QUALCOMM Incorporated (QCOM) Earnings Call Transcript & Summary

Thank you, everybody. Hi. I am Amy Sutter, and I'm sitting in for Rob Sanders, our European hardware analyst who unfortunately, who is not able to be here today, and he covers Qualcomm. So I will be running the fireside chat. And I want to introduce Durga Malladi, who is the EVP of Tech Planning Edge Data Center. who's here from Qualcomm. And maybe just to get started, I know you said you've been at Qualcomm -- maybe give us a little bit of your background, kind of what different roles you've had?

Sure. Yes. Yes. So I joined Qualcomm in '98, so it's my 29th year over there. Inside Qualcomm, we have gone through like at least 2 different DNA mutations over the last 2.5 decades or so. So this is the third one. I ran our research and R&D organization for the longest period of time until about like 10 years back or so, -- that's when Christiano, who was -- who is our CEO now, my boss now. So he said, "I want you to come over on this side and help on the business. So for the last few years, it's been more than a few years, now over 8 or 9 years have been running the product planning. . And at this point in time, I run all of our technology road map across all the businesses, ranging from this lowest end IoT all the way up to a data center. Last year, when we restarted our data center business, I ran the business for 1 year just to kind of make sure that we're in the right direction. And now we have like a full-fledged business unit that's running by itself. So that's my role.

Awesome. Great. I think since the Analyst Day, most common questions, I think, that we've been hearing from investors is how are you going to differentiate in the data center with your newly announced high-bandwidth compute technology -- can you maybe help us understand the genesis of HBC and kind of where it fits into your total strategy?

Yes. It started off. First of all, when we unveiled our HPC product portfolio and the multiple generations that we are working on, -- it's important to understand that it's not something that we just did in 1 year or so. It was something that's been in the works for the longest period of time. But it was last year that I felt like we have reached a point in our R&D where now it's time to commercialize this. But working backwards from there, what is the problem that it is solving. That is something that became quite clear to us by 2021 time frame. When we were observing how the compute in each of these data center racks. And in fact, it also holds true for some of the other places. It's growing quite a bit exponentially, generation over generation, while the memory bandwidth was kind of stagnant or maybe growing more modestly, it's more of a linear growth. And it eventually will lead to a memory wall wherein you can throw as much compute as possible, but it is going to do absolutely nothing when it comes to influence decode portion of it, unless you solve the memory bandwidth as well. And so we invested quite heavily in our R&D. And we reached a point where last year where it's becoming like, yes, now it's feasible. We've scracked a lot of the solutions over there. Keep in mind that we called it as high-bandwidth compute because it is not really a pure memory technology. It is actually about a combination of how do you have the compute right next to memory? How do you design it, codesign it in the right way and then make sure that it interfaces with the rest of the accelerator. And it was a direct take on an alternative way of doing things as opposed to -- and we feel very confident and comfortable with where things are right now, and that's why we unveiled our multi-generation solutions with the first generation coming out next year and the second generation after that. But that is a very unique proposition. And now it's fully endorsed by a large number of the memory vendors as well who've been our partners for a while, but now they are equally out in the open and talking about something like that.

Yes. Great. Yes. I would not have thought Qualcomm as being in the memory game. But talking about it as a compute technology is very helpful. Maybe just to dig a little deeper, when you talked about Gen 1 and Gen 2 and you -- I think at the Analyst Day, you showcased the bandwidth per watt relative to HBM and then you also have some very high numbers relative to SRAM. Can you talk about the unique selling points versus HBM and SRAM that you're talking to customers about in terms of solving the issue with the memory wall.

Okay. So 2 things over there. First is some of the numbers and the second part is on proof points, and maybe I'll shed a little bit of light in terms of how the discussions are coming up as -- so in terms of numbers, we always, always talk about tokens per watt because the power consumption is something that is very important. And today, with HPM, the power consumption is very high. But it's a funny anecdote on that because we talked about 6x improvement compared to HPM in terms of tokens per second to what. And the second part is HPM today is about, give or take, 21, 22 terabytes per second. We had like hundreds of terabytes per second. And when we compare it against SAM, SAM has got its own restrictions. It is the fastest interface, but it's also like it consumes more area, you will need a larger number of racks to do the same thing. So there's a TCO argument to be made against SRAM. These are the 3 things that we talked about in -- at Investor Day. But the important point is, okay, these are numbers. What are the proof points? So the nice thing is that, yes, we tapeout the silicon, the silicon is back in the lab. It's coming along very well. stay tuned on that front as we start unveiling this a little bit more on what the numbers are beginning to look like. But we are pretty confident in terms of the first generation of the products. A bit of an anecdote on this. Like right after Investor Day, we spoke about this, we said, "Hey, here's HPM, and there were like these 2 nice animations, 1 which talked about, okay, this is no good with HPM. On the other hand, with HPC, this is great. Right after that, I was in Korea with the 2 memory vendors, and I fully anticipated them to say, what are you talking about? We have this awesome road map with HPM that's like really great. They're going to -- it's going to be like much better than HPC. That is not the feedback that we received. Instead, what we got is, hey, how can we work with you much closer on this. And in fact, we have been working with them because we need to make sure that we have the right supply as we do the commercialization. But in addition to that, there's been a lot of interest coming in directly from the memory community itself, which has been a very positive and pleasant experience. And that's actually good. It also kind of validates some of the points that we made that we need a different kind of a solution as we move forward in these data center act. The second generation of the product, so the first generation is slated to be commercial and shipping in 2027. We're already working on the second generation, which is going to be 2. And those numbers are even higher. And 1 of the reasons for that as an in terms of the relative comparisons against what we anticipate with HPM 4. And the reason for that is as we kind of look ahead to 2028, A lot of things are evolving in itself. On 1 hand, AI models themselves are evolving, which means the compute that we end up doing in the compute die or the logic die, which is underneath the DRAM stack. We actually are anticipating what we need to do, and we are throwing in even more aridited. It's harder to scale that with HPM, so the relative gain looks even better with that. That's where we are.

Great. Okay. One of the criticisms of it is like with the multiple layers of LPDDR on top of -- with high-performance logic that there's thermal issues. So maybe can we talk about how you're getting around these with I guess, your packaging technology. And then you mentioned Gen 1 and Gen 2, just maybe some milestones for us as investors to think about it's for us to look for in terms of this. And then I know you're talking about talking with the Korean memory players, but also are you engaging with hyperscalers today? I don't know if that's something you've shared yet.

Okay. Short answer is yes to all of them, but I'll start with proof points. So as I said, silicon is back in the lab, looking good. It's coming up along very nicely. I don't believe we have explicitly put down dates as to when this is going to be unveiled. But in the next quarter, I just stay tuned on that front. I think some of it is also within our marketing domain in terms of how we want to expose it. But we will have numbers. We will actually have the silicon validation coming in. And we are on track to ship both from the samples, the engineering and the commercial samples and this technology validation in itself, I think it's looking good. We don't anticipate any issues for the first generation. What's happened is that as we brought this up, and we started talking to every single hyperscaler. Every hyperscaler, they have a lot of their own in-house solutions. Most of them -- some of them are like, "I have my CPU. Some of them are like, I have my CPU and I have my AI accelerator. But very few of -- they always relied upon HPM as okay, it's 1 of those memory bandwidth things. I will acquire that IP and then I'll build my rack on top of it. It's always been like really good when we sat down with all the hyperscalers and they took a look at this, like this is awesome. How do I actually bring this in? And at Investor Day, we specifically talked about these are 3 different tracks, independent tracks. For example, some hyperscaler might come in and say, I have my CPU and my AI accelerator, I need your HPC to be a stand-alone product that works with this. We have a solution for that. There might be someone else who will say, "You know what, I have my Accelerator. That's good. I like your HPC and you know what, your CPU is looking pretty good, too. Maybe I can take both of them. How does that work with me. We have the answer for that as well. It kind of is now at a point where with hyperscalers, we're not coming in with some sort of -- it's all or nothing. No, you can take individual components, if you want, or you can take all of the above. And then we take it from there.

Got it. One of the things when I think HBC, the concern is maybe when you look at prefill versus decode, and I know memory is more focused on the decode side. That's where the bottleneck is. But can you maybe talk about how HBC does there.

Yes. So -- if you are looking at inference in data center, typically, we split that problem into there's the prefill stage and the decostate. The prefilled stage is fully compute dominated. The more compute you throw it, the better it is. And off late, we are seeing something like a 2.5x to 3x increase in compute generation over generation. So you can always solve the problem with throwing more compute at it, HPC will not do much about that 1 at all. On the other hand, decode is completely memory bank we dominated. You can throw as much compute as you want. It doesn't matter unless you have a solution like this, it's not going to work. And that perhaps is the place where in all these discussions, not just the hyperscalers with a lot of the others as well, we see this, "Hey, I have my compute solution. I can use it for prefill, perhaps for decode, as you're doing HPC, there's additional logic that goes into the bottom die. -- can bring in that logic over there. So there's like even more of a mix and match that's occurring in that space. That's how we anticipate at least in the short-term things to evolve. Now in addition to all 3 of these, CPU or AI accelerator and HPC, at Investor Day, we also talked about our custom silicon business, now if you think about everything that I've said, you're almost getting to the point of is it like a custom silicon for something that's going into a hyperscaler. I think there is a very thin academic line between these 2 arguments because once you start putting it together, it does go in that direction. But there's a heavy dosage of our IP that goes in along with mix and match with their IP as well.

Got it. Got it. When you were -- like when you -- I think Hynix and SanDisk are also talking about high-bandwidth flash, though it seems like it's a little ways out, at least that's my take. But is there -- is that -- and it could -- could hyperscalers take a pool of HBM and HBF and run the models more cheaply? Like how does your HPC defend against that kind of TCO narrative, if that's something that becomes a way that we're trying to attack the memory as well.

So when we talk to memory vendors and we kind of have gone into this in depth, with them in terms of HPC versus HBM and what's coming with the high bandwidth flash. I would -- this is my opinion over here at this point in time. HP Flash is trying to solve a slightly different problem. It's not quite exactly the same problem because 1 of the things that -- the argument over there is -- let's also use flash in addition to DRAM to actually do something else over here. That's not the same thing as increasing the memory bandwidth to the point that the decode performance becomes really good. But it has its own use case. There's still more work to be done. That's just the perspective that we have right now. But it is 1 of those stacks, which is kind of independent of this. Meanwhile, with HBM, comparison against HPM. The same vendors who have -- they've been working on HPM for a while, they are like, I need to open a parallel track in addition to this because it's not obvious to me that in the near future, at least for the foreseeable future, not just near future, there's anything like HPC that I can offer through my HPM. Now time may well, maybe 5 years down the stream, it's going to be a different story. But as of today, that's the only solution that actually gets us to this level of performance gains compared to what we have.

What I think has been very interesting for me coming from hot chips earlier this week is just the diversity of solutions that are coming out today in compute, GPUs, CPUs and memory and how every hyperscaler and customer model company is attacking it differently. When I think about HPC, can it be relevant for other end markets outside of data center, maybe auto?

It's a very good question. And we -- when we talked about HBC at Investor Day and of course, at Investor Day, we were talking a lot more in terms of our investments into beyond what we typically have done in edge devices, and that's why we talked a lot about data center. But truth be told, HBC is a generic technology. It goes all the way down into devices. And let me explain why that's the case. And we actually are working, Believe it or not, we are working on HPC on devices that are around us today. We're actually working in terms of what is the technology that makes sense for that. And here's the reason why. In data center, the comparison is against stage PM, and we are talking of what's the best way to solve the decode problem and the memory wall problem over there. There is no equivalent problem in devices. Let's take a smartphone or an agent AI device, but there's a different problem over there, especially with agent AI devices, which are like -- these are devices that are being built as we speak with those who've been in the business of thinking AI first, and I need to build a solution around it. But it's all about ambient AI. It's constantly running. That's why we call it agent. It's always running in the background and just waiting for you to say something and then it picks it up from there. But if you're always on ambient I always on. That means you're always burning power. Now HBC is computers right next to memory. So there's something else that you get. We've completely collapsed that transport latency between memory and the compute and not just that, the power consumption has come down dramatically. It's perfectly suited for ambient AI in all sorts of agentic AI devices that come in now. So the usage of HPC in devices is in the context of ambient AI, whereas in data center, it's about the memory bandwidth that it provides and the deco performance.

And when you say like edge cases? Are we thinking about like something like robotics, where we're thinking about AI there? Or could you think of it being used in a car or a phone?

Going all the way down to smartphones, tablets, PCs, XR devices and yes, into automotive as well. It comes down all the way to those.

And when do you think that happens?

Discussions are going on as we speak. Typically, we have like a lead time on this. Like if you have a commercial product in 1 year, then usually, it's like 3 years before that is when the commercial discussions begin. That's a question for -- once it shows up, you will realize it, but suffice it to say that we're not that far from the first generation of the commercial products coming in.

Got it. I guess then maybe kind of closing out on HBC a little bit. When do you -- I mean, do you think it's with Gen 2 or Gen 1 where you see customers really clamoring for it? And if like if you -- does it become a technology that you might offer to license to third parties to help them scale? Or do you look at this as like some kind of proprietary moat that Qualcomm can offer?

So there's 2 questions over there. So the first 1 in terms of going beyond like Gen 1 and as we're going into Gen 2, where exactly is the action over there. GEN-1 is next year. like that's like literally next year, there would be a few customers will go in that direction. But then as we talk to a lot of the hyperscalers and so on, -- there will be some separation between where the volume is going to be between Gen 1 and Gen 2. We will certainly see traction on Gen 1. I hope we see a lot more because more and more customers are lining up. So the earliest adopters are the ones who are going to go with Gen 1. But now we have so much of incoming interest coming in. I do anticipate more coming up the Gen2 as well. So that's the way that we see it. The second part is in terms of how we see this technology. Just keep in mind that there's actually 3 different components to it. We're not a memory vendor. So someone is still doing their DRAM stacking, they're not doing the DRAM stacking ourselves. What we prescribe is this is how things should be done. These are the TSVs that need to occur. This is the design for it. The compute of course, comes from us. We do the all the logic inside that. We still have to work with the likes of TSMC to take it all together. I think they're all in our swim lanes in our own way. But we are kind of like the overall system integrated and putting it together in the right way. So in that sense, this is our current model. We haven't said anything about how it's expected to evolve, but that's where we are now.

And so if I'm a hyperscaler customer, would I be the 1 who is focusing on the memory like in terms of like because memory is a big bottleneck today. You don't have to worry about that. The hyperscaler worries about that, and you just help get the solution integrated together in any...

So when a hyperscaler is interested in our HPC product, they will have certain volume commitment, some sort of a demand signal that we get, but they talk to us and they will cross check and make sure that, hey, you're all like from a supply perspective, that's great, but they talk to us. And we are like the front end for all of that conversation. And of course, they will be talking to the memory vendors to make sure and to the likes of TSMC to make sure everything else is an in the right order. But this is a product that comes from us to them. They are our customers. So that's how it starts.

Okay. Okay. Great. Switching over a little bit. Let's talk about the drag implies C1000 server CPU. You have some very high claims in terms of performance per watt relative to peers. But it's a second half '28, if I'm correct, right. So you're kind of giving away some stuff -- how do you -- like how do you think about what competition is doing? And does the baseline shift from the other big CPU competitors. Could it be challenged at arrival?

So last year, I remember when we decided to restart this and I was thinking about, okay, what are the differentiations that we have over there. The first one, I was like, okay, we got to go into the memory architecture and memory wall, solve the problem. Our R&D project is good to go. So that's good to go. Denis looking at, okay, where are we with our existing CPU solutions? And what's the right way to scale that. We had something in planning already at that point because over the last 3 or 4 years, we've been investing quite heavily in our custom CPU solutions our Orion family of CPUs. And I was paying very close attention to exactly where each of the data center CP data on center class CPUs are. Starting from -- you can start from Novus we do and take a look at the generation over generation. So what I was looking at is we're not competing against anyone else out there today. I want to compete against where they're going to be in 2028, and that was like the mindset with which we went in, and we need to beat it by a certain generation for it to be compelling. If it's like 2% better, then nobody is going to take it. compared to their own in-house solution or something else that you get, it has to be really compelling. That's the bar that we set for ourselves. And over time, as we kind of built this up, and we did all the benchmarking typically, this is done with Speck and so on, and we have our projections. One of the hyperscalers -- the response from them was your numbers are too good to be true. I was like, okay, so what I'm hearing is, if you do actually show up a drug, you will like them, it was like, Oh, yes, absolutely. So now, we made our claims. You're absolutely right. We feel very confident about those claims. We have absolutely no any sort of a second guessing on that at all. Now it's execution and crunch time. We announced Meta as a first customer for C1000. They believe that claims. Qualcomm is a company where if you say this is what is going to happen. Never have we actually fallen short of it. We have always been on par or better. So they went with our reputation and said we are good to go. There are others like let's see your silicon and then we'll river. So I'm really looking forward to that, but I feel extremely confident.

And so the next milestone will be launched with your first customer?

Launch with the first customer, but even before that, that's the commercial launch is in '28, but with silicon back in the lab, all the proof points and so on, hopefully, we'll be able to share a few more.

The way -- the offerings we see in the market today, some of them are more focused on more cores. Some of them are more focused on single core like hyperthreading or if I get it right. So I mean I don't -- I'm kind of curious from your perspective, when you look at CPUs in the data center, what you think is the right solution or maybe there isn't, like, again, because workloads are going to vary so much, but I would love to get your opinion on that.

Until about 2 years back, I would say that, by the way, the CPU landscape is evolving as we speak, and it's becoming like for the longest period of time, including inside Qualcomm, by the way, it's like our XPU team or the GPU and the NPU teams are like, okay, we are in the AI business. And the CPU team is not quite there. It's a little bit out there, but it's not quite like that anymore. All 3 processes are important, but in a different way. But what's happened in the last 2 years is it used to be that you had 2 different kinds of workloads in data center. General-purpose compute, that means you just run, you light up all the core if you have -- at Investor Day, we said that our racks have like 250-plus cores and so on. So Okay. So let's say that you have 256 cores or so, then you light up all of them. It's a general purpose, it's like a work horse. Then there is a different configuration, which is an AI head node, where actually most of the AI workload is being done by some XPU that are sitting out there. The CPU's job is to do some management and kind of get out of the action and let the GPU or the NPU perform whatever it is that's needed. All the AI inference occurs over there. But that led to an asymmetry in terms of number of CPUs to XPU. It was like 4:1. It used to be 8:1, became 4:1. But with agent, that's changing quite a bit. In fact, nowadays, we are more in the 2:1 configuration for every CPU, there's like 2 XPUs, maybe that's not 4. And I wouldn't be surprised if it actually becomes even less than that, like it becomes more and more symmetric. So as these conversations evolve, we initially started, hey, your course looked really great, excellent for general purpose compute. I think I got it for AI head node. I probably don't need you, but general purpose compute is great. But over time, now we are getting like -- let's actually talk about not just general-purpose compute, but also agentic AI because I would like to use those CPUs over there as well. So that's the evolving landscape with CPUs and the workloads themselves are shifting. They're becoming a mix and match of general-purpose compute and agent AI, not just AI head node, but agent workload, which look pretty different.