QUALCOMM Incorporated (QCOM) Earnings Call Transcript

So post-lunch session. I'm Samik Chatterjee again. Thank you for being here. I have the pleasure of hosting Akash Palkhiwala, CFO of Qualcomm.

I know you guys will have a ton of questions. So I'll just start it off with a couple here for Akash and then I'll let the audience take over and sort of dive into all the smartphone questions for you.

I love those.

Let's start with AI. A big driver of discussion at the sessions we are hosting today. You've talked about Generative AI in some of your investor outreach as well. Can you outline your thoughts on why it will be important for edge devices to have AI capabilities? What drives the confidence to have AI like just have that confidence about content associated with AI on the edge devices.

Yes. So first of all, thank you, everyone, for attending, and happy to be here. So our view on AI is very simple. It's very much like processing, right? You have processing that happens on the cloud, processing that happens on the device. And we end up -- we see AI kind of happening the same way. We are going to start with the cloud. Eventually, a lot of the AI processing will move over to the device for several use cases. The advantages of doing it on the device are very straightforward. Cost, of course, is a massive advantage. It's -- in some ways, it's sunk cost. You bought the device. It's sitting there in your pocket. It could be processing at the same time when it's sitting there. So that's the first one. Second is latency. You don't have to go back to the cloud. Privacy and security, there's data that's user-specific that doesn't need to go to the cloud when you're running it on the device. But beyond all of these, we see a different set of use cases playing out on the device. So if you think about pervasive AI, whether it's a copilot for an office application, or whether it's a copilot or AI being used on top of content that is being created by users, all those use cases requires AI to be on the device. And so we just think there's a whole set of use cases that demand the need for AI to be on the device. We have really strong technology. We think that enables that. Of course, we have the CPU. We have the GPU and both of those cores can do Generative AI, but more importantly, what's going to be critical in the longer term in our minds is to have a very low power neural processing engine that can do AI on the device. And if you're going to do pervasive AI that happens several times every minute. It needs to be extremely low power and the advantage of having our neural processor is, it can enable use cases like that. So that's kind of our general view on AI. You're still going to do a lot of training in the cloud in my mind, but then the trained models need to be inferred, and the inference running on the device makes a lot of sense. Eventually, in the long term, you could see models being trained with user specific data on the device as well. But that's probably a second step, we'll start with [ inference ] in the first step.

Okay. And how broad do you see the device ecosystem being creative to like just smartphones, how much of a broad reach do you see in terms of these capabilities moving beyond smartphones?

So a lot of these use cases and almost like categorize it into consumer, enterprise, and industrial. And the use cases are going to be pervasive on all devices. So if there are consumer use cases that apply to handsets, there are other consumer use cases that are going to apply to PC. We know that PC is going in the direction of becoming an AI PC device already. Same is going to happen to handset. You're going to industrial applications. There is tremendous opportunity to use Generative AI for new use cases, new capabilities that you could not do before, now you'll be able to do. And then finally, enterprise, obviously, Microsoft has a very clear plan that they've outlined on enterprise applications and having a copilot within each application to enable use cases. So this is not something that remains in one device. It is not something that remains in one tier of device. It goes across tiers, it goes across devices and it becomes something as ubiquitous as just normal processing or computing. Just like every device needs a GPU, it needs a CPU. It is going to have use cases that require a very low-power AI engine, which, in our case, is the NPU.

Your announcement with Meta to optimize execution of Llama 2 language models directly on the device. What's the broader vision there? What are the use cases that you're thinking of when you're sort of looking at that implementation?

So Llama obviously has tremendously broad application. As it became one of the open -- key open source models, there's a lot of ecosystem that uses the Llama model to run applications and use cases on top. So for us, it's relatively straightforward. We're going to take our advantage that I just outlined, bring Llama for on-device AI. So this is never goes to the cloud. The entire model runs on the device. And then that automatically brings with it all the ecosystem of applications and use cases that are going to be created. I won't sit here and profess that here's the specific use case that we think is going to drive the adoption. It's not going to be 1. It's not going to be 10, it's going to be millions of use cases. I think that will be enabled in broader ecosystem, and they'll all run on the device in one form or the other.

Okay. Interesting. So any thoughts -- early thoughts here of what your TAM for AI-related content would be associated with end markets like if you want to quantify smartphone versus non-smartphone? How are you sort of even thinking about the size of those opportunities?

Yes. So if you -- I think coming up with a specific number, it's too early in the cycle, but here's the framework of how we think about it. First is we've seen a tremendous increase in phones, smartphone content, especially for us, especially in premium tier devices over the last 3 years. And so this, to me, kind of drives the content growth going forward in addition to other things, this will become a big driver of how content will continue to grow in smartphones. And so we used to get this -- a lot of questions about, we've seen great content growth. What is going to drive the growth going forward, this is going to be one key driver for us. So that's the first one within phones. Second is, I think from a competitive perspective, we have technology that's unique. And so this allows us to continue to maintain our differentiation and our advantage as use cases get deployed. In new markets like PC and XR and industrial, where we don't have a very large presence today, I see this as almost like one more advantage that we'll bring with us when we go into those markets. And so probability of success hopefully improves because now we have something else that's unique that the market is demanding. And then finally, I think you could also make an argument that once as these use cases get deployed, it will drive replacement rates in various edge devices, assuming there's broader adoption. And so we think of that as an upside to not just our handset market, but also the other markets we're going to participate in.

Okay. So another way to look at it, I mean, at your last Analyst Day, you had outlined industry CAGR for your QCT handset to be double digits. And at that time, you had mentioned you were assuming flat -- smartphone market to be flat. Obviously, we all know that hasn't remained the case. But when you think about content growth, you were sort of embedding at point, is this sort of part of that content growth that was already sort of envisioned at that point to be double digit or is this upside really to that double-digit growth? How should we think about that?

Yes. So the content growth happens in multiple ways, right? Like one is just look at a specific tier and then generation over generation. And if you look at the last 3 years, and we talked about this last year, is the die size if that's a metric for content growth has grown at a 10-ish percent CAGR for the last 3 years. This clearly kind of positions us to stay on a curve like that going forward. But it's not just about what happens within a tier. When these new use cases come in, we see a shift across tiers, which improves the mix. And then when you look at a weighted average content growth, it also helps from that perspective. So we have a very broad view of this. We don't want to enable this just in the premium tier devices. We're going to enable it across all tiers of devices. And as people upgrade the next set of phones, we think there's an opportunity for upgrades or people buying higher tier phones as they do that.

Okay. And then taking that forward, when you sort of see the differentiation Qualcomm has to bring to this market it also raises the question that clearly, the smartphone OEMs would also look at this as an area that they can differentiate on themselves. Going back to a lot of the discussion sort of going around about in-house capabilities at these smartphone OEMs. How do you think about their intent to then invest in-house capabilities relative to the processor and just sort of look to even sort of create that differentiation on their own rather than wait for or rely on Qualcomm.

So if you think about how this technology works, it's integrated into our SoC. It relies on CPU, GPU and NPU and depending on which model and which use case you could use any of those engines. And so when we think about the competitive landscape, we've done this for a very long period of time. Number of competitors, if anything, has gone down rather than gone up. . We've also competed with several internal efforts at the OEMs. And what has held true through the entire process is if you have technology leadership, if you are delivering a chip that is leading node, all technologies integrated, you have a better CPU. We'll have a custom CPU coming in shortly. A better NPU, wireless leadership is intact. When you put all those together, we think we're extremely competitive, and we continue to have the advantages we've had in the past. So we like our position going forward.

Okay. Let me open it up to the audience and see if anyone has a question. Any questions? Go ahead.

Maybe just stepping back to the year ago when we started seeing some of the challenges in the supply chain. At that time, you had quantified sort of the inventory impact of about $3 billion to $4 billion, maybe $3 billion of that smartphone, the rest in others. If you look at where we are, is this possible to figure out how much we're undershipping the industry by specifically in smartphone? Because you haven't seen a recovery off of that units are down. I understand that you look at fiscal '23, your comments...

So the one framework that I think is relevant is if you look at the 2 years combined, fiscal '22 and fiscal '23, there was a build in a year, there is a drain in the next year, looking at the 2 years combined would be like one framework. The adjustment that you'd have to make to that is the market is lower than what the weighted average would be. So adjusted for market size, I think that's a reasonable way to think about what -- how much we're undershipping and what comes through soon, and you can debate what the timing is, but it almost in some ways when you're looking at the company long term, it almost doesn't matter.

But just going back to the $3 billion to $4 billion sort of number, I understand we have just more, Huawei had a smaller market. $3 billion of that is smartphone. Is it fair to say market is 10% lower, Huawei is a $1 billion. So that $3 billion is more like $1.5 billion impact this year, it's not really growing, so that could be the level of your undershipping the market.

Well, rather than getting to the specific math, I think the framework that I outlined would be the right way of thinking about it. So if you do the math of that framework, I think it will probably put you in the right spot.

Akash, just to sort of carry on with the smartphone market questions. I think one of the perceptions that investors have right now of the modem or the SoC market is that it's as good as it can be with Apple continue to use Qualcomm right now. Samsung is clearly sort of giving you more share at this point. It seems to be the perception that this is all going to sort of move more steadily towards in-house, Apple moving in-house, Samsung sort of overall coming back with Exynos in the market? Who are we starting to in-source, how do you think about sort of that broader perception? Is that [ accurate ] that this is sort of a period of time in the cycle where things are as good as they get in the merchant SoC market?

So we've lived this market for a long period of time. And I hear this argument every 2 years. And the reality is history would tell you that our position in the market has grown stronger over a long period of time rather than become weaker. And just you go back to pre-COVID versus where we are today. We have seen a tremendous increase in scale in our business, but more importantly, content has increased in a very big way. So we feel very confident that from a technology perspective, we're in a great spot. From a content increase perspective, we're in a great spot. We have our custom CPU coming in shortly. We're going to have the AI neural processing engine that is going to be very important. And when you put all of those together, we're in a very competitive place going forward. There'll be ins and outs through the years, but just when we look at the trajectory, we like our technology portfolio in handsets and our ability to take that portfolio to other markets.

Yes. I mean does, in your view, the geopolitical situation sort of change how things have played out versus the prior cycles, particularly when it comes to some of the Chinese smartphone OEMs. I mean yes, you've sort of continued to manage to showcase your technology and sort of expertise to sort of drive that content gain higher, but does it now change because of where the geopolitics is and where customers want to go in the long run?

Well, so I think you can look back at the last year and how things have played out, and we've continued to sell our portfolio. For us, what we control and what we think puts us in a good place is the roadmap, is the technology. And as long as we are leading in that, there's always going to be a demand for it. And we're continuing to work with all of our customers, Samsung and Apple, but also the Chinese OEMs and we feel like our competitive position continues to be strong. If you look at year-over-year, what has happened to our market share within the Chinese OEMs between fiscal '22 and fiscal '23. If anything, we have strengthened our market share.

In terms of the kind of longer-term opportunities for what could be done on the device. How important is like phone operating system working in conjunction with SSC. Just curious like what frictions there or what.

Is the question specific to AI, I assume.

Yes. Specific to kind of either a personal assistant running on device or kind of bigger opportunities that you think could be really revolutionary for handsets or PCs. Like what work needs to be done from an OS tied with core processor.

Yes. So if you think about our kind of key partners as we start working in this direction. Microsoft is going to be a key partner, and we've announced a bunch of work we're doing with them. It not just applies to our PC efforts because they have a vision of pervasive Gen AI within all their productivity applications, including an assistant -- digital assistant -- AI assistant. So they're a key partner as we bring them handsets as well. Second is Google, of course, Clearly, this is a big priority for them as they deploy new models and new use cases, especially within their own apps, we'll be working with them to bring that to market. And third, we talked about Meta and Llama, which is -- has become one of the more important AI models out there. And so applications are being built on top. So we're going to bring them to the device as well. So for us, it's important that we are working with each of these ecosystems. And each of them take advantage of the hardware capabilities we are putting in our chip.

Is there any advantage to doing that, I guess, to put it a different way, like -- do you see a future where Qualcomm should be driving development of the models because having a closed ecosystem of the underlying software and the hardware being from one vendor like there's things that you could accomplish that way that you couldn't accomplish if it were a third-party model running on your hardware?

So if you go back to the different hardware that can run models, right, you're going to have CPU, which is more of a standard approach to running models on. Same thing on GPU. There are standards that will be used to run the models. NPU will be our advantage in the long term because of it being a specialized engine to run these models at very low power. And that is where the opportunity to partner closely will be is to put these models that would run pervasively on the device on our NPU optimize it and allow the performance to be better.

Got it. So maybe ask in a different way, like do you see any potential that for model providers or for people who are trying to drive the application ecosystem that it would make sense for them to try to design an NPU type competitor and like build a device where, like we're doing CPU, GPU, modem, but the NPU is like OS specific NPU.

In our minds, what would make sense is to have all those engines all available to the application ecosystem and to build it in a way that there's a single SoC that's serving all those applications. Just history [ lesson ] and phones used to have a bunch of different chips, everything gets integrated into one piece of silicon because size matters, power performance matters, consistency across applications and use cases matter. And so architecture, the integrated architecture has always won in the past. And so we think -- we continue to believe that integrated architecture is going to win. If anything on the PC side, as you look at it, there's actually more integration happening rather than the other way around.

Got it. Okay. No, that makes a lot sense. I guess the piece of like I've always tried to kind of -- you guys do, obviously, a very competitive CPU, GPU and now NPU. But the one piece that like has actually proven very hard to replicate is the modem. So as much as you guys, I feel like kind of try to shift the focus to the other pieces, there's 2 modem providers left standing, and it's been really difficult for even very talented silicon design teams to replicate there. So I guess I'm just trying to understand how you think about the durability of a leading competitive position on edge, CPU, GPU, NPU relative to what we've clearly proven to be a durable industry-leading position on the modem side.

Yes. First of all, thank you for acknowledging our leadership in modem like that. I would say it a little differently. What makes us strong are not the parts but the whole. And so that is being very good at parts -- and of course, if you say what is Qualcomm best at in the world, is the modem, no question about it. But the race in phones is about doing the whole -- that is why there have been fewer and fewer players in the industry and then it's down to a couple of merchant players versus we used to have 10 competitors at one point. Doing everything in one piece of silicon is what is difficult. And so -- the way we think about advantage is we should have the best core, and then we deliver an integrated solution. And so the -- I think breaking down the competitive advantage into one or the other core is actually maybe missing the point on what makes us unique.

[indiscernible] smartphone business's ex growth. It's not -- and -- so it's a no-growth industry. And...

Units or content.

It's dollars, let's just say dollars, units no growth or low growth, but you might think otherwise, but just not really growing. But with AI, if you're running LOMs, I don't know, 5, 10 growing [indiscernible] and the edge, let's say, someday every [indiscernible] and your NPU next-gen versions can run these models because you don't want to push it all the way back to data centers, you want to do a lot with the computer at the edge, it makes total sense. Can you at least get this would probably imply a bigger die, bigger ASP and this will be an ARPU -- ASP driver of edge inference compute, do you see that happening? When does it happen? What do you need? What kind of changes need on that NPU computing power envelope to be able to run multibillion parameter models in the edge on your phone.

Yes. Everything you said, the answer is yes. We definitely have the strong view -- and I think I mentioned this before you came in that -- going forward, we see these models running on device. We've talked about 10 billion to 20 billion parameters models running on our devices next year. And the way we are making that -- enabling that is through a much more powerful NPU and infrastructure around it. As I said earlier, it will run, models will run on the CPU, on the GPU as well, so it is across all 3. But when you want to run something pervasively, power becomes extremely, extremely important, and that's when the advantage of dedicated processing engine becomes important. And so we definitely see it as a content growth opportunity. Just when you look back and you look at our revenue per device, it has got very significantly over the last 3 years, and we continue to expect this as one of the key drivers going forward, just as you outlined.

My question was on the smartphone industry. I think everybody said it's ex growth, but markets shrunk 20% in units in the last 2 to 3 years. The last market that was supposed to be explored PC went through an upcycle down, that -- how do you think about it over the next 2 to 3 years? Is this -- is the right assumption just to think about it flattish or at some point, the upgrade rates.

Yes. So of course, I'll have a somewhat optimistic view of it. But let's go back to history because usually, there's good things to learn out of it. What has happened in the cell phone market through history is when a new use case or a new inflection point comes in, replacement rates go up. And then over time, it normalizes and then goes back up again if there's another new set of use cases. So what is important for us is, as we look forward, there was a 5G cycle we went through. During the 5G cycle for Qualcomm content actually grew more on the application processor side than it did on the modem side. As we go forward, we think to Tony's earlier question, like Gen AI and the use cases that, that could bring could drive a whole set of upgrade cycles in the phone going forward. And that's the history of the phone market and the replacement rates is -- we had 2 factors driving the market. We are the total market growing as more people were buying phones but then people who had phones replacing at a certain rate. Replacement rate has tempered down a bit, we think there's an opportunity for the trade to go up as you go forward.

Yes. I think on the last Analyst Day you think about a multiyear CAGR, where we thought it was going to be flat it's sort of done this. And if that flat number is right, it has to at some point, recover -- or should we think of this as a new reason.

I think that's one way of thinking about it. The other way is a bunch of people upgraded their phones in 2021. If you look at a 3-, 4-year replacement cycle, you should get a redo of that next year or the year after. And again, I'm not like saying something unique. It's just that how the market operates. Gen AI obviously sits on top of that as a reason to upgrade as well.