A podcast about the business and market of semiconductors
Ben Bajarin (00:01.084)
Hello everyone listening! Welcome back to the circuit. I am Ben Beharren.
Jay Goldberg (00:07.663)
Greetings, internet. I'm Jay Goldberg.
Ben Bajarin (00:10.288)
We are joined by a wonderful guest, someone that we have known for a long time, and I know every one of our listeners will be very excited to hear from, Renee Haas, CEO of Arm. Thank you so much for joining us, Renee.
Rene Haas (00:25.719)
You are welcome. Hi Ben, hi Jay.
Ben Bajarin (00:30.168)
So let's dig right into it, the meat of the conversation. You know, when we, and we've done a number of episodes on Arm, which you may or may not have listened to, some were positive, some were not negative, but you know, maybe looking at some of the deeper questions, but they were always very well listened to episodes. So like, my point is there seems to be a lot of interest in Arm, and so I think having you...
sort of broadly talk about this is going to be really interesting. But from some of our broader conversations with industry, even with you, I do feel like there's, call it still a misunderstanding about arm in the market today. And so maybe I'll just, I'll just jump off and let you kind of just clarify, right? How do you think about the company? How do you think about arm? It's not the same arm for, of, of the past. This is a new arm marching forward to a new future.
And I know you maybe sometimes feel these misunderstandings, so maybe just lob that out there. Like, what do you think some of those misconceptions are and how would you address those?
Rene Haas (01:34.974)
Yeah, well, thank you for giving me that opportunity. For a 33-year-old-ish company, it is one that, as you said, is not always super well understood, both in terms of what we do, what we deliver, what our products are. So I'll try to cast a little bit of clarity on that. The way I like to think about the company is we are essentially the brain that powers everything in your life. When we look at the world's population,
Jay Goldberg (01:47.355)
deliver what our products are. So I'll try to cast a little bit of clarity on that. The way I like to think about the company is we are essentially the brain that powers everything in your life. When you look at the world population...
Rene Haas (02:05.13)
and the amount of people who touch ARM in some way, shape or form, we estimate that probably 70% of the world's population probably touches ARM. And if you think of every electronic device needs a brain to operate and the name of that brain and electronic device, the CPU, that's what ARM does. We do CPUs and we are, I think, probably the most pervasive popular CPU that's ever been shipped or delivered or used.
Jay Goldberg (02:08.971)
We estimate that probably 70% of the world's population probably touches on it. The thing that every electronic device needs to believe and operate...
Rene Haas (02:34.494)
And probably in our lifetimes, we'll probably maintain that. We are in cell phones, we're in laptop PCs, we're in automobiles, we're in data centers, we're in IoT devices like thermostats and security cameras. And we were talking earlier about the camera on Jay's desk. It probably has arms somewhere inside. So we're literally everywhere. And the CPU is the brain that essentially allows that electronic device to operate.
What people get a little confused about is, okay, you do the CPUs, but do you actually build anything? Do you build chips or do you build systems or do you build these devices? And we don't, our business model is we design this CPU, but we then take that design and we license it, meaning that we have a contract with someone who ultimately does build a chip and we allow them to basically,
Jay Goldberg (03:20.407)
design and we license it, meaning that we have a contract with somebody who ultimately does build a chip, and we allow them to basically take that design, put it in a chip that they like, and then that is our big good model.
Rene Haas (03:29.986)
take that design, put it in any chip that they like. And then that is our business model, which essentially allows us to license the product. And then the other part of our business model is we have a royalty that we collect on each unit of chips. And Jay and Ben, I'm gonna hold for a second here, Samuel, your screen went a little batty on me here.
Jay Goldberg (03:39.379)
License the product.
Rene Haas (03:56.534)
Sorry, hopefully you guys can edit. Hopefully you guys can edit from that.
Ben Bajarin (03:57.416)
No worries. We can fix. Yeah, no problem.
Jay Goldberg (03:57.439)
Okay. Yeah.
Jay Goldberg (04:02.063)
So I understand the importance of what you're saying. I agree with it. ARM is pretty much everywhere. And the brain is a pretty important function. But isn't there a disconnect then between the value that ARM is providing and the value that you capture? I think, like, I looked at the last earnings, you had revenue divided by the number of chips. It works out to something like $0.12 per ARM chip shipped in the world.
And I know that's not perfect math. It's not the right way to exactly look at it. But it seems like there's an awful lot of value created on top of ARM. Is ARM getting its fair share of that value? And how do you increase your share of the value capture in everybody's brains?
Rene Haas (04:48.526)
So one of the things that the company was very purposeful about in the early years was adopting a business model that what we best way to describe was very horizontal. So any end customer that's going to try to build a chip and put that brain inside it, that brain was pretty general purpose. And the business model associated with that was pretty broad and general, meaning that if you're putting that brain into a security camera,
or you were putting that brain into a high-end data center, the business model was the same. License fees were the same and royalties were the same. One of the things we began to observe, particularly after the company was taken private under SoftBank, we looked at some of the new markets that were growing pretty rapidly, cloud data center, networking, automotive. And what we saw was these horizontal brains that we had developed.
Jay Goldberg (05:39.195)
Thanks for watching!
Rene Haas (05:45.166)
for general purpose applications simply weren't enough for these new markets, meaning that for the cloud data center, we needed to develop something that was much more purpose built for these products, things that needed scalar instructions that could support multi-core designs, massive switch fabric. So we designed a bunch of new products called Neoverse. Same thing in automotive. These IVI systems, ADAS systems, they required functional safety compliance.
They needed safety islands. We ended up designing some very specific products for the automotive market. So by developing those new products, we then had the opportunity to essentially have the royalty rates commensurate with the value that we were bringing to the market. So the royalties for a Neoverse product are very different than what goes into a mobile phone. Same thing into automotive. So we think over time, by not only growing market share in these new markets,
but royalty rates that are more commensurate with the value delivery, that we're gonna be able to go and grow faster than the market. The other opportunities I think, Jay is gonna allow us to grow faster than the market, is delivering more of a solution to the end customer. Now, one of the things that's happening in our industry is that these SOCs are becoming extremely complicated to build billions of transistors.
Jay Goldberg (06:51.92)
The other opportunity I think is going to allow us to go faster in the market is delivering more of a solution in the end customer. Now one of the things that's happening in our industry is that the SOC is not going to be able to do that.
Jay Goldberg (07:10.216)
Billions of transistors, a lot of the time, is taken not only doing the design, but actually verifying the design.
Rene Haas (07:12.062)
A lot of the time is taken in not only doing the design, but actually verifying the design, verifying that these 128 cores that you've built are going to perform it. They are going to run at frequency. They are going to run at timing. That the way you've designed the interconnect is all just going to work. And when you take the fact that the designs are taking longer to build, and then the processing cycle times, that is after I tape out a chip, from the time it takes me to get that chip back from a TSMC or a Samsung.
Jay Goldberg (07:25.135)
design the interconnect is all of this. And when you take the fact that you're going to be longer to build, then the processing cycles, that is after I get out of the chip, you kind of think that they get that chip back from the DSMC or Samsung, or Intel, where it may have been 12 to 16 weeks is now 16 to 26 weeks depending on the complexity of the project.
Rene Haas (07:42.314)
or an Intel where it may have been 12 to 16 weeks is now 16 to 26 weeks, depending on the complexity of the project, you added a lot more time to the time it takes to get to money. So there is now an increasing demand to help us design these chips faster and get them to the market sooner, which gives us an opportunity for ARM to deliver full solutions. So back to this Neoverse application.
Jay Goldberg (07:53.627)
uh...
Rene Haas (08:11.646)
We can put together 128 CPUs. We can put together the interconnect. We can put together the fabric. All of that done, verified, and completed, so it just works. Because in these SOCs, there's a lot of other IP that isn't the ARM portion. So increasingly, we have customers who say, gosh, if you can save that much time off our development time, and in some cases, it could take over a year, or even more than a year off design cycle time.
Jay Goldberg (08:24.145)
So
Jay Goldberg (08:35.803)
In some cases, it takes over a year or more than a year on the second half to take that solution. And by delivering that solution, we've got even more attractive rules to write. So that will allow us, I think, to get a better value.
Rene Haas (08:40.782)
they'll take that solution. And by delivering that solution, we've got even more attractive royalty rates on those solutions. So that will allow us, I think, to get a better value going forward and more percentage of dollars that comes out of SOC. But I'd like to say, you know, in one way is the customer paying more money? Yes, but in other ways, we're saving them a lot of money because they're getting to market a lot faster. So that's really the benefit both ways.
We win and they win.
Ben Bajarin (09:12.72)
I think the clarity there in terms of the historical precedent being it was generally the same IP and you could take that same IP and go to all these different areas, including high ASP areas like you talked about and now productizing these segments better is probably one of the biggest takeaways that I hope people grasp because I feel like it's significant because you said product, right? But at the same time, we talked about that disconnect being, well, when they say product, it's not end ship, but it's that IP is actually productized.
to a end market. And I think that's a helpful way to think about, again, this business model shift, if I'm using maybe not a great term, but it is, it's an evolving business model for the modern world that stack, that time to market, but really that productization by category is probably the best way to think about how you're making now revenue increases on these new categories of products.
Rene Haas (10:05.198)
That's right.
I think that's right. Because when you think about what has been growing for ARM, our units shipped per year has definitely been growing. In some of the recent years now, you're over 30 million chips in a year that ship, that have ARM inside. Some of these ARM chips that ship, if you remember calling on ARM in the early days, guys, it would be remarkable if we had 16 to 30 CPUs in the chip.
Jay Goldberg (10:22.8)
Thank you.
Rene Haas (10:35.87)
Now we have some that have well over 160, 192 CPU cores. If you have more chip shipping and more cores that are shipped inside the chip, and you're getting better value on a per core basis and your chips are growing, you should be able to see the top line grow, which is why our royalty growth has really been the dramatic growth factor for the company. And it's not only the units that are growing, but it's the number of CPUs because more and more complexity per chip.
And we have better economics because we've now delivered a whole different value proposition, whether it's Neoverse or Automotive or these subsystems.
Jay Goldberg (11:05.36)
And we have the record.
Ben Bajarin (11:16.98)
Hmm.
Jay Goldberg (11:18.203)
OK, that makes sense. So you have a bunch of new products, new services coming on. What about new markets? What do you think are the important growth drivers for ARM going forward in terms of end market industries? Autos, servers, data center?
Rene Haas (11:38.158)
We've got a couple of things going on here. The traditional business that we're well known for has been in mobile. But the areas that we're pretty excited about in terms of growth adjacent to mobile are PCs. So we've taken a lot of market share in one particular operating system. There's another ecosystem that is now starting to move to ARM in a larger way. We're very excited about the opportunity there.
We talked about the cloud data center, also networking, automotive and IoT. The other thing that's layered across all these markets, of course, is AI. And AI has different meanings depending on what you're doing in a different market. If you're talking about cloud data center and you're talking about generative AI and you're talking about training, increasingly, that is a place where ARM is going to have a footprint. Just need to look at...
NVIDIA's Grace Hopper design, where they're now integrating 72 ARM CPUs or in their Grace Superchip CPU, up to 144 CPUs that now work in conjunction with their GPUs for a very, very high-performance training engine and for inference. I think what we're also going to see in the Cloud data center is these general purpose ARM CPUs being used for inference.
You're seeing again, big announcements that have come from Microsoft with Cobalt or AWS with Graviton, they're Graviton 4, increasing usage for ARM in the cloud data center. And again, that's AI inference that can run on ARM. And I then think that this AI piece finds its way everywhere. You're going to have to have some level of capability inside the machine.
to run these workloads. And I think as you start going away from the cloud towards the edge, it's still early days as these LLMs are being developed and such. But just look at the recent announcement made by Google with their Gemini platform, where they talk about Gemini Ultra all the way down to Gemini Nano. Gemini Nano is running on a smartphone. So Gemini Nano is running AI algorithms, running that on an ARM CPU.
Rene Haas (13:52.534)
So I think AI is going to be yet another growth driver for the industry. I think at Stacey Obvious, where you see everything that's going on. But I think as it starts to find its way down across different endpoints, particularly around what inference looks like, what we see is going to be a need for more and more CPU to be used for some of these workloads. Because generally, the applications don't really know exactly how to use other hardware inside the apps processor.
So most of this ends up defaulting onto the CPU. And the CPU is a very good engine for all that. So I think AI is going to be another growth driver for us unquestionably.
Ben Bajarin (14:27.356)
Hmm.
Ben Bajarin (14:31.624)
Hmm. So before I want to make a couple of clarifying questions on AI, but before we do that, I want to maybe just pick your brain on how you think about the size of kind of this upside for things like autos and servers. And so one of the way, when you look at people build these models, they say, you know, and the number is debatable, but they'll say that the content in like cars or the content in semi-conductor in servers today is X and it's growing to, you know, some big number, right?
a silicon land grab as we digitize everything, right? So more silicon content in all of these areas. Now tying that back to how you think about productization, do you view this the same way that as more and more dollars go to data center and go to auto that this approach, this productization approach for you to develop IP that goes after those specific segments is an easy way to kind of see that revenue upside to you as that silicon content.
or whatever grows.
Rene Haas (15:32.59)
I think so. I think the trends are all moving in the right direction for us because if you think about the broader semiconductor market of call it $500 to $600 billion or whatever the number is today and projected to be a trillion dollars by the end of the decade, that's not a stretch at all. If you look at everything going on with AI and finance way into all workloads. So then you extend it out into, well, a large part of that trillion dollar TAM.
is going to have to run some level of operating systems, some level of applications, some level of AI workload. Well, that's good for ARM because we're going to find our way into pretty much all those areas. What we do know is that you start getting towards the end of the decade as these trends of designing these complex SOCs are probably going to get harder, not easier. You're still going to want to have a brand new galaxy or iPhone every year that's not going to...
Ben Bajarin (16:21.255)
Yeah.
Rene Haas (16:27.466)
that cycle is not going to come out and say, well, new phones come out every three years. So as the pressure on getting products out on the same cadence continues, the need for more and more compute continues and the opportunity for ARM to do more and deliver more of a solution. I think that all lines up very well for us, to be quite honest. You know, we kicked off this compute subsystem strategy in earnest after I took over. Oh gosh, it's going to be two years coming up in February. I can't believe it's been that long.
And we rolled it out towards the early part of our first year of me taking over. And I can tell you that now that we've done this, we are oversubscribed in terms of opportunity. We've rolled out a partner program around CSSes where we have lots of different partners who work with us. And I know we were going to chat about this, but I'll preempt it. One of the things that's been interesting about this
Ben Bajarin (17:09.999)
Hmm.
Rene Haas (17:23.021)
Clearly, OEMs who are starting their own chip efforts see massive benefit in this. Because if you're an OEM and you're going vertical and you're building on your own chips, you look at it and say, gosh, the ARM CPU complex, I don't have experience doing that. That's not something I've done for years. And you're ARM, can you just go off and do that? But we're actually seeing the silicon guys. And this is what people originally thought we were gonna see most friction with is silicon companies who were building their own SOCs and had built them on ARM for years
say, hey, this is the area that I've spent years and years optimizing on, this is my space. We're actually seeing a lot of those partners coming around and saying, hey, this is probably something that's best managed and done through the ARM ecosystem because time to market is everything. And if I can put my valuable engineers on other parts of the IP of the system that I own and develop and let ARM optimize on the part they own and develop, everybody wins.
And I can say the last big benefit we get from this is by doing the subsystems and doing the early work relative to the definition of the hardware platform, we can get the hardware out there faster so that software can get developed on it. Because at the end of the day, any architecture, whether it's an accelerator or a CPU, is only as good as the software ecosystem that essentially runs on its architecture.
Jay Goldberg (18:44.872)
Hmm.
Rene Haas (18:49.994)
So that's the other big benefit we get from doing these subsystems is that we can actually get out to the developer community much earlier than we have in the past.
Ben Bajarin (18:58.588)
Hmm. I think, I think that's a good point, right? That you made, and this is consistent with range of conversations that we've had, which is, you know, anybody that's doing, and in this case, like you said, is this is, this is more custom approaches wants to just double down on where they can add value and they can differentiate. And that's not always going to be the CPU cores. It's going to be something else, right? The custom basic, the AI accelerator, et cetera. And, and, and I'm curious on your, your take on this.
Jay Goldberg (19:01.627)
Thanks for watching!
Ben Bajarin (19:24.476)
before we move on from this segment, but we keep hearing, and I'm just curious if you hear the same thing, because if so, it feels like this is an even bigger opportunity for ARM, that a lot of auto companies are moving to more custom solutions. Now again, that might not be everything, but it's gonna be some parts of that stack. And I think that's perhaps an underappreciated reality, that this, what we're talking about, time to market, verified designs.
focus on your differentiation for a industry that's quite large. And we'll have, again, a lot of money going to cars actually taking less of a merchant Silicon approach. If, if, if that's the right even term for me to state it in, but still building their differentiation based on, on arm product or arm IP.
Rene Haas (20:08.458)
Yeah, when you think about what's going on inside the automobile industry, two gigantic inflection points are taking place. One is obviously the transition away from combustion to EV, which is an inevitable but painful transition, just given how combustion vehicles are designed and built.
That is a gigantic opportunity to move to something that is going to be more and more arm-based because of all the things that we do around the electronic control units and the subsystems and the batteries. Then you get into the situation around ADAS and IBI. These are the differentiation pieces for automobile manufacturers, far more than other parts of the vehicle. And even things such as supporting stock.
Jay Goldberg (20:40.655)
then you get into the situation where I paid out, I paid out.
Rene Haas (20:56.166)
interfaces such as Apple CarPlay or Android Auto, you read about the automobile OEMs wanting to control that part of the interface, wanting to really have what that entire digital cockpit experience looks like. That's a great place for ARM because, back to the software ecosystem, which is really the strength of the ARM architecture, the software ecosystem that...
essentially runs the backbone of these vehicles is all largely been optimized and tuned to be performing on ARM, whether it's Android Auto or automotive Linux or other areas, whether it's RTX. All of these have been optimized to run on ARM. So the automobile manufacturers, yes, they want to do something a little bit more vertical, and they want to stay on ARM because of the software ecosystem. And these subsystems are of a great benefit to the ARM.
Ben Bajarin (21:32.892)
Mm-hmm.
Jay Goldberg (21:41.219)
So the automobile manufacturers, yes they were.
Rene Haas (21:52.706)
excuse me, to the automotive players, because that's just not areas they've spent, you know, decades optimizing.
Ben Bajarin (21:59.676)
Hmm. So we'll sort of go back to AI for just one quick point, right? And I want to hit this on the perspective of again, how you design IP and it's a product, right? So do you think AI is just not necessarily its own product when it comes to ARM IP? Or is that the NPU? Like, do you think there's, yeah, we're, you know, ARM SOCs have this great, you know, bit of NPU IP they get from ARM. And so that's kind of the...
product that we build for AI or, or should we not even think about it in productization because it's going to fit into every architecture. And you're just going to build that into all, all platforms, for example.
Rene Haas (22:40.69)
So we do a standalone NPU today, and that NPU is essentially very, very good at matmols and transformers. I think the other thing you're going to see is workloads that can be pushed onto the CPU wherever possible, particularly where you've got issues around space and power. What I think will happen over time is...
And again, if you think about the notion of accelerated computing where there needs to be an accelerator and then to some extent the computer itself, I think that is an architecture that as let me say it this way. I think that is an architecture that will be somewhat pervasive over time. So you can expect ARM to be quite active in that market, aka accelerators plus a CPU. I think both will be fairly general purpose in nature.
Because I think one thing that we're learning, particularly with these LLMs and then video-based models, multimodal, that it's very, very hard to design very, very specific engines that are very, very good for one kind of model. So general purpose generally is a very good way to go, so long as you're managing the memory bandwidth and how you ingest into the engine, et cetera, et cetera.
Jay Goldberg (23:46.749)
It's very, very hard to design. Very, very sophisticated.
Ben Bajarin (23:52.911)
Mm.
Rene Haas (24:02.646)
So I think for us, our strategy, and we've seen that with CPUs, that over time we've added a number of extensions and things that can get run into the CPU that will help on the offload. And in some cases, you may want to have a dedicated accelerator. I think one of the strengths that we bring to bear, whether it's the accelerator component over time and or the CPU, is because we are essentially a platform, it makes it very, very easy for developers.
to be able to manage on top of that basis. So that's where I think you're going to see sort of the market going. AI on one level is a workload, end quote, but it is a workload that depending on what you're doing might be very good to run on accelerator and or could be very good to run on the central CPU.
Jay Goldberg (24:53.069)
Hmm.
Ben Bajarin (24:53.831)
Nice.
Jay Goldberg (24:55.895)
Hmm. So I want to shift gears slightly. Since it's 2023 and we're talking about semiconductors, we're required to talk about China. How is I mean, there's a lot, obviously, there's always lots going on with China regards to semis right now. We've heard a lot of seems like there's a growing amount of friction with a lot of big US chip companies worried about losing revenue in China worried about
that dynamic, how is ARM dealing with the US sanctions on China? And how does that sort of, how do you think that plays out?
Rene Haas (25:35.022)
I think one thing that's different about 2023 and maybe 2022 is lumped in there in 2021, certainly compared to five, 10 years ago, is CEOs who knew the first names of everybody who were in the cabinet relative to defense, commerce, who's in export control, one of the latest restrictions coming out. We've all become much more expert in this area than I ever thought any of us would imagine as CEOs, and I'm no different.
talk to government folks with a much higher degree of frequency today than I'm sure my predecessor did. And I know that anything from an arm standpoint that we thought about it before. So it is, I would say on one level, J, arm is not any different than pretty much every major semi-co and that we have to really have our eyes wide open on this. That being said, how does it impact arm directly?
Not as much as some of the other companies, particularly simply around the fact that a lot of our product is not designed in the United States. And because our product is not designed in the United States, the export control restrictions are a little bit different. But certainly whenever a restriction is laid out and put out there, we comply 100% if it affects us.
And we're always sort of looking around corners to make sure we understand where that where the puck is going on this thing, but it's, it's tricky, uh, for sure to navigate because the rules change often and it is a, it's a pretty high flash point for us, our China business. I think now is about 20%, uh, ish, maybe a little bit less. Um, our last quarter was very, very good in terms of the, the non China, non China area, um, but, uh, it's certainly like every other tech CEO, it's a
Jay Goldberg (27:15.717)
Yeah.
Rene Haas (27:25.326)
It's something I gotta pay attention to.
Jay Goldberg (27:27.531)
And I mean, arm in China is obviously, there's been a lot of history there, which we don't have to get into, but it feels like there is, there's a lot going on in arm China. One of the things I think there was a little bit of disruption over the last five, six years, so much was going on, both in terms of what was happening specifically in arm China, but also happening with arm changing ownership and.
in Nvidia coming and leaving. My sense is that the this big wave of new China chip companies, a lot of them moved away from ARM or never, never really looked at ARM. They're new companies. They couldn't afford the license or they just, they were worried about IP or US restrictions. And so it feels like a lot of these companies, especially in IOT and embedded went with risk five.
Jay Goldberg (28:28.427)
I mean, do you agree with that? Do you disagree with that? Am I like, how is that dynamic shaping? Like how much, where is ARM doing best in China in terms of end market segments? Where are you really seeing the traction with customers, with licensees?
Rene Haas (28:44.33)
One of the biggest areas of focus that we put into the company after the SoftBank acquisition, kind of going back to the Neoverse and automotive examples, was in addition to very, very specific product focused areas that were very, very purpose built, was around where the software ecosystem matters. So if you think about the server ecosystem,
having full support of all the Linux distros. There's now hundreds of Linux distros that are fully supported and performing on ARM. Kubernetes, all the virtualization software. Back on the automotive side, we started an initiative around SOFI, which is our code name for an initiative that allows people to define cloud-based interfaces that will make it very, very easy to develop automotive applications.
So we've really focused on the software ecosystem. And if you then go to the question of, well, where are we successful in China? Those two areas have been very good for us. People who are doing stuff around the cloud data center, we're seeing very, very good market traction in China. Very similar to what we're seeing in the rest of the world. We have deployments with Alibaba, Tencent, other major cloud manufacturers.
everything going on with electric vehicles in China. All of that has been very, very good for ARM, again, in the same areas that we've seen across the rest of the world. And that sometimes is something that people don't always understand or maybe can intellectualize relative to China, that they actually tend to look a lot like the rest of the world when leveraging a global software ecosystem.
Jay Goldberg (30:08.955)
Thanks for watching!
Jay Goldberg (30:29.125)
Yeah.
Yeah.
Rene Haas (30:32.066)
when you look at mobile phones, obviously of iOS and Android, that that's what runs in China. So it's not then unnatural to think that ARM is successful in similar markets in China. On IoT RISC-V, I think if there are applications, particularly in some of the deeply embedded areas, where a software ecosystem is not critical, you're not running applications, you're not worried about security, you're not worried about external...
software environments, then RISC-V is just fine. In fact, that's where we see most of its traction, either in China or in the rest of the world. It's an area where typically the programming interface or the interface in terms of developer, it's not exposed. So someone using the chip doesn't know it's RISC-V until the marketing blurb come out and said there's RISC-V inside. That is probably the area where.
Jay Goldberg (31:28.302)
Right.
Rene Haas (31:30.446)
There's been some traction or at least the momentum in China. But to answer your question directly, it's really automotive and server where we've had a good growth.
Ben Bajarin (31:38.28)
Hmm.
So question on the emerging relationship with Intel for IFS. Obviously, it came out that Intel owns roughly 1% of ARM now buying into the IPO. And obviously, Intel Foundry would love to make millions upon millions upon millions of ARM chips at some point in time. So maybe just.
talk about that relationship and then also sort of moving to what I think is a really interesting question and doesn't get talked about as a part of that, which is how do you see ARM in a chiplet era? Because it feels like if we're talking about in servers or data centers, that some of these chips and automotive are going to get really big and we're going to put 50 chips on a die, for example. Well, that's more than just one ARM chip that could go into these things. So maybe just think about that because people don't talk about it, but I think that's a
is an even bigger sort of growth story if Arm has a bigger role in kind of the, the chiplet and advanced packaging world as well.
Rene Haas (32:43.242)
Yeah, so with IFS and Intel, we work very closely with them now. They have been working with us very closely to essentially expose their latest design kits around N18A for our advanced mobile designs. So we're doing a lot of work with our engineering teams to make sure that IFS is a first class citizen, an option relative to the ecosystem, choosing to partner with them, should they?
I think it's good for the industry. I think it's good for the U.S. I think it's good for geographic resiliency of supply chain, let alone geopolitical resiliency of supply chain to have IFS be successful. So we're very committed to working with them. Pat and Stu have been great. It may appear that's that strange bedfellows for Pat Gelsinger and I to be talking about how to make ARM more successful.
But I think he realizes that for IFS to be successful building logic chips, it would make sense that he has access and is working very closely with us. So we are, we're working really closely with them. We are friendly competitors in the sense that, you know, they think they have a position in the server market. We have a position in the server market. We talk about laptop chips as well, but I think broadly speaking, they're very serious about making IFS successful and so are we.
Ben Bajarin (33:43.325)
Mm.
Rene Haas (34:05.538)
And it's been a great partnership so far. To your question on chiplets, that's an interesting one, right? So if you go back to the idea of doing these compute subsystems, and these compute subsystems have lots and lots of ARM IP on them, you can imagine in a world where some of these subsystems go onto a chiplet, and you look at it and say, gosh, most of the IP on that chiplet is from ARM. Well, could ARM be doing the chiplet?
Jay Goldberg (34:27.119)
Some systems don't want to split.
Rene Haas (34:35.458)
Maybe we certainly could do something all the way to a GBS2 handoff, where it's taken all the way to a physical design. And not only have we taken the subsystem approach in terms of verification and testing and validation, but now we can take it all the way to the step of it's guaranteed to run at this frequency, run at this voltage, and we'll deliver this level of benchmark performance. We could and we may.
Ben Bajarin (34:41.035)
Hmm.
Rene Haas (35:03.662)
I think that's an opportunity for us. The world is moving fast. We'll move wherever the market needs to relative to getting products to market sooner. So it's something I think would be a really interesting opportunity.
Jay Goldberg (35:16.827)
Very cool. So we're up against time. I just want to I just have one question to wrap up with. It's something that I always wonder about, which is, do you have all the ARM product names memorized, or do you have to carry around a cheat sheet like the rest of us?
Ben Bajarin (35:17.435)
Nice.
Rene Haas (35:35.578)
I guarantee you I have to have a cheat sheet. We, in particular for our GPUs, our GPU design team is based in Trondheim, Norway, and they choose Norwegian gods as sort of the code names for these things. So unless you're a Norseman or a, or very, very well versed in what's going on in Norwegian lore, so there's no way on the GPUs, a little bit better chance on the CPUs, but no.
Jay Goldberg (35:37.403)
Hehehehe
Ben Bajarin (35:52.474)
Awesome.
Rene Haas (36:04.298)
A cheat sheet for sure.
Ben Bajarin (36:06.196)
That's awesome. That's awesome. Awesome. Well, thanks, Renee, very much for coming on. We greatly appreciate your time. We know you're busy. This was awesome, and we'd love to have you back at some point.
Jay Goldberg (36:06.315)
Yeah, that, yeah.
Rene Haas (36:18.075)
Sure, always happy to do so. Thank you so much for having me.
Ben Bajarin (36:20.764)
Thanks for listening everybody. Like, subscribe, say something nice. Until next time.
Jay Goldberg (36:20.92)
Thank you so much.
Jay Goldberg (36:25.903)
Tell your friends. Thank you.
Ben Bajarin (36:29.553)
All right, just.