geopolitics and export controls?

Yes. I mean there are a lot -- there, at least we are tracking all the China EDA companies there about 10 or 15 of them. But still, it's they're small, and so they're -- and more focused. So we continue to track it, but I think it's not a -- I don't expect that to be a short or midterm issue because all the big China customers -- I mean, they may have some engagement with these companies, but most of it is with the primary flows.

Is piracy an issue for you guys in China?

Yes, that's a big issue. That's what we explained to the government also.

How do you deal with that?

Well, we haven't -- we have some piracy revenue, and it kind of -- it's difficult to predict. It goes up and down. So sometimes we get a flush of upfront revenue. We have some -- but typically, we try to -- we have some kind of monitoring software. It doesn't work in all cases, but in several cases, it will work. So we will talk to the customer saying that, well, we noticed you have a lot of usage and -- so we work with like that way. And then the other thing in China, which you may be interested in is all these export regulations. So we follow all the U.S. regulations, but the impact to cadence is fairly limited because we are very diversified. And most of the export regulations are targeted towards manufacturing, not design activity because of all these piracy and other issues, I think, is the government has decided correctly to focus on the manufacturing side. So the impact of the design side is fairly limited. And all that is in our guidance. So this year, we are growing about 14% with all the effect already included.

Got it. You have the targets for this year for the guidance. You don't have like -- you have never put out like a model, have you for like long term, like we're going to grow X, we're going to have margins of Y. That's not something you guys have done.

No, we like to print than predict. But one thing we have said though is -- and we did that like a while ago when our margin was like 25% or something. We put out a long-term target. But now I think there's only 1 company I checked IR team in terms of software space that has 14% revenue growth and more than 40% operating margin. So -- but one thing we have said is last few years, we have what we calculate what is called incremental margin. So let's say, if we add $100 million of revenue. So that has been -- so our goal is 50% plus. So it has been like 55%, 56% in the last few years. So our goal is still 50% plus. So I still think that in the next few years, we can continue to grow margin. And of course, revenue depends on all this activity. But we haven't given a long-term model, but we are confident to where we are right, yes.

I want to ask a little bit about the near term. So yes, your earnings a little while ago maybe a little bit more challenged than we've seen in a while from you guys. Could you maybe just talk about what was going on in maybe its revenue mix recurring versus upfront lead times like that kind of thing. What happened?

I think what happened in the earnings is that so our full year guidance is -- was a little over 13% and we raised that to 14%. So it was a beat and raise quarter.

For the full year?

For the full year, right? And what happens is when we give guidance beginning of the year, like February, we always guide Q1 and the whole year. We never guide by quarter, okay? So I think Wall Street had a different model for Q2 than what we had. So that was the -- but we never guided for Q2 to be...

Okay. So I guess what are you saying is Q2 no change from your expectations for where they've been in the beginning of the year.

It's better than our expectation. So what happens if you look at Q2, the model was slightly higher than we guided. But if you look at last year, forget this year, we are still up from last year Q2 to this year Q2, we are still up 13% plus. So maybe we should do a better job of giving like quarter-by-quarter guidance, but it -- maybe look at it or it's just -- because I think that the Q2 is the trickiest, but then as you get towards the end of the year, if you tell -- Q3, you can calculate Q4 and -- so I think Q2 is the trickiest one, and maybe next year, we'll give more guidance on what maybe first half should be. I think that's -- honestly, that's the only thing -- but from our standpoint, it's better than what we started off at the beginning of the year.

Yes. Okay. Okay. I wanted to go back a little bit to what you were talking about in terms of you've got more opportunity, and there's ideally opportunity to take more of R&D spend. And so this is an issue that I think the industry is wrestled and not just in the EDAs as costs are clearly going up. We've been in an inflationary environment the fundamental costs are going up, capital intensity is going up, EDA intensity may be going up. And the same thing you've had customers consolidating bigger guys -- ultimately, who pays for all of this I mean, I guess, in a perfect world, it's the customer, the end customer. Like does that happen and even given Moore's Laws slowing down. Is this a risk like -- if it's no longer cost effective to keep pushing the envelope do we slow?

No, no. Because -- okay, there's one important thing that because you guys may already know that when people -- you see all these charts of costs are going up like this, right? Design costs are going up, EDA costs are going up, TSMC manufacturing cost or Samsung or Intel, okay? I think one thing that people sometimes don't take that into account is the volume because this is design cost amortized by volume. The volume is going up faster than the costs are going up. So I can give you a lot of examples like we worked with a car company, and they were trying to decide whether to do their own chip design, and it will cost X million dollars, maybe $100 million or $75 million. But the volume of -- in their calculation, if the volume was 1 million cars, it would make sense. It would make sense. Of course, the reason they do it is because of differentiation, right? The product is better, like a custom product is better than a general-purpose product. So that's the first reason they do it. The second reason they do it is because of schedule, you control your own schedule. But the third reason is at a given volume, it is actually even more cost effective to do it. And that's true for a car company. That's true for a phone company. That's true for a laptop. So how many cars have more than 1 million volume? At least 10 car companies. There are 130 million cars sold in a year, okay? So it makes sense for at least...

Not anymore, but...

Okay, maybe 120 million. okay? It's not -- it's more than 100 million, right? So there are at least 10 companies with more than 1 million cars sold okay? Now same thing will apply to phones. And so I think the volume piece is missing from that calculation. So and there's going to be more and more volume of semiconductors consumed in the world. So I don't see that -- and a little bit slowing of Moore's law is actually not bad. If you go from 2 years to a 3-year's cadence, so everybody has more time to monetize their investment.

Yes. I mean clearly, it hasn't been bad for the industry like it's been fine. We've got about 10 minus. We actually have a lot of questions. So should we go to the audience who are likely around?

Absolutely, yes.

So I'm going to tone this one down, but I'm going to ask it. EDA has been talking about capturing value created forever. Why is this time different?

Well, that's a good question. I mean in some ways, EDA's automation and automation has happened for a while. I think the reason it's different is that, I would say, probably 3 reasons. So one is that there is consolidation in the industry, like you asked me, right? There are 2 major players. So I think there will be different dynamics. Second, I do think this AI and reinforcement learning is a step change in what we can provide, like the whole explanation of workflow automation, I mean it is actually true and the customers like -- even like last week, to give you an example, we just concluded a contract with another big cloud companies, and they bought a lot of these AI tools in this new contract. So I think we are getting actual proof that there is more value. And the third thing is more of the system companies. So it's a different scale of optimization. So I do think, of course, we will watch and see how it progresses. But I think for these 3 reasons, I think it should be different, yes.

Actually close into the next 2 questions. So number one, have you seen any meaningful pickup in interest from system companies directly associated with Generative AI? It sounds like the answer must be yes.

Absolutely. I mean, absolutely. So there are 2 benefits. One is, they always -- they are starting new, right? And so they're always looking for -- because one of the issues is they don't have enough people when you start new and you don't have any kind of baggage of the previous flow -- so they are much more open to a Generative AI-based flow. Plus anyway, they are deploying it in their own companies. So they are much more -- deployment of our AI tools there. The second reason is that they are all now building. I mean their own chips for AI and compute. I mean this is all public, right? so that drives a lot of -- yes.

Yes. Got it. How would you characterize the growth opportunity with non-semi customers, EG, Tesla, Meta, Apple who are designing their own what's the outlook for growth for these customers versus traditional semi? So I'm going to assume that question is asking semiconductor EDA at traditional non-semiconductor companies?

Yes. No, that's a very good question. So the system -- so now system companies are about 45% of our revenue.

Was that high? Okay. Does that include semiconductor design and system companies? Is that just the system revenue?

The system revenue is about 12%, okay. The system revenue, meaning non-silicon revenue, okay? So that's about -- I expect that to be about $500 million this year out of $4 billion. But out of $4 billion, 45% is from system companies. So a lot of it is also silicon design in system companies, okay? So maybe 3/4 is silicon design, 1/4 this system, but both will grow, okay?

Who was that number like 5 years ago?

It was less. But the good thing is that the percentage is more tricky because the silicon companies have done well, too, but it was lower percentage. But the system company move, I think, is only we are in the second or third inning of that. Because in the beginning, like now, of course, some companies are more advanced, like the big phone companies, they have done this for more than 10 years. But a lot of the other companies are still only a few years into it. So we -- first, they do 1 product and they're all successful with it. Look at like Amazon has Graviton. Google has TPUs, Meta has all these chips. So first, they will do like 1 product, and maybe that product is done with the ASIC house, but then they will do more products and then move more of it in-house. And then more companies will do it. So there are at least 3 levels of growth, yes.

It sounds like you think that the general trend of more and more companies trying to do their own silicon design, do you think that is a -- it's a growing trend?

It's everywhere at this point, I think. Now that doesn't mean that system companies won't -- semi companies won't do well. I think there is too much value from a customization, from a schedule and from a cost standpoint, not to do your own. And also our job is to make it easier for these companies to do it. And we have made it a lot more easier than 10 years ago, right?

Yes. Got it. Can you size the total domestic demand from Chinese customers? And how would the decoupling of China impact that part of your business?

So we report this. So our China business is roughly 15%.

Okay. And that's from local? Does it also include like multinationals in China, would you look like if it was high necks. Would that be China revenue? Or would that be Korea revenue?

No, it includes multinational China. So some of it may be if it moves around, if people go to other, then that might move?

Is it based on where the engineers sitting?

Yes, that's how we -- but there's also a lot of -- I mean, there are a lot of local China companies. There are more than 1,500 fabulous companies in China and there is still growth. So again, all these regulations right now are primarily at manufacturing. So the impact to Cadence is fairly limited. So right now, I still see a lot of activity in China, yes.

Got it. The question that always comes up is why Cadence over Synopsys. What are the true competitive differentiations in any historical context along the way where strategies have diverged?

Yes. Like, first of all, I think we want the whole industry to do well, right? I mean that's -- because semiconductors are so important, okay? And I think what people don't realize is that none of these semiconductors and electronic systems are possible without EDA. What I joke around is that if there was no EDA, you need all be like riding horses. It will be like Game of Thrones or something. You don't realize how pervasive it is. So I think, first of all, it's good for the whole industry. . Now we are better positioned for several reasons. One is, we are truly analog, digital, we are the broadest. So if you're doing memory design, you doing analog design and you can look at our customer list, you're doing high-end digital, they all using Cadence. And then this move into systems, okay? It's truly unique to Cadence. And we have done this for like 5 years now, and it's growing like 25% a year. And then when we apply AI also, not only we apply it to chip design but to 3D-IC and to systems, okay? So one thing I didn't talk about like in systems, you could barely simulate like a car or something because now we can do that with our software or working with NVIDIA. But there is still opportunity to apply AI there like we are working with McLaren to optimize the -- in the F1 racing car, the critical part is the front wing. I don't know if you -- it controls the air flow of the car to keep it down. So we are using AI optimize the front wing, which is never possible, we could barely simulate the car, okay? Now you can do optimization, which is the history of EDA. So this AI, which I described for chip design is as or more applicable on system design, which is also unique to Cadence. And then the last thing I'd mention is that our financial performance is just better, the take-home pay is better. Our SBC is lower, our margin is higher. I mean it's a unique differentiation.

Yes. This is an interesting question. Can AI help new companies compete with you?

New EDA company, that is almost impossible. I mean I don't want to...

The question is can AI help -- like is that -- does that give others, I guess, a boost up? Can you help new companies form that can compete with you?

No, I don't think so, It's impossible.

Too hard.

Because we have done this for -- there are all these little startups. See, what you have to realize is, okay, this AI thing. So AI algorithms are okay, give me like 2 minutes to explain this. So AI algorithms are the training algorithm is conjugate gradient, okay? Inference is matrix multiply, okay? We have done this for 40 years, okay? The placer in Innovus is conjugate gradient, okay? The big thing in AI right now is sparsity, okay? EDA did that in 1970s before I was in kindergarten. Okay. So there is nothing in AI is, Oh, Wow, we don't know that, okay? So we are the one who can leverage AI and we have done that so fast because all the 6,500 people in R&D, they can all do AI, okay? Now the other thing with AI that people forget like when you apply a self-driving to cars, you need a good car also. You can't apply self-driving to your bicycle, okay? Maybe you can -- so you need like a good Mercedes or Porsche or Tesla, whatever you want. So you can't apply AI in a vacuum so the base tools are almost impossible to replicate because now there's full flow at 3-nanometer, 2-nanometer. So any EDA company competing with this AI is almost impossible. And sometimes I get a question that will the customers write software like some of the big -- they don't do that either because it doesn't make any sense. So that's not going to happen. I think what AI can enable though, and this is what the customers tell me, there could be a lot more design activity. See, because our tools will get easier to use, anyway they have become easier to use in the last 20 years. So like more system companies could do. So there would be more silicon design activity would happen. But the EDA landscape is not going to change with the -- it's going to change fundamentally, but you're not going to have a small company suddenly put AI and...

And be a big -- got it. All right. I think that is -- we are out of time. I've got more to ask, but I think that's as good of a place to leave it there. Thank you so much.

Yes. Thank you. Yes.