Lattice Semiconductor Corporation (LSCC) Earnings Call Transcript & Summary

Great. Good afternoon, everybody. My name is John Vinh, I cover semis here at KeyBanc Capital Markets. We're pleased to have Lattice Semi here with us. We have Lorenzo Flores, CFO. Welcome, Lorenzo. Great to see you.

Thanks, John. Great to have you. Always a great event.

So let's start talking about kind of your data center growth, right? In the quarter you just reported, coms and compute grew 80% -- over 80%, which would obviously suggest that AI data center is growing faster than that. Can you talk about how much of this growth has been driven by general purpose server versus AI?

Yes. So I'm going to start with my best understanding, but I think I'll provide a little bit more color around the overall growth dynamic, which might actually be more helpful, right? So at the beginning of the year, I think everybody in the industry was looking at industry expectations from about 15 million units for total servers, something about a 20% AI mix, and then it grew up 16%. And now like the last past 3 weeks, we've seen the estimate for overall servers about 20 million units, and with a more -- higher AI mix to them. And I would say that in parallel, as we've gone through the year, we've continued to see our demand going up and up and up. And at first, we were a little bit surprised and tried to -- went back and tried to understand it. And what's really going on is, sure, AI servers are growing, but also just the more standard foundational, CPU-based server infrastructure that is enabling this agentic workload being broadly deployed. And at the highest level, both phenomenon, growth in the AI server and the more general purpose service is very beneficial to us. We've historically been really strong and had a pervasive footprint and standard servers. And as you know, and we've talked about many times, we have a really great opportunity in AI servers as you build out the racks with many, many more attach points for us, up to 100 attach points or even greater than that, in an AI service rack because of the complexity. So when we looked at what's going on and honestly, in real time, it's a challenged to parse those different drivers because our ship timing and the end customer ship timing don't necessarily align. But what we saw is a really strong unit volume growth in our products going to servers. That is accompanied by -- or that's driven by our attach rate is growing. As we've talked about, it was, historically, let's say, 5, 4 years ago, 1.x, went 2.x, then 3.x and it's now even higher. And we've been providing higher-value FPGAs into those opportunities. We've gone from our MachXO 2, XO 3, XO 3D, XO 5, and some Nexus parts as we are now seeing an expansion in our ASP from 3 to 4 and so on. And so you have a multiplier effect for us, greater unit volume growth driven by higher attach rates as the complexity of the servers in both AI and general purpose applications grow and increasing ASP. So that's really setting -- we've seen great growth. It's setting the foundation for future growth. So I know I didn't only answer your question. I thought you'd get there anyway.

No, that's super helpful. Do you have a sense of what your mix of general purpose versus AI is right now?

So what we've been looking at is our overall AI penetration, and some are on the data path somewhere in the server domain. We don't do very much direct AI compute. So that's grown into the high 20% of our overall business, which means a greater percent of the compute business. But again, we know where our design wins are. We know where customers are shipping to, how they're actually using the devices, and not yet fully transparent to us, but it's, in a way, it's kind of like we'd like to know. More importantly, we want to ship all the units to them and drive the growth.

Great. Maybe just to double-click on just your attach rates, right? I think when you're at 1% to 2%, right, a lot of your core functionality, my understanding, was driven by Root of Trust and maybe I/O expansion as well. So as now you've gone to 3x and now you're saying it's even higher than 3x, what are some of the functions that are driving the attach rates to high levels right now beyond just Root of Trust?

Right. So I think actually, back in -- when it was a lower attach rate, it was primarily bridging an I/O expansion type of applications. And what we see then, we saw the security capabilities, including Root of Trust, layer on. And then now we are beginning to see an expansion into manageability applications like power and cooling. And then the basis for our acquisition of AMI builds on that growing functionality that we've seen in the FPGA business and the broader applications that's available to us, and their capabilities, which really augment us. So there -- we used to say we're first on, last off on a machine, and they're first on, last off right with us. And then they layered on top of their boot capability, this management capability, and that meshes really, really well with how we see our FPGA application set growing. So that's how the attach rate grows and then that's the growth path that AMI had intrinsically. And then for both of us together, that's this 1 plus 1 equals 3 phenomenon we're trying to drive with bringing more sophisticated solutions to the customers. And right now, we think it's very acutely needed in the [indiscernible].

Great. I do want to touch on AMI in a little bit, but just a couple more follow-ups on data center. Your levers on data center is obviously market TAM, ASPs and attach. So if you think about your data center business, I don't know, 3 to 5 years out, what's going to be a bigger driver of growth? Is it going to be ASPs or attach rates, do you think?

Right. So I'll answer that in 3 parts. One is FPGA part, one is my part, and I know you said you're going to get to that, then I think the solutions together. So for the FPGA part, if we were thinking only about silicon devices, unit volume growth will be significant and will continue to be the primary driver. And we will be able to have some expansion. What's going to be driving that is the number of attach points in the server for different functionalities. And we'd already been working on things, and we've talked in the past before, and I'm sure some of you have heard this, like things like leak detection, which 18 months ago nobody thought about, 12 months ago, people were just kind of wondering about it, now is a really critical application. So an opportunity didn't exist before. We're really well positioned to drive that. The same thing with some of the more advanced management capabilities that [indiscernible]. So we would see continued attach rate growth on top of any industry unit volume growth, right? And our ASPs will continue to progress. But I think the unit volume will be the bigger driver in the FPGA business. in the AMI business, their footprint will also expand, but also their capability -- their management capability is growing and they have the opportunity to, again, expand the value. So that might be more of a driver than footprint. But there's opportunities they have as they integrate open-source capabilities with their inherent capabilities to also grow unit volume. So there's that. The third piece of it is, it's very important because it's core to the strategic value of the AMI acquisition, which is the solutions that we'll integrate together. And that is a unit volume play, but also a really high-value application we can bring to our end customers. So you'd see much stronger revenue growth per unit from the solutions that we provide than from an FPGA or a manageability [indiscernible] by itself.

Great.

Is that greenfield or is that just a combination of the first [indiscernible]?

Totally greenfield. Well, that -- there's a nuance in the answer to that question, which is the problem statement for the hyperscalers is the same. How they would solve it by using, say, an FPGA and some of AMI's firmware, could be the same. But integrating it and adding more features and functionality, that's greenfield. So it's an upscale in value that we think we can capture. So yes. So the solution may look like the same, but I think the overall value will be much greater than the individual components.

I've got one more question for you on data center is when you think about rack scale, I think the teardowns have shown that you've got an outsized number of FPGAs in a rack scale rack. I guess the question is, how are you thinking about this going forward? Is this content sustainable? I know at Xilinx, I know this is a very different company, there was this phenomenon base stations where there was kind of an ASIC reversion. I know that you're more insulated from that for a number of reasons. But do you think it's sustainable? And then is there an opportunity for you to grow that content with newer generations of racks?

Right. And I agree with your premise, it is different than -- for those of you who haven't followed FPGAs forever, the history has been FPGAs get to a certain volume and the customer says, "Okay, I'm going to design you out and put an ASIC in place." But I'll tell you, even in my prior company, where that did not happen was where the pace of change was such that you really couldn't afford to do that. In our situation, given that we've taken -- deliberately taken this companionship strategy to the market where we are not going head-to-head for core processing capabilities in the data center, but we are surrounding that with capability that makes real economic sense to offload, right? You don't want to spend 2-nanometer silicon, your silicon die budget on I/O expansion. It makes way more sense to do it in a 65-nanometer FPGA. And so by doing that, we, in a way, insulated ourselves, just as you said, from that phenomenon. And because you've also seen an acceleration in the design cycles for these racks, from 2 years to 18 months to 12 months, the need for flexibility is an inherent advantage for maintaining the FPGA footprint. Then you layer on top of it just the performance characteristics of our particular product, which is very low power, right? If you add hundreds of devices to a rack and they're watts in consumption per, that's a problem. If you have milliwatts, right, it's actually pretty helpful to the overall solution of the power budget for the rack. So that helps us as well. And then the last piece of the sustainability does again come back to being able to integrate functionality that we can bring from AMI into the overall solution and put it in the program. And then you do the updates and upgrades taking advantage of the capability of the programmable FPGA, and you're in really pretty good shape to provide a flexible solution for the customer for longer periods of time. You always bring me out to altitude.

Yes. Yes, we like altitude here. It's a thing. So maybe pivoting to AMI. I think you talked about that you've already engaged with 100 different types of customers, hyperscalers OEMs, ODMs, other ecosystem partners. And it seems like feedback is pretty positive on the acquisition. Maybe what customer problems are generating the most amount of interest in the combined Lattice-AMI solution from these customers?

And it's the -- it will be a very similar problems that the hyperscalers and data center operators would be complaining about to everyone, which is how do I increase my uptime, that will refer back to inherent capabilities of the manageability that we have where you can look at what's going on in your system and take action to do it. Security, a huge issue that we have to continuously advance our capabilities on and provide customers with solutions that our hardware and firmware base, because threats keep expanding. And then there is power, cooling that you need to optimize how your rack consumes power. And can you keep it at the right temperature for optimal performance? And if something happens, how do you prevent catastrophic failure? You talk about leak detection often. If you can respond instantaneously and shut off the source of the leak and manage a -- have a managed power down, your system is in much better shape. And then back to the uptime. If you can bring up your rack all simultaneously versus a serial blade-by-blade boot, you're up and running much faster. So these are all capabilities we've been talking to the customers about in different forums. And then finally, in the end, because of the demand as such, the hyperscalers are looking at trying to get their systems built and they're going from they're going to a mode wanting to be able to use a heterogeneous source of supply, so different OEMs, different componentry, mixing old infrastructure with new infrastructure. And how do you manage that all together effectively? You need the manageability at the firmware layer to do that. And that's the very strong value proposition from AMI.