MAGILLEM DESIGN SERVICES (MLMGL) May 2026 Conference Transcript

OK, so welcome to the Arteris session. So I'd like to talk about the system IP and how that accelerates the creation of semiconductors. So the usual disclaimer. So basically, the thesis here is that the future is based on a foundation of silicon. And that all of these business models like automated driving, robotics, drones, and space applications. Basically, they just contain a chip after chip. And it's not just that. There's new emerging businesses, some of which are speculative at this point. So for us, the most exciting ones are the humanoid robotics. Having run a robotics company, I'm a little bit skeptical about that, But those look like cars in terms of silicon and also the orbital data center approach, which may become prevalent in the future. So and after this will come other things. So the future is based on semiconductors. Semiconductors basically do three things. They process data and the main players there are ARM and RISC-V and x86. They store data, and there's memories like SRAMs, high bandwidth memory, DRAMs, and those kinds of things. And then there is the system IP, which is where we play, which basically is the part of the chip that moves and manages the movement of data. So what is system IP? So system IP is basically all functions that essentially allow the chip to move data. So from our product line perspective, there is something called network on chip, and Arteris is kind of the inventor of network on chip. So we were the first to apply networking techniques to on-chip data movement. There's cache coherent network on chip, which is basically for processor subsystems. There's the SOC integration automation software where you're basically packaging the rest of the IP in the system. And you're essentially generating the RTL or the Registered Transfer Language for the exit ports of all the IPs. And then you have created a package model of these for integration. And then in January, we acquired a company called Security, which is basically doing hardware security assurance. And security is going to be part of everything that we do going forward. And securing the data movement is a major thing that has to be done. The company has been public since 2021. We've been pretty much ahead of guidance on all parameters or most parameters ever since then. We have about 350 employees or more. and we are in 4 billion of chips that have been using our IP that has been shipped in commercial systems and we have global coverage in terms of both engineering and in terms of support. And we have a very blue chip ecosystem. We're working with a number of partners because system IP is in the middle of the chip so we have to work with foundries and EDA companies and processor companies and design houses and we have been able to achieve a blue chip customer base that basically has chosen to keep using our technology over and over again and we have about 90% plus customer retention rate. So the strategy for further growth is basically based on four parameters. One is our organic growth is based on basically creating two enhancement releases a year of our product line, existing product lines, the one I just talked about a little bit before, and to create one new product basically every year. So last year was a product called FlexGen, which allows you to generate network on chip automatically. For 2026 is for us the year of the chiplet. And this is a structure where the silicon is no longer on one die. It's on multiple dies that have been combined into one SoC in a relatively complex package. So that's our product for 2026. We are focused on the highest growth markets. So we've had very strong presence in automotive. We support both ARM and RISC-V, NX86. And we are now heavily focused on data center applications, which is where the chiplets are seeing the first adoption. We are also focused on financial performance. Last year, we promised the street that we were going to be free cash flow positive, and we achieved that, and we're going to be free cash flow positive this year. And our goal is to become non-GAAP profitable in the fourth quarter of this year while funding a very robust R&D program. We are essentially doing R&D for 10% of the SOC industry, And so we have to be investing continually in R&D as the company grows. And then we have been supplementing our organic growth, which is about 20%, 25% a year for a number of years with acquisitions. And so our plan is to make an acquisition on the average about every two to three years. And so these are three companies that we acquired. And what I tell people is that the deals are hard, but the hard work starts after the deal is done. So you've got to make those work after you buy those companies. In terms of our five-year plan, this is a slide that's been heavily modified by our lawyers. So you basically have the 2025 numbers, which they did allow us to put in there. And we were $71 million in revenue in 2025. the the end of this is 2030 and the y-axis is basically 250 million and we basically want to achieve that out of three components so there's a very detailed plan underneath this and so the dark blue is license growth the medium blue is royalty growth and the light blue is basically growth coming from two further acquisitions so just so that you believe us I just put up this slide which is basically shows you the royalty growth so there's been so in Q3 we will probably reach 1,000 design starts so our customers reaching a thousand design starts it's probably happened gonna to happen in Q3 of this year, and 60% of the designs that have been done, or actually even a little more, because recently the design starts have been growing quite a bit, are still waiting to generate royalties, but the royalties, so only 16% of the designs that have been done with our IP are generating full royalty, which means they're like in a third or fourth year of royalties where it typically peaks and then there is a bunch of designs that are just starting to generate royalties and so it all adds up to this particular curve where the work has already been done many years ago and these things are just going to ship in the market as part of our customers chips and their customers electronic systems so we think that just to show you that the royalty projections are backed by real data. In terms of guidance, basically what we said is that we're going to be for 2026 somewhere between 95 and 99 million. We beat and raised by about 2 million this quarter. and basically we are going to be free cash flow positive and we're going to reach non-GAAP profitability by Q4 or in Q4 of this year. That's a major objective of the company. So we have been demonstrating for a number of years now a solid financial performance. We have a very strong position in the market segment that we have chosen, which is the Semiconductor System IP, and we have a plan for becoming a much larger company in the future. In a market, it's about 1.5 billion TAM, and that basically we have a long way to go before we achieve leading market share, but we intend to be the leading company in this particular segment and basically be considered along synopsis and cadence and ARM as one of the major companies that helps, that sits at the table when our customers are defining their next generation of chips. So with that, thank you for listening.

Q&A now, if that's okay with you. I guess maybe to start, more near-term dynamics, there's clearly a need for much more compute, and you guys help enable compute, come to market, and come to market faster. How much has business activity, I guess, for you guys picked up in the last year or so? And any meaningful changes? Like, how should we think about you being tied to the AI thematic that everyone knows and loves?

Okay. So, I personally believe, and you may disagree, is that AI is not a separate market. AI is going to be a part of every, it's going to be a feature of every market. So right now, we have announced on our first quarter earnings call that 67% of the design starts in the quarter have been AI. This may reach 80%. And that's because AI is going to be part of every vertical. Now, some of the applications for which AI is being used are not going to pan out. It's going to turn out they're not going to be economical. So, I wouldn't be surprised if there's a downdraft at some point. But when all the smoke clears and all the hype dies down, AI is going to be a feature of every product. And we're going to be talking about AI in terms of the vertical applications, not as a separate technology in a separate market. And automotive is an example of this, right? The ADAS chips, which are used for automated driving, are AI chips for automotive. And people talk about automated driving. They don't talk about automotive driving AI.

And how much, I guess, of their future royalty revenue do you expect to be in a data center? You've talked about increased hyperscale engagements. You have the engagement with AMD. But you have that rich legacy of auto that isn't slowing down. I mean, should we expect that mix to continue to grow meaningfully on sort of the, I guess, data center side of the business and enterprise?

Yeah, so data center royalties will grow substantially, probably faster than other segments. but The endpoints are much more numerous than the data center blades, right and so for us the data center is is primarily a license play and Automotive continues to be and consumer and industrial are basically the royalty place But the data center royalty is going to increase Substantially, but the the data center chips are not as numerous They're much higher priced, but they're not as numerous as the endpoints.

And how do the design cycles change? I mean, auto is notorious for being extremely long. The data center has accelerated seemingly to an annual cadence now. I know you mentioned it's much more of a licensing play, but should we expect contribution to come faster than what we're used to with our tourists, given it's not an auto cycle that takes three years and then ships for seven?

Yes, so absolutely. So yeah, the the automotive royalties are on kind of the long end and the data centers are one of the shortest And that's because the large language models are evolving very very quickly and some of these chips the desired Cycle time instead of three year two to three years is one year And so the data center chips are going to get to the market much faster, which means the royalties are going to start quicker But they're also not going to last as long. So the automotive royalties can last 10 years or more. The data center royalties, particularly for AI inference, are probably going to only last two to three years. So it's going to be a much faster churn.

And you mentioned, I guess, with your lawyer's blessing, that $250 million target in 2030. No, the lawyers did not give me blessing. Okay. You said that, not me. How much of that is signed up, licensing deals that you have visibility to now? How much is aspirational? I think you mentioned there's two more deals you expect in the other chart you showed.

Yeah, so we're looking to make two more acquisitions in the time frame. In terms of the licensed revenue, we have a rateable model. So basically, we recognize the expense is point in time. the revenue is recognized over the life of the contract the shorter the contract term the higher the revenue which the data center actually helps more than the automotive for that and if we were point in time we would be already profitable if we recognize the revenue right up front but then we'd have more spiky variability in the revenue curve which guys like you don't like so we rateable revenue is actually much better for that. So, so yeah, so that's that's kind of the dynamic that we're that we're dealing with on the revenue side.

Okay, and so you have the legacy of chip interconnect and cash coherency. You mentioned trying to compete more completely against synopsis and cadence.

No, no, no. We view cadence and synopsis as partners.

Okay.

We, in fact, our strategy is to avoid competing against arm cadence and synopsis and to be their partner and actually be additive to their success and we have joint development programs with all of them but i think we're dealing with the age of co-op petition right so your main partners are also somehow competing a bit with you and we accept that what we don't want is people at arm or synopsis or and says, okay, let's go out and kill Arteris because they're bad for us.

I mean, can you talk about that dynamic, though? Because I do think part of the pitch that you have made is you can break the ARM bundle a little bit and use Arteris IP instead of paying a full royalty rate to ARM. Can you, I guess, peel back, go under the hood and give us some insight into how those dynamics happen and what your pitch is to basically make some mind share, I guess, with ARM?

So basically, Arm has shifted its strategy over time from being an IP company to being a subsystem company that basically bundles all of its IP into subsystem called CSS, and now they announce that they're going to build chips, right? So the business for us is changing because we're going to be connecting to Arm subsystems, not directly to the processors. and for most companies the processor is chosen before the IP system IP right so we're kind of following the customer's decision about what they're going to use rather than dictating it except in some cases and so we're we're basically working with ARM on a couple different projects one is they're trying to establish the multi-die protocol called CHI C2C so we're working on that with them. And we're just trying to avoid competition and stay out of their way.

Okay. Um, so on the recent earnings call, you, you highlighted wins with a global hyperscaler and a high bandwidth memory player. Um, can you, I guess, how meaningful can these be? I mean, people hear hyperscaler and HBM and think huge numbers. Um, could you, you know, what, what was unique about these deals? What wasn't unique about these deals? And, you know, what, what are you bringing to these customers to help them get to market with more competitive solutions?

So data movement is a category, but a data movement functions different for different chips. And so what happens in the inference AI applications is they're not particularly processor-centric, but they're moving huge amounts of data between HBM memories and the processing elements. So the demand on the interconnect is you have to move billions and even trillions of weights. You have very high data payload bandwidths. And so we're building features for those kinds of data center chips to be able to handle those kinds of AI workloads. And so, yeah, those relationships involve us basically keeping pace with the industry and allow them to achieve the performance characteristics of those kinds of AI chips that basically are very memory-centric and complex SOCs.

You're seeing legitimate revenue synergies across the product families where that is physically possible with the chip architecture?

Yeah, I mean, our pitch when our sales force goes in to talk to those customers is, hey, buy everything from us, right? And that winds up to be a fair amount of money. So we have ability to security, connectivity, register, management, cash-coher interconnect, non-coherent interconnect. And if you buy everything together, it starts to approach $2 million per project. Basically have problems getting funding. So we shifted to large companies. And so there is a finite number of very large companies. And we are essentially doing a lot of repeat business with those kinds of with those kinds of accounts right, so The side benefit of this is when you're selling to a startup when we love startups They often don't make it to production in volume Whereas the big companies When they launch a project they either cancel it or they have lots of customers for it And so the royalty stream becomes much, much more predictable when those designs are basically done by large, you know, top, you know, we have an index of top 30 technology companies. Those companies, when they build a chip, it will sell and it will ship.

I guess, how do you deal with resource allocation and, I guess, financial projections given some of that volatility? I think one thing that I've learned from covering Arturis is your visibility is Especially in the licensing side is much greater than we've appreciated because a lot of it is re-upping licensing engagement Yes, can you speak to sort of you know that backdrop of your visibility and how you prioritize with finite resources?

Yeah, and and so the royalties are very predictable because those designs are already in I mean they're affected by the sell-through volume of the electronic systems, but There's enough diversification now that the royalties are fairly predictable. The licenses also, when we start a year, because of the rateable model, we basically have somewhere between 75% and 80% of the royalty already on our balance sheet waiting to be amortized from our PO onto the income statement. So we have about 80% predictability of what's going to happen in that year, and the following year it goes down to like 60% or something like this. So, yes, we should be very predictable. And if you look at our guidance and our kind of delivery of what we said we were going to do, we've basically been pretty much on track because of that predictability.

And shifting gears, I mean, the AMD engagement you guys announced, it's actually been in, you've had that engagement for a long time, but it got a lot of attention. I think it was last summer when there was a press release. AMD seems like a company that we wouldn't expect to need help with chip interconnect and chiplets. Any details you're able to share on what IP you're providing, what type of applications that you're helping them with?

Yeah, they actually allowed us to say that publicly. So basically, AMD has a very capable cache coherent interconnect called Infinity Fabric, and they're going to keep on using that. So our system IP is always designed for mis-and-match. What Arteris provides is a non-coherent piece of it and a chiplet piece of it. So the chiplet connectivity and the non-coherent network on chip and also magulum. And so we're basically coexisting with the internal AMD product. And they're happy with that. We're happy with that. They could have taken the Infinity Fabric team and done what Arteris has done. They have enough budget to do that. but it would have taken too much time and you know arterias is a proven capability in that in that area and so far it's been working very well and that deal we announced there's been subsequent deals in other parts of amd since then and so uh it's been a it's been a good strong relationship but we're not uh you know somebody uh dino says you don't want to be an amd tracking stock right yeah and uh we have many other large customers including intel nxp and and samsung and many others.

I mean, is it similar types of engagements? I mean, obviously, Samsung and NXP are going to be different, but with Intel, is it similar where you're helping them with a non-coherent interconnect?

Intel has a huge variety of designs. So at Intel, it's all over the place. Okay. And also, Intel is changing a lot because they're spinning off Ontera, they're spinning off Mobileye, they're concentrating on the design data center and chipset and probably a little bit of AI, so Intel's changing. So ask me that in a year, and I can ask that question more accurately. But Intel's one of our largest customers.

OK, I'll try again with AMD then, I guess. Should we think about your progress on the royalty side as tied more to the CPU side or the GPU side of their product portfolio?

More on the CPU side.

OK, got it. And then I guess, how aggressive is this shift to chiplets from your customers that we've seen go on? It seems like it's getting talked about a lot, But, you know, behind the scenes, is this something that you guys have observed happening for a while? Or is it a more recent phenomenon?

So, chiplet makes sense for certain types of designs. It doesn't make sense, for example, to put analog mix signal functionality on the TSMC N3 or N5 processes, right? Same thing. Memory has its own process requirements. And so particularly for data center, where you have a huge variety of kind of data traffic and data processing, it makes sense to move to chiplets. So we're seeing chiplets in PC client chips, in data center, and a little bit in automotive. So we think that right now chiplets are about 10% of the design starts. We think they're going to go to about 30%. But from the hype, you would think they're going to 100, and that's not going to happen. I mean, there are companies that are saying we're never going to chiplets because of the economics and the packaging and all this stuff. So, ultimately, 30% of the designs are going to be chiplet designs, and that's a very attractive market for us.

Okay, and then, you know, you've made a more recent push into, like, the microcontroller and edge and physical AI. Can you talk about the progress there? How meaningful is that? And, you know, you specifically called up physical AI as a growth vector. What are you seeing there that gets you excited?

Well, the key question in physical AI is, you know, what does the robotic future look like? Do we get humanoid robots? Do we get, you know, other type of robotics out there? Because if you look at the humanoid robots, they look like cars. There's an AI from an electronic perspective. They don't look like cars. I mean, they got hands and all this stuff, but There's an eight-ass chip. There's very complex MPUs SOC is kind of forth for the hand control and there's basically all kinds of motion control Microprocessors and FPGAs in there so there's a suite of Seven to nine chips maybe ten chips twelve chips that are going to be needed for this and if it starts shipping in volume This is a very attractive market for us. And so the reason we went to the microcontroller market is that The the high end of the microcontroller market is becoming very sophisticated They start to have AI chips AI sections 32-bit microprocessors those kinds of things and so, you know, you asked about traction. So we announced the Infineon deal a couple years ago We announced on the last ring called the MIPS deal which is mainly a microcontroller deal, even though we have some common automotive and other customers. But MIPS is going to be heavily involved in the micro-MCU market. So we're sort of going one by one and showing some penetration in the MCU market. And it takes a while because these MCU families are designed like two or three every two or three years. But they're shipping in huge volumes, unlike the data center chips. So we think that it's an exciting market for us.

Okay, and we've got a couple minutes left. I wanted to ask about Sycuity. So that was a pretty noticeable expansion outside of your traditional interconnect IP into design stage security. Was that something that you guys thought you needed to add? Was it something that your customer said would be great if you had this asset and we could help bring it to market? Can you just any background on how that deal came together?

Absolutely. So security, so we always have functional safety, which is one of the reasons why we're so strong in automotive, and that's protection against environmental risk like gamma ray strikes and things like this. The customer says, we need you to protect the data movement part from man-made attacks. And you need to help us be ISO 21434 certified for cybersecurity. And so we we are expecting more security solutions from you. And so this was a response to this But security is a little bit more than that, right? So one is we want to make security part of everything we do and protect the data movement also from man-made attacks at the hardware level But it turns out that all of our two hundred and thirty-some customers active customers need security and so They basically So, this allows us, one, to sell to our entire install base, two, it allows us to enter the government market where we have been traditionally weak, and it allows us to deal with some companies who don't want to buy our system IP, but who need security. So, we're very excited about the potential for the Sycuity technology.

And how does the go-to-market differ? Is there just a traditional license and then there's a royalty, or is there no royalty It's an EDA model, so far.

Got it. But the kind of the great thing is right now, CyQ is being sold to the security experts in those companies. We would like to make it part of the verification suite. So basically, there is a cybersecurity sign-off, not just a functional sign-off and not just performance sign-off, but there's a security sign-off. And that would make the hardware assurance segment much, much larger than it is today. It would go from relatively few licenses per customers to hundreds of licenses per customer.

All right. Well, right on time. Well done Thank you so much Charlie for joining us. We really appreciate you coming to our TMT conference and best of luck Josh.

Thank you very much.

Thank you