00:00Let's get the sort of handset smartphone and memory situation out of the way, if that's okay.
00:06You know, clearly this is impacting outlook for the smartphone market across this calendar year.
00:12How do you think it showed up in the earnings that you posted?
00:16And would it have been slightly better if the situation were different?
00:20Yeah, good morning. Good morning, Ed.
00:22So first off, the quarter was amazing for us.
00:25We had not only record revenue, but record royalties, $1.24 billion in revenue, nearly $740 million in royalties.
00:36And if you just stack that up, that's 27% year-on-year.
00:40Our data center business is exploding.
00:42We were up 100% year-on-year, probably north of 100% year-on-year.
00:47So to your question, no, the memory situation isn't really impacting ARMS so directly.
00:52And you might ask, well, why is that?
00:56Well, number one, we are spread across a lot of different businesses.
00:59As I said, the data center business is increasingly growing for us.
01:02And we think in a few years, it'll be our largest business.
01:05Secondly, our products are used across the board.
01:09So what we tend to see in memory shortages is the bottom end of the stack starts to get impacted.
01:15And that's actually lower royalty rates for us.
01:18So as a result, we don't get impacted very much.
01:20So no, short answer is no impact this quarter.
01:24And we raised our guidance for the next quarter because, as I said, we're just not seeing that kind of impact.
01:30But I'm not denying that it's there.
01:32But for the ARM business, the impact is quite minimal.
01:36Rene, thank you for that.
01:38Data center royalty revenue grown 100% year-on-year, as you outlined.
01:41I get the same question for you from everyone, sell side through to buy side, which is give me a specific year.
01:49Mark it on the calendar when ARM goes from being handset to data center and it accounts for more of your business.
01:56Is that inflection point clear on your calendar, at least?
02:01It's very clear.
02:03And that's why in the earnings call, I said in a few years.
02:06As you know, we don't typically provide forward guidance on an annual basis.
02:11But we have a line of sight that says we see it coming.
02:15And it's probably coming sooner than we had thought.
02:18And the reason for that is the data center growth.
02:21But if you click below that, why is data center growing so rapidly?
02:25It is the, A, presence of ARM CPUs in the data center.
02:29We're now over 50% market share with the hyperscalers.
02:32Secondly, the CPUs that are being used are using more ARM CPUs inside the chip.
02:38192 cores, for example, inside a Graviton chip going from 96.
02:45Vera, NVIDIA Vera, going from 72 cores on Grace now to 88.
02:49So what does that mean?
02:51That means more cores, means more royalties and high growth rate.
02:54Well, why are there more cores?
02:56When you think about agentic AI and everything associated with agents moving workloads across
03:01systems, managing workflows, et cetera, et cetera, that's the kind of work only CPUs can do.
03:07So to your question, while we're not giving a date, we can see it.
03:11And it's coming sooner than we had thought, even probably six to nine months ago.
03:14And probably on the back of when you look at an alphabet increasing its capital expenditure
03:19by another, well, potentially $185 billion, Rene, is the AI bubble, this worry about AI
03:26infrastructure build out a thing of the past to you?
03:28Or are there still some bottleneck issues that you're concerned about?
03:31Yeah, so it's a great observation, Caroline.
03:34And yes, there's huge growth in Google's CapEx.
03:37Not all of that $180 billion is coming to us, of course.
03:40But what does that say?
03:42We continue to see investments in the data center, whether it's what Meta had announced
03:47through Microsoft announced, and now Google Alphabet, that exceed what people had originally
03:53thought was even possible, even a couple of years ago.
03:56If you go back to when we announced the Stargate initiative in January of 2024, $500 billion
04:03over a number of years, people looked at those numbers and said, how is that even possible?
04:06Now, 18 months later, it looks quite, quite credible.
04:11So, bottom line is, we are seeing such an investment because the AI opportunity, the way I like to
04:19think about it, is kind of the final frontier relative to technology.
04:23When you think about what could be beyond AI, it's really hard to imagine.
04:28So, also, the fact that AI has really had very little penetration, let's say, into large
04:33enterprise or into our health systems or things around drug discovery.
04:37So, there's a long, long runway.
04:39I think people are a little nervous about it, as decidedly so, because we've not seen
04:44it before, right?
04:45We've not seen the order of magnitude we're talking about.
04:47So, it's somewhat natural for people to look at it and say, well, gravity's got to hit it
04:51sometime.
04:52But then, if you intellectually look at it and say, my gosh, is AI going to be something
04:56that will change and transform how everything is done?
04:59Well, of course.
05:00And then, there's going to be investment to support that.
05:03And that's what we're seeing.
05:05Investment to support that.
05:06We mentioned Stargate.
05:08That immediately makes me think of Massa.
05:11And you were clear on the call yesterday that you've spoken to him.
05:14You don't think SoftBank is in any way going to sell down its stake in Arm.
05:17But how much is that an overhang in the longer term?
05:19And how much, actually, do you want to see more free float of your stock?
05:23Well, I have one very large shareholder who I talk to quite frequently who loves Arm and
05:29he's very long on Arm.
05:31So, he has no intention to sell at all.
05:34And I shouldn't even say anytime soon.
05:36I don't know when that would be, quite frankly.
05:38We do talk about it.
05:40He said he's got no interest in selling.
05:42How that impacts the float, et cetera.
05:44You know, that's not something, candidly, I spend a whole lot of bandwidth thinking about.
05:47But he's very, very long on Arm because he sees the enormous opportunity that we have.
05:54And if you think about what we've done since we've gone public, Caroline, in the fall of 2023,
05:59we told the world we'd be growing at 25% year on year for two, three years.
06:04And people didn't believe it was possible.
06:06Here we are now a couple years later.
06:08And we're beyond what we told investors during our roadshow.
06:12And things are even stronger than they were at that time.
06:14So, as a result, he sees it.
06:16He sees probably more of the data than anyone does.
06:19And that's why he's not selling.
06:21Rene, I want to go back to CPU.
06:23January 26th, Bloomberg speaks to NVIDIA CEO Jensen Wang.
06:27And he speaks for the first time about selling to the market via a CPU as a standalone.
06:34And that seems critically important.
06:36Because if you think about the pipeline for Blackwell and Rubin, right,
06:41those are GPUs coded in ARM-based architecture.
06:46And those building, they don't want to sort of have to cross-pollinate x86-based platforms with their ARM-based GPUs.
06:54It seems to be really substantive.
06:57Could you explain how much of a boost that was for ARM?
07:01And whether this idea that the Vera CPU goes into the market as a standalone product changes the trajectory at all?
07:09Because at the same time, NVIDIA talks up about how much content it owns proportionally in the overall system.
07:16Just take it and run with it, please.
07:18Yeah.
07:18So, first off, thrilled to see Jensen say that.
07:22They're a great partner.
07:23And Vera is a great product.
07:25And as I mentioned earlier, Vera has gone from now 88 CPUs inside it, CPU cores from 72.
07:33So, why is that a really good thing for us going forward?
07:37Well, primarily because when you think about the data centers and how they're evolving to from what is general-purpose compute moving to AI compute, a mix of training and inference, and those inference workloads are agentic, that's right down the center of the plate for what the CPU not only is good at but can only do, can only do.
07:57So, what that means is inside the data center, you're going to start to see movement towards a homogeneous type of structure where people would love to have the ARM stack running almost everything.
08:08It's just easier from a maintenance standpoint.
08:10It's easier from an upgrade standpoint.
08:11It's easier from a cost standpoint.
08:14So, and also the flexibility it can afford.
08:17You can build a tremendously efficient custom system based on ARM.
08:21The Vera Rubin platform, compared to Grace Blackwell, uses 6x the number of CPUs.
08:28And that's when you look at the storage, the DPU, the offload.
08:32That's a huge increase.
08:34So, yes, selling loose Veras, that's a great thing.
08:37We're super happy to see that.
08:39Briefly, Rene, you have such a bird's-eye perspective because you're working with NVIDIA, but also you're working with OpenAI.
08:45It was reported potentially was looking with you at your technology to be working with its own custom chip with Broadcom.
08:51You've got Meta looking at custom chips, of course.
08:53How do you see this evolving?
08:57Yeah, I have a lucky job.
08:59I get to work with just about everybody in the industry across the planet, whether it's in the Fab Foundry area, whether it is a chip company, whether it's an OEM, whether it's a software provider.
09:11We work and talk to everyone because ARM is just so pervasive.
09:15The common theme we see is continued investment in AI and not just at the data center.
09:22It's trying to figure out how do you run those AI workloads everywhere.
09:26How do you run them in wearables?
09:28How do you continue to make them more efficient in a smartphone or a PC or in physical AI, whether it's autonomous driving and or robotics?
09:35So we're involved in all the conversations and the common theme is heavy, heavy investment.
09:42And one of the challenges that we see in the chip world around this is, as you know, it takes a couple of years to build a chip.
09:49The IP that we developed that goes into the chips, maybe a couple of years in before it, trying to predict what the architectures look like four years after the chip has been designed, where the AI models are going, is really very tricky.
10:02That lends itself also, though, very well to ARM because we are programmable, we are flexible, and we are low power, meaning that wherever the chips drop, no pun intended, we should be in a very good position to be able to drive those workloads.
Comments