Arm CEO Says AI Is 'Not a Bubble'

Bloomberg

Watch Arm CEO Says AI Is 'Not a Bubble' - Bloomberg on Dailymotion

Transcript

00:00Let's just dwell on your numbers because revenue is strong, up 34% for your fiscal fourth quarter.

00:04We saw more than a billion dollars being brought in and indeed a profit beat, but

00:07investors taken with what seems to be this cautious outlook. How uncertain are things

00:13from a demand perspective that also perhaps you're not giving that annual target that someone hoped

00:18for? Yeah. Hi, Caroline, and thank you for the kind words. Yes, we just came off an amazing

00:24end of our fiscal year, the first billion dollar quarter, which was a record. Record royalties

00:31over $600 million, record licensing over $600 million, and then we finished the year at $4

00:37billion, which was also a record for the company. For Q1, we're looking at very good royalty growth,

00:44actually, Caroline. We're projecting anywhere between 25% to 30%. On the licensing side, I think,

00:50as you know, the deals can be a little bit lumpy. So as a result of looking forward in terms of

00:56whether deals close in the June or July timeframe, we decided to be cautious and give the outlook

01:01that we did just based upon licensing. But the overall health of the business is really, really

01:06good. We have, obviously, it's coming off a great quarter, and we have really good visibility into

01:13the current quarter on royalties. The reason we chose not to give the yearly guidance, I think,

01:17as you know now, the vast majority of our business is increasingly royalties, and that really comes

01:23from people who make either the end chip or the end equipment. And that, in the next two to three

01:29quarters, is a little hard to predict. In fact, we're not seeing that guidance coming from our

01:34partners. So as a result, given that their visibility isn't as strong as it usually is around

01:40that, we decided not to go forward and give guidance beyond just this first quarter.

01:44And just for the people who always need to be reminded of the business model, you are designing,

01:50ultimately, people that are licensing using these designs. You are not directly impacted by tariffs,

01:56but of course, the people that make the smartphones eventually might be. Are you hearing anything about

02:00a worry on demand in the future? Are you thinking that we will see a downward trajectory for some of

02:05the electronics parts that you've diversified the business into?

02:08Yeah, thank you for pointing that out. Yeah, exactly. We are a provider of the designs that go

02:15into chips that go into end products like your iPhone, your Ford F-150, your earbuds, the AI data

02:22centers. So we're not directly impacted, but we are certainly indirectly impacted, and that would come

02:28on the royalty side. To your question, are we seeing anything relative to folks not clear in terms of

02:37where the tariffs are coming from? What we're seeing is kind of a lack of clarity as opposed to any one

02:43indicator or the other. So that's just really driving it. It's just the lack of visibility that

02:48our end partners have.

02:49René, have you any plans on changing how you actually charge your royalties? Maybe you'll move

02:55it to a per-device basis. And also, how much pricing power do you have in an inflationary environment?

03:01Yeah, thanks for asking that question. What we have done is transition to more comprehensive

03:07products, what we call our compute subsystems. And think of that as a design that not only are we

03:13licensing the individual blocks of IP, but all of the other blocks of IP required to make a

03:18subsystem that would go on a chip. Those command higher royalties generally double from what we

03:25see on traditional individual blocks. The benefit to the customers is it really accelerates time to

03:32market. Chip development time goes down. Huge value that they see in terms of predictability of the

03:37design. So that's what we've been shifting to much more in terms of our royalty model, which is why

03:43the royalty growth has been as good as it has been.

03:45I'm really interested in the AI data center part, because this has been a real area of focus,

03:51of growth. And I'm interested in the momentum that you're seeing and how resilient it is.

03:55You're talking about uncertainty when it comes to smartphones and electronics in the future.

03:59Has there been any signal of uncertainty on spending on AI infrastructure?

04:05Yeah, not that we've seen, Caroline. You've got Stargate, obviously a huge program that's being done

04:11between SoftBank and OpenAI, which is all around capital and power for data centers. You're seeing

04:16large CapEx spending by the major hyperscalers. But probably more importantly, I think we're nowhere

04:23close to really good enough in terms of the capability of what AI can do. Obviously, the advances have been

04:30amazing. And if you look at just what OpenAI has done with ChatGPT, it's been remarkable. But enterprise usage

04:36from a mass deployment standpoint is still quite modest. A factoid I like to refer to is ChatGPT used

04:44a petabyte of data to create the model. JP Morgan has 150 petabytes of data sitting inside their

04:50enterprise in one way, shape, or form. Harnessing that data, turning that data into real valuable

04:55information, and then helping productivity, there's a lot of white space there. And I think still there's

05:01a long way to see growth here. There's been some uncertainty, of course, around ultimately how

05:06those that use your designs can ultimately sell their chips. And we've now got a pullback from the

05:12AI diffusion rule, as it's called here in the United States. It would seem that's going to be

05:15announced anytime soon by the Trump administration. Is that going to be a boost for your business? Do you

05:20think that leads to more certainty or actually lack of it? Because now we have countries having to

05:24negotiate on a case-by-case basis. Yeah, I don't know that I could predict it. I haven't actually seen

05:29what the outcome looks like. The wonderful thing about our jobs these days, CEOs, is that we're

05:36reading the news and hearing that from you at real time. I think we'll just have to see is the short

05:41answer. I don't think AI demand is a bubble in any way, shape, or form. And as I said, I don't think

05:48we've gone anywhere close to reaching its true potential. When you think about end devices that

05:55will run AI models, everything will run them. They'll run in earbuds, they'll run in cars,

06:01they'll run on mobile phones. They're certainly going to run in the cloud, but they won't run

06:05exclusively in the cloud. And I think we still have a long way to go in this area. The progress has

06:15been remarkable, shocking in some ways, just how good the models have become. But there's still a

06:20long ways to go in terms of, I think, true AGI and then ultimately ASI.

06:25Rene, your top five customers are responsible for 60% of revenue, and are in China, 20% of that.

06:31I'm curious as to how you're making an effort to diversify. That's a phenomenally concentrated

06:36client base, particularly when you consider that China, you know, you might have some IP

06:40concerns around China.

06:42Yeah, the industry that we're in has been seeing, you know, quite a bit of consolidation,

06:47you know, for quite some time. The amount of capital required to make chips is quite a bit.

06:55So you look at the number of companies who are, their end products are chips. It's shrank over

07:00the last number of years. And at the same time, you've got large hyperscalers, the Microsofts,

07:05the Metas, the Googles, the Amazons, that, you know, there's not many people have that capital.

07:11That being said, the number of chips, the demand for chips, the compute has actually gone way up.

07:15So I probably don't think so much about the number of people who buy. I would get more worried if the

07:21number of, if the quantity was going down. And we're not seeing that. And why are we not seeing

07:26that? The insatiable demand for more and more compute driven by AI, I think, again, is, we're in

07:34very early stages of it. You also have to remember that running these AI workloads, you still need to run

07:40them on top of running a Windows operating system, on top of running an iOS operating

07:44system, on top of running an IVI instrumentation inside a car or autonomous. So it's not that AI

07:51compute replaces traditional compute. It's on top of, which is a strong demand signal for

07:57semiconductors.

Category

Transcript

Be the first to comment

Recommended