Arteris, Inc. (AIP) Earnings Call Transcript & Summary

All right. Charlie, Nick from Arteris IP, a recent IPO that Cowen was involved with and a really good story about on-chip interconnect IP and the growth of AI, and particularly automotive chips for the industry. So Charlie, maybe we could just initially kick off for the benefit of the folks online as well. And I think we're going to continue to get people streaming in here given the elevator capacity is limited, let's just say. But anyway, it's great to see you both. Thank you for spending some time. And Charlie, if you don't mind giving a bit of an introduction to the company. Maybe not completely introduction, but just kind of an update on the business.

Yes. So we make connectivity components for essentially semiconductor-based electronic systems like cars, TVs, computers, those kinds of things. And specifically, what we do is we deliver network-on-chip with the pioneers of using networking techniques inside semiconductor chips. And additionally, we make software that allows our customers to deploy their IP libraries. So we call it IP deployment software. And together with the NoC interface IP, there are some specific blocks that connect to the network-on-chip and make the chips to work better, that makes up our system IP business. So we're a system IP company. So we're designing to about 70% to 80% of all ADAS chips for automated driving. We see strong demand in machine learning, 5G infrastructure, consumer and industrial. And we've -- basically are in about 3 billion chips -- 3 billion systems using our products. So it's quite a bit. So the drivers of our business are basically complexity of the SoCs that is increasing dramatically as SoCs are allowed to make decisions rather than just process data. There's also what I call the democratization of the SoC business, where the IP methodology has allowed system houses to be more successful in building chips. So there's more companies building chips through our systems. And used to be that the semiconductor companies put out a chip and the people would program for it, now people build a huge software structure -- infrastructure and they build chips to support that software. And of course, the globalization is dead. So you've got regionalization of the semiconductor industry between the U.S., EU and China. And finally, a lot of the players only do a portion of the design cycle. So I call it disintermediation of the semiconductor design cycle and that speaks well for IP deployment software. So that's our system IP business, and we think we're the leader in that space.

And just to add to that, Charlie. I mean one of the things that a lot of you don't realize is the complexity of NoCs, networks-on-chip, is increasing dramatically every single year. And so that pushes people our way, to our solution, because you can't do it with simple discrete dies anymore. And there's a scarcity of NoC engineers, interconnect engineers. And so it's very hard for people to build their own internal teams.

No, it makes sense. I think there's a lot to dig into with your company, especially given the evolution in AI chips and also in the automotive market, in particular, going on with your customer base. So my observation is that in the automotive market, all of your top customers are going through this transition from being, what I would call, microcontroller-centric companies to SoC-centric companies, right? And if you're building an MCU and you've been building one forever, and it's kind of the same MCU over and over again, you probably don't need a whole bunch of interconnect IP. You make that transition to a much more diversified SoC that needs more capabilities at higher process nodes and, all of a sudden, there's a build or buy decision that comes into the play for the customer. They need the transitions happening quickly. They need on-chip IP. Is that the driver of the business basically, is they need that IP and you guys have proven at scale that the IP works and can provide them that service over a long period of time?

Yes, absolutely. So we see 2 types of teams in automotive, right? One is the team that has been building automotive MCUs forever and their next project is a system-on-chip, with a processor, machine learning sections and all that stuff. And that's a very, very big step to go for those kinds of small devices to an automotive SoC. And then the other team is people who are used to building SoCs, but they don't want to think about automotive. So they want functional safety. They don't know anything about kind of the automotive standards and those kinds of things. So we can -- we essentially kind of bridge that gap, and we deliver functional safety-oriented technology called Resilience, which allows the chip to meet the different levels of ISO 262 (sic) [ ISO 26262 ], Standard ASIL B, C and D, depending on how far they want to go. And so we've been early in developing those technologies. And so we're designed into -- I think we have about 36 automotive customers, and there's more and more coming into the market because it's the biggest opportunity in tech over the next 20 years, I think.

Yes. I mean you've shared through the whole IPO process, and even since then, you guys have updated sort of the logo chart of where you are in automotive. But just for this audience, if you could just kind of remind folks the diversity of the customer base. I mean it's kind of the who's who in auto semis.

Yes. And so we have, as I said, about 35, 36 -- 36, I think, automotive customers. So some of the names you would recognize are Mobileye, NXP, Bosch. We have a number of leading Chinese companies, Black Sesame, Horizon Robotics, Cambricon and SemiDrive, right? And a bunch of smaller companies that are doing automotive. And then many that we're not allowed to announce, right? So there's ride-sharing companies. There's automotive OEMs, of which we've been only allowed to announce, one of which was BMW, which we announced in the last quarter's earnings call. There's Tier 1s and there's semiconductors, and they're all designing chips in some form or the other, because the car is becoming a supercomputer on wheels.

So we'll get back to the BMW conversation here in a second. But I think it's important that -- for folks to understand the consistency of what you guys have been doing with some of these automotive customers as well. It's not just newly won designs. There are certain flagship and large customers in your automotive base that have done 3, 4, 5, 6 generations of products, have consistently used you guys. One of your largest customers was acquired by Intel and is now potentially going outside of Intel, at least partially, and they had opportunities to explore different types of IP when Intel bought them and very consistently stuck with Arteris for a long period of time. So maybe you could talk about that track record, what complexities happened with your customers over time? What the ASP for your solutions has done over time and sort of the stickiness of the IP?

Yes. So the technology is very sticky, particularly in automotive because, basically, the interconnect is the data highway, the data backbone of the chip. And so if you were to replace the interconnect, you actually have to requalify the chip. So the only time you can ever come out is between generations. And we have the largest support force in the world for system IP. We have consistently delivered one new product per year for -- some of that for the automotive market. We make 2 to 3 releases of every major product, update releases per year. So we've been able to address the needs of those kinds of customers. And consequently, we don't lose customers. Mobileye, we've been in their chips, every one of their chips since 2011. We've been in multiple generations of NXP chips, multiple generations of Bosch chips and those kinds of things.

No, perfect. Nick, I wanted to -- I think a lot of the investors are familiar with some of the really large IP businesses in semiconductors, Arm being the flagship one. And there was always the licensing fee upfront for the IP, the royalty stream that would come later. Your business has very much those dynamics, and then you have the additional piece of the IP deployment software on top. So if you can just kind of walk folks through how the revenue model works? What the metrics are that investors should be following? And I use Arm as the flag post because it was a large cap and people were very familiar with the model. But if you could just kind of compare and contrast what the metrics are that we should be following?

Yes. So we have multiple streams of revenue. As you rightly pointed, we have license revenue, which, unfortunately for complexity, is different if you're selling interconnect where it's ratable, so it's spread over the contract term as opposed to IP deployment software, which is point-in-time because of the nature of its -- of the agreements currently. We'd love to move that to a ratable model too, but rat takes a little bit of work. But -- so that's the one piece. The point-in-time element of that can ebb and flow. And we've seen a really strong Q1, as you saw in the -- outperform a very strong Q2 in the guide anyway, sort well above consensus. And that's really down to the nature of the cadence of the quarters. The metric that we encourage people to look at which basically removes that -- the peaks and troughs -- more peaks, obviously, than the troughs, is to use what we call annual contract value, ACV, and we add on to that trailing 12-month royalties. So annual contract value is simply the total fixed fees under contract divided by the contract term. So if it's $600,000 3-year license, it becomes a 2-year $200,000 ACV. And then trailing 12-month royalty is kind of a proxy to revenue is the way we look at it, because it basically -- it smooths out the point-in-time revenue versus the ratable, and it puts on the same level playing field. And once you add in trailing 12-month royalties, then you have a true revenue proxy. So last year, for example, we ended up with around $38 million of GAAP revenue. ACV plus TTMR at the end of the year was $50 million. So they're quite different scales of numbers as well, but much more consistent. If you look at the development of ACV over the last -- I think we've published numbers for the last 2.5 years, it's a very, very consistent growth path, whereas due to a couple of very large point-in-time contracts in 2019, 2020, the GAAP revenue is a bit more lumpy [ on balls ].

One of the things that caused a little bit of those lumps was a couple of things in China, right? There's no -- and Charlie mentioned globalization is maybe not dead but under question, that's for sure.

U.S. dollar.

Yes, exactly. So the challenges were around HiSilicon and around a couple of companies in the drone market. So if you might just walk through those and assure folks that it might have affected comparables, but the effects are essentially behind you.

Yes. So HiSilicon was a May 2019 deal, added -- it was a $10 million point-in-time revenue in -- on that day we signed the deal in -- or delivered the software back in May Q2 of 2019. DJI was similar. It was added -- and this is because they joined the notorious banned entities list. And therefore, we were unable to support them in the field. It was illegal for us to give them support, FAE support in the field to develop the use of the interconnect in their chips. So those have to be taken as point-in-time. So if you look at 2019, huge revenue, huge bottom line, because it's pure $10 million right away down to the bottom line. Same thing happened almost at the end of December 2020, which is DJI. It was a $7.5 million -- just over $7.5 million deal, all point-in-time. So those are obviously bad comps when you get to the end of 2021. It looks like '21 over 2020 is not really growing very much, which is again one of the reasons why we encourage people to look at ACV. Because from an ACV perspective, both of those deals were treated as if they'd been ratable revenue essentially. They -- but the HiSilicon $10 million was spread at $3.3 million evenly over the 3 years, and it just dropped out last month.

Yes. There's also the effect of the China lockdowns a little bit, which affected some of the volumes. But China cannot stay locked down forever. So Shanghai has just opened up. And the principal impact on us was that, in China, the contracts have to be stamped with something called a chop, which is a stamp unique to each company. So it's not a signature that matters, it's the stamp. And so people were not able to get to their stamps because of the lockdowns. But that's all over. So we could have done even better than we -- than the numbers that we announced probably by a little bit. But that effect, I think, is going to be behind us, I hope.

Availability of stamps, that's a -- there's...

Ask Arm.

Yes. Yes, exactly. They have to find the stamp. The board is available. So Charlie -- thank you, Nick, for the conversation about the metrics. I think what's maybe more important to me is the pace of innovation and the product portfolio. You talk about bringing out new capabilities in the product portfolio year-over-year, things around cash coherency and other features that bring not just higher licensing costs, but hopefully higher royalties out the back end. How has the development been going? I know that the -- how is the road map, everything delivering on time, I assume? And then the hunt for talent in the Valley is competitive, let's say. So how have things been going in building out the team and executing towards the longer-term road map?

Yes. So in terms of the team, so we've been, in some sense, diversifying our risk because we now have about half of our engineers in France. We started as a French company. So we have a team in Campbell, a team in Austin and a fairly large team in 2 locations in Paris and a small team in Sophia-Antipolis. So we have some diversification there. The product portfolio is -- and we were adding management for those teams as well. So we've hired some very impressive leaders for some of those teams. In terms of the technology, things are getting very complex, right? And you now have cash coherency, you have noncoherent traffic, you have machine learning traffic, and soon, inter-chip links, where people are trying to get multiple dies to work as a single system. So that makes the interconnect even more complex and introduces fourth data traffic class. So we're essentially doing R&D for 10% of the SoC industry, right? So we've been delivering all of those products. One of the things that we're focused on is on talent shortage because there's just not enough interconnect engineers whatsoever out there, right? And the universities aren't producing any. There's some process, of course, a few courses like that, but nothing that -- there's no professor teaching system IP. So the only answer is automation. So our goal is to start automating some of these processes to a much higher level so that people can get their NoCs faster with less manpower or same manpower and getting many more NoCs to be done of different types. So automation is kind of the direction for the evolution of our product portfolio. And as I said earlier, we -- our goal is to ship one major new product every year. We shipped one last year. We're going to ship one this year, we're going to ship the one after that and to update our products 2 to 3x a year. We've enhancement requests from our customers.

So it's interesting, you mentioned the fight for talent and the lack of talent in interconnect. That could be a challenge for you guys if your bigger customers or potential customers use their balance sheet to hire all of those people. But it also can be a huge tailwind if each of them can't find people, and they just like, well, let's push these folks and have them go to Arteris and the whole industry can benefit from that scale and consolidation and R&D outsourcing as it were. Where -- what's that teeter-totter look like?

Well, I mean -- so if you add up sort of last year's revenue, commercial revenue just in interconnect, not in IP deployment or NoC interface IP, you have about 170. The business has grown there. But the total market is about 600. So you have about 470 million of interconnect, right? So that's a big opportunity for us to address some of that. The customers, though, are having increasing difficulty keeping those teams together. And it was okay when you had to do one type of interconnect, but now you have to do cash coherency, noncoherent machine learning and inter-chip links in one company, working together as a single system for one set of products, and that's starting to be difficult. So we think that all this stuff is going to get outsourced. And the talent shortage is actually creating decisions in our customers that say, okay, where do I put my best engineers? Do I put them on the processor side? Do I put them on the memory subsystem side? Or do I put them on the interconnect and on the system IP side? And so I think the talent shortage is actually kind of working in our favor because we can deliver those solutions at a much lower cost than hiring your internal team.

And we have actually added some really significant people in the last, and we've done a little bit of recycling to upgrading in R&D in the last year, roughly, starting with the appointment of the COO from Qualcomm, who's really the sort of SVP of Engineering is a better way to characterize them. And he's managed to bring in some people, some ex-Mobileye people and others who are really, really top people, and they've got a great Rolodex. And so it kind of becomes sort of a virtuous circle in a way because the more you attract the top talent in, the more people from the outside see you as a good place to land. You just have to be competitive. You have to be competitive in some way, shape or form with the likes of NVIDIA or an Apple because -- and Google because they all want the same people. So it's not cheap. And that -- we do use our presence in France as a lever. So it's very valuable. It's about 2/3 of the cost of it and the same quality, generally speaking, as a Valley engineer.

That's really interesting. I call to -- I guess, just finishing this one. Like wage increases with everything that's going on with inflation and whatnot, have you -- Nick, have you kind of contemplated that in the guidance for OpEx? Are there still like employee cost increases that come just from a wage perspective? Like anything we should think of that there would be a step function there? Or...

We've had the step function at the beginning of this year. We raised, on average, around 5%, 5.5% globally because it's not just the U.S. has wage inflation, it's France, it's all across Europe. China is massive. And so, yes, generally speaking, wage inflation has been running at about 7% or 8% globally, as you know. Not all that's flowing down to all of these positions. But we have tried to stay competitive. I think it's important. So -- and that's built in -- to your question specifically, it is built into the guidance already.

Okay. Please.

I wanted to -- I know there's always so much you can say, but I want to ask about BMW engagement. How meaningful, I guess, are your engagements with the auto OEMs? I guess it's somewhat surprising to see them engaging with you at the level of technical expertise to be designing SoCs that complex. I know you don't really care who wins probably, but how meaningful are auto OEMs internal chip designs to your longer-term business?

Very meaningful. And BMW was actually the first OEM who allowed us to say that they're a customer or a user, but there's probably 5 or 6 others that we have as customers. And basically, the car is becoming a supercomputer on wheels, and it's becoming an endpoint in a network or of Internet of cars, right? So electronics and software, the chip design, architecture, at least, is becoming very, very important. The car companies are generally not designing chips, they're either designing 1 or 2 IP blocks or they're just designing the architecture, and then they're having their partners do the chip design, right? So this is actually not a bad news for the semiconductor companies for the design houses. But in a sense, they drive what is built, and they kind of determine what IP components are used in those designs. So for our strategy, dealing with the ridesharing companies and the car OEMs is very, very important. And also, you're gaining a lot of knowledge at the end-user system level, what is needed that can drive your development, right? So it's also about having a faster learning than other companies. So they're very, very important. And everyone in that automotive food chain is going to do some level of chip design.

And our approach is actually to have -- is a ubiquitous one. So we will address OEMs. We'll address Tier 1s. We'll address semis. And sort of saying we have 36 or even 37, I think, now, people in that space at all 3 of those levels. And to your point, we -- I mean, of course, we care who wins. We hope it's going to be the people that we service, and we want the good guys to win, but it's -- we are relatively ambivalent on that because we supply that whole ecosystem.

Just to follow up on that. You mentioned BMW is the one that allowed you guys to actually speak publicly. What's the hesitancy of some of the ones that you do larger volume with of being public about? You wouldn't think who they used for interconnect would be supersecret. So just wondering the hesitancy to be public.

The reluctance is that they also buy chips from semiconductor companies, right? They also are going to buy some chips from Tier 1s, who're starting to design chips, even though slower than they should have. And so there, they just don't want to upset the apple cart of their semiconductor supply chain.

Plus they're run by logos. I mean, frankly, there's a lot of logos involved, domestic but likely there. You try and get a logo consent out of Apple. Yes, I mean...

Good luck with that.

Never been done. So I think there's an awful lot of that. It's an easy answer is just to say, no, we don't give that.

And then to just touch on the whole conversation, you don't want to give it out.

Yes. But the real reason is that they -- some of them are afraid of saying, wait a minute, we're putting this investment in this OEM relationship, but you're doing your own tariffs. So that's probably why. But I think eventually, more of those relationships -- we're hopeful that more of those relationships would be allowed to become public.

No, that makes sense. Really quickly, I think we only have a couple of minutes left. We didn't get to the AI/ML space, and I think we should spend the time we have left on that. Like how is it different or similar to the automotive opportunity for you guys? And what's the pipeline look like there?

So actually, the ADAS chips are ML chips, right? They have to do machine learning, processing and those kinds of things. But the machine learning market is much broader. And we actually don't think that there is a machine learning market. We think that machine learning is going to be a feature of every product. It's going to make things a lot more usable and have people manage -- help with the complexity of dealing with these sophisticated systems. The reason it's different is that basically what you have is you have something called convolution neural networks, which is basically mesh configurations, and those have different traffic considerations. Using meshes, tree mesh, ring, whatever, is it the same? No. The peer-to-peer communications have some specific requirements. And so you have to deliver features for things like avoiding data packet congestion in certain regions of the mesh and those kinds of things, or helping people automatically assemble these very large structures and those kinds of things. So there are some specific features that have to be delivered for that market, and we're doing that. And ultimately, that is the biggest market, right? It's going to be -- and it's not going to be -- it's going to be so big that it's not going to be its own market, it's going to be a part of every market.

So the majority of the customers there are merchant chip companies trying to sell into those markets? Or most of them are -- the majority of the opportunity is sort of with the broad hyperscale companies look to build their own stuff?

Yes. But I would say the biggest license growth right now is from machine learning start-ups, leading specific edge, edge machine learning, SoCs for specific applications.

Super interesting. There's a lot that we could talk about. I think we're getting the red blinking dot of death here from the clock. But thank you all for listening. And Charlie, Nick, it's great to see you. Thanks for spending the time with Cowen. Hope you have a great day here, and we'll be in touch.

Thanks, Matt. We appreciate it.

Thanks.