Good afternoon. Welcome to the 27th annual Needham Growth Conference. My name is Charles Shi. I'm the semiconductor design software and IP analyst here at Needham. Joining me here today is Arteris. We have Charlie Janac, the CEO of Arteris, here with us on the stage. Charlie's gonna make a presentation, and I'm gonna moderate the Q&A a little bit later. Charlie, the floor is yours.
Okay, thank you. Thank you, Charles. So, essentially, in terms of the highlights, financial highlights, we achieved $65 million annual contract value plus royalties, and we delivered $1.1 million of free cash flow, which is something we're focused on. Basically, what we do is we have, well, we obviously have a very strong customer presence. So, these are some of the customers that we can disclose in a Q3 earnings call, because right now we're still in a quiet period. So we've received orders from VeriSilicon. We've also received orders from Tenstorrent, which is a RISC-V player. And also from NIO, which is a Chinese automotive OEM. So they allowed us to disclose who they are. So we're very grateful to NIO for that. Our strategy is to announce about one RISC-V partnership a quarter.
And for this quarter, we announced a relationship with SiFive, which is one of the leaders in RISC-V processor technologies. So, in terms of revenue, we've essentially 11% growth year to year, and as well as growth in ACV royalties, and a kind of an interesting category called RPO. So basically what happens is when we book a deal, it goes into RPO. It goes into deferred revenue because we have ratable revenue. And then it goes is utilized in non-GAAP revenue measure, which is ACV, annual contract value plus the royalties that we obtain. And then from then it goes on to ratable revenue. So, it's kind of a, in some sense the RPO is kind of a canary in a coal mine where it basically is a predictor of future revenue.
We're also pretty interested in controlling operating expenses. So we've done quite well in doing that. But we don't just want to control operating expenses. We really want to control operating expenses while doing extensive R&D. So we really want to be able to deliver one new technology per year, which means we have to be working on two new products per year. So our expense control is basically framed within that goal. We're also focused on free cash flow, generating cash. We are very pleased with the system IP segment that we're in. And so the longer, the more cash you can generate, the longer you can essentially address this growing opportunity for semiconductor system IP. This is our guidance.
The ACV plus royalties, which is the non-GAAP measure, we're basically targeting somewhere between $63 million and $67 million for 2024. You know, basically free cash flow, which may be somewhere between negative $900,000 or so and positive $1.1 million. And an operating GAAP income loss of about $4 million to $5 million. Now, this sounds like a lot, but basically we're now recognizing expenses in a point in time, but we're recognizing the revenue over the term of the contract. If it's a three-year deal, we recognize that revenue over three years, but we collect cash within 45 days, right? That's why we're so focused on free cash flow and then let the financials be as they may.
I think this is just a detail of some of the financials. So how did we wind up here? That is interesting. Okay. Let me just kind of explain a little bit the fundamentals of the business without some of the slides. We are the inventors of something called network on chip, which is use of networking techniques inside semiconductor technology right now inside die, but also increasingly between die. And so these networking techniques have advantage of essentially being lower power, lower area, higher performance.
In 2020, we acquired a company called Magillem, which makes something called semiconductor integration automation, which is packaging the rest of the IPs in the chip, essentially controlling the connectivity or specifying the connectivity and also controlling the registers, which are the exit ports of the chip. So that basically is a separate class of IP compared to the IP blocks like processors, GPUs, memory controllers, IOs, and so on and so forth, which are essentially dominated by adjacent companies to Arteris in terms of the markets that we are focused on. Obviously every chip needs to communicate in order to be a chip. So this IP is essential to the functioning of these SoCs. And it's a separate category to the processors.
One of the things that's unique about system IP, it changes about 9 to 10 times per project and it lives within a project. When you start a project, there's no system IP. You have to configure all the networks and all the connectivity, all the system, address maps and so on and so forth very, very quickly within the project. Then you're basically finished. After the project is finished, then you basically have to start all over again. You can still use the same setup, the same Lego pieces, but you need to configure them in a different way. That makes the system IP category unique. We're essentially the leader in this. In terms of commercial players, it's us and Arm, and that's basically it. Our main competition is internal solutions.
Obviously, people who build billions of chips, their chips communicate. So some of the stuff is being done internally. But the thesis behind the company is that over the next five years, the system IP is becoming so complicated and so expensive that the market's gonna switch from about two-thirds internal, one-third commercial, to about two, you know, it's gonna flip the other way. It's gonna go two-thirds commercial and about one-third internal. And so that will drive further growth for Arteris and also for some of the other system IP participants. In terms of the market, we are essentially addressing pretty much every chip. There's few architectures that are not addressed by what we do, but they're, those are rare. But we're primarily focused on, on two segments, and I'll talk about a new one. We're heavily focused on machine learning, so AI.
We're in about 200-some designs, AI designs, and we continuously develop features for AI. One of the things that we announced in the third quarter was availability of something called tiling, which allows you to build a section of a chip and then replicate it across. We also have some additional features for meshes, regular structures that are used in convolutional neural networks and so on and so forth. These were features that were demanded by our AI customers, both for cache coherent and non-coherent networks on chip. AI is a big focus. We're in lots of AI chips, lots of AI customers. You know, we think that ultimately, the AI market is not a separate market, that basically all systems will have features for their AI features.
So we're not gonna be talking about AI as a separate category, you know, three to four years from now, but it's gonna be part of everyday life. The other thing that we're focused on is automotive. In fact, the way we got into AI is via automotive, because essentially ADAS chips, automatic driving assist system chips are AI chips, and they make decisions for, you know, whether the car is gonna switch lanes, whether it's going to brake, whether it's gonna do some kind of an avoidance maneuver or give you a warning. So those are machine learning chips, as well applied to automotive. But it's much more than that. We have a strong position in radar. We have a strong position in car modems. We're in infotainment, dashboard control, engine control, chassis control.
The thing that's happening in automotive is maybe two years ago, a car had, you know, one point one plus level driving capability, which is sort of super cruise control. And there were maybe three SoCs and about a million cars. By 2026, you're gonna start to get, or 2027, you're gonna start to get level three plus and even level four cars. And those have 20 to 25 chips per car and SoCs plus about 40 to 50 microcontrollers. And so the car's becoming a supercomputer on wheels. And ultimately, out of the 90 million cars that's being made a year, there's gonna be probably about 60 million cars that are gonna be essentially electronically enabled.
And it doesn't matter if it's gas or electric, even though the electric car seems to be engineered as computers on wheels from the very beginning, whereas with the gas cars, the electronics are just added on. But they still use the same amount, same amount of electronics. And so, automotive is a revolution. We are participating at every level of the value chain. So we actually provide our IP to other IP companies that target automotive. We target, we have many semiconductor customers. We have tier ones, which are the subsystem suppliers to the automotive companies. We have automotive companies themselves, and we either even have pretty good business with the ride-sharing companies. Some of them we can name and some we cannot. So we are participating across the entire, entire value chain.
And similar kind of technologies are basically used in drones, in trucks, in logistics delivery vehicles. So this is a very, very large business over time. It doesn't move quickly. It has more hype than it deserves at the moment. But it is gonna be. It's one of the historical trends. And it's the application of artificial intelligence to basically the transportation vertical, right? So you'll be able to have vehicles that make decisions by themselves without input from human beings. Now, it's not gonna be, as I said. It's gonna take a while because the highway driving test case is actually very manageable with the current technologies. So over the next few years, you're gonna see hands-off, eyes-off type systems on a highway. But inner city scenarios are much more complex.
My personal opinion is that the cities are gonna have to be redesigned for automated driving in order for there to be full reality. So some of the hype that you see around automated driving, I think it's going to take quite a long time to realize. The other thing, so automotive is going to continue. It's not gonna be a linear path, but it's one of the historical trends here, which is the application of artificial intelligence to automation or to transportation. The other thing that we announced is movement into microcontrollers. Microcontrollers are becoming very, very complicated or much more complicated than they used to be. And these are chips that are used to control machines, machine tools, medical equipment, industrial equipment. They're used in automotive, of course, in transportation. So microcontrollers are used to control machines.
The reason we actually have a number of our customers that already use Arteris for microcontrollers, but now we're making a concerted effort. What's driving us is that a lot of our customers are using us for SoCs. The discussion is, okay, if you want a bigger relationship, not only do our SoCs, but do our controllers, do our simpler chips. We're targeting kind of a relationship with the larger companies in this space, with the leaders. So we've basically, you know, anticipate that we're gonna have some significant success over the next couple of years in the microcontroller space. This market has its very own dynamics. The chips are lower cost, they're very high volume, they're highly optimized, they're latency sensitive, they're cost sensitive.
So we're gonna have to be developing features using our existing product, but developing new features for microcontrollers. So that's pretty much on the market side. On the ecosystem side, we continue to try to be the Switzerland of IP. The ecosystem that we work with are companies that make the processors, that make the EDA tools, that make other types of IP, the foundries which provide the libraries, and also the design houses that do design for other people. So this kind of ecosystem takes a lot of work. I think right now we have 86 ecosystem interactions with other companies. So that takes a fair amount of money and a fair amount of investment in order to make an ecosystem work for our customers.
Because since we're in the middle of the chip, we really have to work with other companies to create a bigger solution and make sure that our customers have tested products, and this sort of goes to the kind of issue of competition, so, you know, most of our smaller competitors have been purchased by others. Intel bought NetSpeed, Meta bought Sonics, Google bought Provino, and so the main commercial providers are us and Arm. Arm is obviously focused on other things besides just system IP, which only represents about 5% of their business, and so, you know, we have a very favorable competitive position, and there's also about nine other startups that are trying to do what we're doing, but now the problem is that the scale that you need to really provide a solution is very high and very expensive.
It takes a lot of people, takes a lot of cash. And the market that we serve is about $1.2 billion. And so there's not gonna be a lot of people spending $50-100 million trying to build solutions for a $1 billion market, successfully. So we think we have huge barriers to entry compared to others. It takes a number of years to design a solution that is proven. It takes a number of years to get a design in. It takes even longer to generate a royalty stream. So basically we think we have a pretty good barrier to entry compared to others. But our success is not preordained, right? You have to work at it, very, very hard in order to keep up with the request of customers.
So we have to do a lot of R&D. We have to do a lot of customer support. One of the things that we have is we have about 70 application engineers now, and that is probably the largest system IP support force in the world. We try to be focused on quality so that our AEs can really focus on helping the customer define their architectures and get their chips to market. One of the biggest things here is for our customers to be successful in producing chips with Arteris IP. And then, of course, as I talked about, this ultimately will generate, you know, we have to turn that into financial success, so our first milestone is to be cash flow positive.
And the second milestone, which will be achieved in the first half of 2026, is to be non-GAAP net income positive and eventually GAAP the following year, sometime, GAAP positive. So, it's a big enough market. We can become a much larger company. It's a valuable market. And we're essentially our goal is to be a leader in that and to be one of the largest IP companies in the world, as we go on this journey. So with that, I'd like to answer any questions. All right.
Thanks, Charlie. I actually think with our slides it's much easier to follow your story.
I'll tell that to our marketing group.
Yeah. Well, that doesn't mean we don't need the investor presentation. We do need that. Let's open it up maybe first to some questions, please.
As you're thinking about microcontrollers, you're gonna sell IP not silicon.
Correct. It is, I mean, you know, it's amazing, that some people who are very, very experienced don't understand that it's really hard, almost impossible to be both a chip company and an IP company. And a number of people have tried it, but it's extremely difficult. And, because in the IP business, you're basically helping customer deliver the chip. If you're a chip company at the same time, you're also competing with your customers. And so this is exceedingly difficult. So we are not gonna be a silicon company. The thing that we could become over time though is a chiplet company, right? Where we bundle a number of our IPs together into chiplets.
The one that we're most qualified to do is something called an IO hub. If you have a multi-die SoC and you have enough dies that do different functions in a chip, you need a traffic cop kind of IP. And that's composed mainly of the interconnect and some last level caches and things like this. So we could do that. That would be hardware. There would be hardware, but you can deliver that in a Lego form where you say, okay, here's a bunch of Legos that have been, you can assemble that yourself. You can give them fixed RTL, or you can even give them GDS II, but you don't give them silicon. Let somebody else make the silicon.
But being a chip company and an IP company at the same time, very difficult and in my opinion, not very advisable. But you guys have seen more companies than I have. So you may differ in that. I think I remember last year you said you have 77 patents. Yes. Still have about the same patent. No, we file about 10-15 patents a year. So this year is probably a few more. But we, you know, we don't ever wanna be a patent troll. I hate patent suits, haven't been involved in a couple of them. And so we're really trying to build a very strong defensive patent portfolio. That's what we're really doing. But we do file about 10-15 patents a year.
If you're in a business in a long time, eventually you wind up with a pretty large patent portfolio. Is that strong enough, do you think, or do you need, are there some more things you do as you try to expand into new areas? I mean, because 77 or nine doesn't sound like, you know, if you look at companies, other semi IP companies, that's a small fraction of their patents. That well, the cost that they would have would be much higher than defended. So just capitalization wise and IP wise, how are you gonna really compete? We have more than that in terms of patent properties. I think now it's probably pretty close to 200 in terms of patent properties.
But you know, one, I don't know who came up with this saying, but I really like it, is that winners sell, losers sue. Okay? And we are very cognizant that we wanna have the best solution. And my goal is to have such a good solution that I don't have to sue anyone. But usually companies that lose their position start thinking about suing people with their patent portfolio. And we're not in that space. Are your customers the big ones like Samsung? They are. So Samsung is one of our largest customers. Intel is one of our largest customers. We have, I think, something like 780 chips that have been built with our IP. And there's, I think, 230 active customers right now doing projects. So we have a pretty broad customer base.
So, do they pay you based, is it on a unit basis or is it just a fixed royalty for? No. So, well, the royalties are. There are various models, but essentially when you start a project with Arteris, you pay us a license fee. Yeah. And then when you go into production with that chip, you pay us a royalty. Based on units? It's based on, it's a per unit royalty typically. With a cap or? Oh, no, no, no. Caps are the death. So when a salesperson comes in with a royalty cap proposal, we usually send them back to negotiate further. So no, royalty caps are bad. So you don't have any royalty caps in your? I cannot totally say that, but no, generally no. Yeah.
The reason, one of the things you have to consider is that if you're spending, I think, you know, it's public, we're probably spending $70-$75 million a year, something like this. Companies that have a high value, high capital, high development cost IP, you have to have royalties, otherwise you eventually go out of business. Right? So that model has to be maintained in order for Arteris to keep growing and maintain its position. If you have a, you know, SiFive tried it, they basically said, oh, we're not gonna charge royalties on our RISC-V processors and their financials look like hell. Right? And so you pretty much, every high cost IP, high capital development IP needs royalties in order to maintain the growth and be able to do the R&D that is necessary to evolve the product.
So for example, every product line that we have, and we have a couple product lines, we enhance it, we make two enhancement releases a year. So all that stuff has to get funded somehow. And an IP company goes out of business, doesn't do anybody good, not the customer and not yourself and not the investors. So you have to know how to run a business. Yes, please. Are suitable nodes? This is basic questions. And does NoC cover advanced nodes as well? Yes. So we start, we go from architecture to RTL. So we need the advanced libraries, but we don't do physical design. So we generate, the RTL then goes to synthesis and it's going to the design flow. Now, one of the features that we've shipped last year is physical awareness or second generation of it.
So we can estimate physical effects in it when you design the NoC. But we have customers that are on five nanometers, seven nanometer, eight nanometer, 12 nanometer, 14, 16, 28, and there's also now discussions of three nanometer designs. So we can handle all of that with our product, but we don't get into the specific physical library issues that companies like Synopsys would have to handle. Right? We can use the three nanometer libraries, but we only use relatively few parameters from those libraries in order to design the NoC. But the NoC works fine at, should work fine at three nanometer. Any other questions? Yeah. Average revenue per customer? So the average revenue per project at a customer is right now running at about $560,000-$580,000 per project. And number of customers have multiple projects.
So that's why it's kind of hard to, hard to say. But the project ASP is about 560-580, and it's going up because these chips use more IP and they use more new products. And the chiplets are gonna make it even more complicated because now you not only have to have on-die interconnect, but you also have to have intra, inter-die interconnect. So the ASP's gonna go up even further with, with chiplets. Those, those project revenues you mentioned, are those not inclusive of royalties? Not inclusive of royalties. Yeah. This is just an initial license fee. So typically what would you expect in terms of a royalty flow versus the initial? Depends how many units the customer sells. Yeah. I know. But, so, so I, I really can't give you a figure for that, but you know, if you're obviously, huh? More or less than? Directionally, yeah.
More. More. Yeah. It's more than the license fee typically. But it depends. If you're building something like a Pachinko machine or a server chip, but versus building a smartphone, right? So it just really depends on the volume. But yeah, it's more than the ASP typically. And the reason we get that, we're one of the few royalty bearing IPs because our goal is to offer a 10X payback to our customer. So for every dollar you give us for the project, we wanna save you $10 over the life of the SOC. And that's why almost very few people switch from the technology. So once you adopt an NOC, you don't go back.
Yep, please. Basic question again.
So what is the use case for lagging nodes or traditional analog, like more like lagging microcontrollers?
So overlooking like the most advanced one or like? So you know, I just talked to a customer recently, and their microcontrollers are on five nanometer. Okay? So it just depends. But there's a couple of reasons for using trailing edge nodes. One is availability, one is cost because the 28-nanometer fabs are fully depreciated, right? So if you're building a microcontroller, your costs are less. Also what's happening in China is the U.S. is starting to do a relatively effective job of banning the Chinese companies from accessing anything below 12-nanometer. So you're starting to see lots of trailing edge projects in China because SMIC is able to produce 28-nanometer in volume. Right?
There's a geopolitical reason to use a trailing edge node. Also, there's a number of functionalities, which is why chiplets ultimately have a really good shot at being the mainstream methodology is you only need some parts of the chip to be on the leading edge process. For example, the processor, you want the maximum performance and the lowest power. You wanna put that on the leading edge node. But your memory has a separate process, which would be different than the leading edge process. Also if you're doing mixed analog digital chiplet, there's no reason to put that on 5 nm or 7 nm. Right?
So this is why the chiplets ultimately over the next few years ultimately are gonna become a major methodology where essentially the function has the process that it needs, not everything on the same process. So it's easy to change. Yeah. So exactly. Yeah. So for example, in a chiplet, it could be that the chiplet is reused continuously because it's not evolving, whereas your processing chiplet is changing with each project as you need more performance. So the chiplets give you a lot more flexibility. The only problem right now is that it's costly. Right? The ecosystem really isn't there yet, and the early projects are just starting. And because the whole thing hasn't been proven, the multi-die projects are a little bit more expensive.
Any other questions in the last six and a half minutes that we have? Yes.
Do you kind of have a target where you can get this company to what size in like three to five years? Is there?
Yeah. I mean, my goal is to get it to a hundred million within the next, I would say three years or less. Right? Revenue. Right? Bookings, we can probably get to a hundred million pretty quickly. But because of the ratability, it takes a while for that to flow through to the revenue line. But over the next three years, we can definitely reach a hundred million. And you know, I was employee number two of Cadence, right? And if you told me that Cadence was at some point gonna be $3 billion, I would have laughed at you.
I think you and I were there, right? We were there at 500-600 million, something like this. And it, I would have thought it was a joke. Right? So this is one of these companies, that like the EDA companies can just grow and grow and grow. It's not gonna grow that fast, but it can grow to significant size over time. So, when we hit 100 million in revenues, how much is your R&D expense gonna increase and how profitable will it be? So we are essentially targeting a growth rate on the revenue side of about low- to high-teens to low 20%. We're targeting the growth in expenses to be about half that. So if we're growing 20%, we're happy to grow expenses 10%. And eventually that produces a lot of financial leverage.
We're trying to make the expense line focus on things that make a difference, which for us is engineering development and customer support. That's kind of where we're happy to increase our expenses. So, you know, at a hundred million, you already achieve some pretty good leverage. So you would probably wind up with 30% operating margin, something like that. Any other questions? All right. All right. Thanks.
It's a very good Q&A.