Lattice Semiconductor Corporation (LSCC) Earnings Call Transcript & Summary

Great. Welcome back, everybody. I'm Joe Moore, Morgan Stanley Semiconductor Research. Very happy to have today the management team of Lattice Semiconductor, Ford Tamer, Lorenzo Flores. I don't cover the company, so if I ask questions that I shouldn't ask, just answer the question I should have asked. It's fine. No harm done. But we've spent a lot of time in the FPGA space, and I like the Lattice story a lot. So Ford, I think you wanted to start with some opening kind of commentary? Or you want to share...

Sure, if you'd like me to. So first, thank you for having us, Joe. And for those of you who don't know, it's Joe's birthday. So happy birthday.

That's getting a lot of advertising, but thank you.

So excited to be here and see some familiar faces. I'll take you through -- I've been at Lattice now for 17 months. So let me take you through the past 17 months and then fast forward to today. I joined 17 months when the inventory situation at the time was 6 months of inventory in the channel. The revenue had decreased from $730 million to $500 million in 2024. We had to cut costs. We took a 14% restructuring. And at the time, we said what we were going to do. And at the time, we said we were going to grow 25% in '26 compared to 2025. We're going to drive inventory in the channel down to 3 months by end of '25. and we'll keep bringing this down. We said we're going to be growing in 2026 big time in data center AI and Nexus, which is our small range FPGA product line. And then in 2027 is a year of physical AI and Avant, our midrange. And we basically did exactly what we said. End of '25, our inventory in the channel was down to 3 months. We're back to growth and consensus is now above that 25% number for '26 growth. Margins still very strong for both gross margin, operating profit, EBITDA and free cash flow, and very excited about where the future can take us. Future is bright.

Great. Thank you. So I guess you described Lattice as the everywhere companion chip. Can you describe what that means? Can you describe the philosophy underlying that?

Yes. So the realization when I came in is, look, we don't want to be the chip. The FPGA in the past has been wanting to be the chip, the NIC, the wireless processing, the near-edge AI. We want to be partners to all the major chips that are driving the growth in both data center AI and physical AI. So the MVP, the most valuable player is the GPU or the AI accelerator or the network switch or the NIC or the baseboard management controller, the microprocessor, microcontroller or the different sensors, and we're a companion chip to all of these chips. We call the GPU, the MVP. MVP is a very important player, the big player on the field, but they cannot win a game, let alone a championship without a team. And we're that team. We're that team providing all these functions to the tunes of -- it used to be tens of FPGA per data center rack to now hundreds of FPGA per data center rack all the way from booting, power sequencing, control management, I/O expansion, security, power and cooling, et cetera, to the tunes of tens of FPGA in a humanoid providing the major control and connectivity to the various sensor, LiDAR, RADAR, and the tens of motor controller there are in these humanoids, robotaxi and various physical AI applications. So we do this. Being a companion chip doesn't mean you're -- when we first started calling ourselves companion chip, our team was worried, oh, that means we're weak. Well, we're not weak if we're a companion chip. We're actually -- we provide a very powerful function, because we are Switzerland. We provide it across all the various vendors. We provide it across all these functions that I just took you through. We're everywhere. We provide this companion chip function anywhere from the comms and compute market that are growing the fastest. This year in 2026, it will be over 60% of our revenue. In industrial, automotive and all these new physical AI emerging spaces, humanoid, robotaxi, medical, aerospace, defense, autonomous vehicle and even AR/VR in the consumer side. And we do it across everywhere as well in every application. So if you look at application, we're not selling FPGAs anymore. We migrated the company to be a solution provider. So we sell you security, and it innovates the discussion to a different level. So we go in there and we talk about security. We talk about rack management. We talk about power and cooling. We talk about PQC, post-quantum cryptography, which is protecting the assets against quantum decrypting in the future. We do it near every image sensor, every LiDAR, every RADAR, every ultraviolet, every infrared image sensing. We do it across a variety of other industrial sensors like temperature, pressure, et cetera. We've started recently to have these partnerships that are driving us into these high-volume applications. So a partnership we discussed is a partnership we have with NVIDIA, where we have a solution that we support NVIDIA with called Holoscan, where we can feed, let's say, 15 different camera feeds into our FPGA, some preprocessing source in there that preprocess data before feeding it to Orin and Thor, making them more efficient. We've announced a partnership with NXP, where we had a joint meeting of our customers in Europe in January and the VP Strategy from NXP flew from Austin to Munich just to be with us, be the keynote speaker before flying back. And we're putting a partnership where you can have big FPGA, small micro; big micro, small FPGA; medium, medium, it doesn't matter; providing the best solution to the customer. We're announcing many partnerships with image sensor and LiDAR, RADAR, where we make these solutions easier to adopt and easier to install for customers. So that's the companion chip story across every function, across every market, across every application, everywhere to benefit the customer and eventually the investors.

That's great. NVIDIA and NXP probably doing pretty good these days. I hear about this NVIDIA company. The attach rates for FPGAs per server, I think you are expected to move above $3 with ASPs above $4. Can you talk about the longer-term trajectory there? And what ultimately caps that number?

Yes. So what Joe is referring to is the fact that our revenue is growing faster than CapEx. So last year and this year, CapEx in cloud, CapEx for data center is growing 50% to 60%. Last year, we grew our server business at 85%. We grew our communication business at 60%. So growing revenue faster than CapEx. And we're going to continue to do so this year. And what's driving this is the following: Number one, the number of servers is growing pretty fast. Number two, AI as a percent of revenue is growing fast for us. So we went from, in 2024, AI being 10% of unit, to last year being 12% of unit, and this year 15% of unit being AI servers. And just to put things in perspective, AI drives a larger percent of revenue. So last year on 10% of unit for revenue, we drove 20% of the revenue dollar. So hence, providing higher dollar per unit. Number three, attach rate, we keep finding the new applications. So security has been a very strong application for us, where we started with Root of Trust. We added attestation. We're adding post-quantum cryptography. We keep growing in security. We found this new application on cooling, for example, that we recently adopted and that's, again, growing very fast. So the attach rates across applications are growing. The needs are continuing to grow, hence, bigger FPGA and hence, bigger ASP. So overall, you could see, if you put all these together, the multiplicative and revenue grows faster than CapEx, which is already growing at a pretty fast rate. So that's what's driving us.

Okay. Yes, I guess how do you think about that longer-term trajectory then with hyperscaler CapEx? This year, we're sort of looking at the hyperscalers cash neutral. I'm pretty bullish spending will still be up a lot next year, but how do you think about that dynamic affecting you guys?

Yes. Look, we're booked for the year and through mid next year. So I mean, bookings are extremely strong, and we see them to be strong for the foreseeable future. So we don't see a slowdown right now. We're up and to the right. And this year has been the year of data center AI for us. Next year is going to be the year of physical AI, where you've got a whole bunch of these new applications I just described that go to market. And it's interesting, some of the discussion we've had at this conference, Joe, is can humanoid be bigger than data center, because people are saying, look, we've got as much content in a humanoid as we have in a rack. And so 15 million server, 3 million racks, the numbers for humanoid could be anywhere above that. I was listening to the Figure AI CEO on the Moonshot podcast. He's predicting more than 1 humanoid per person. So that would be 8 billion humanoid. So whatever the number, even if we're wrong by 2 orders of magnitude, 80 million is still a pretty big number. So all of the new physical AI applications coming to market next year could be super interesting. And we're seeing real productivity gains. I mean, us at Lattice are using Claude for our own development. We're seeing tremendous productivity gains. We have all the tools, Cursor or Claude. But Claude over the past few months has been amazing. And so you're seeing this accelerating. These new applications are coming to market and really the rate of change accelerating in a pretty big way.

I think the irony of all of this is that your 2 biggest competitors were bought by compute companies with the idea of moving FPGAs more into the compute space and that you're having all the success. I mean, is that in a way because of that, because they're focused on that central processor and trying to use FPGAs for that?

Look, I mean, our success is directly related to where the sweet spot for FPGA is. The sweet spot is in the low to mid-range FPGA non-SoC. That's exactly what we do. We're focused on this single mission. And we're going to take the company in the solution space. So you're seeing us invest in solutions. We've got people that understand these solutions across all these markets, and solutions where we're going to take the company and grow it in a big way.

And when you talk about solutions rather than FPGAs, I mean, it's still, at its core, a programmable FPGA, right? That you're orienting around those solutions.

Absolutely. At the core, we're still providing future-proof. So let me give you why FPGA. So security, the algorithms are changing very fast. You want to be able to change -- like the old algorithm like AES-256, when quantum shows up, it is sticking like a gum to paper. It's like no protection whatsoever. So you've got a huge imperative to move to this post-quantum cryptography like today. And these algorithms for post-quantum cryptography, PQC, are changing still very fast. So we've got these Lattice algorithms, and no pun intended, these are the name of the algorithms. These are 3D algorithms, but they're changing. So having the ability to change these algorithms in the field for future proof is very important. When we talk about humanoid application and physical AI, FPGA is great, because you need low latency. So if the roaming humanoid needs to stop, it can stop very fast. You need to have parallelism for fast performance. You need to have accuracy for applications such as screw driving, bin picking, welding, you need high accuracy. You need connectivity to be -- we have EtherCAT that can connect all these motors and vision sensors together for perfect synchronization. You need futureproofing again. So the FPGA provides all these functions. For all these different various applications, FPGAs are critical, but you also want a low-power FPGA that's going to be small size and cost-effective and fast boot. And by the way, you want to go talk solution. You don't want to talk FPGA. You want to talk security. You want to talk vision. You want to talk motor control.

Okay. Very helpful. Lead times are stretching out. Can you talk about what you're seeing there and how that's impacting your business? And any opportunities around that to see higher pricing, anything like that?

Yes. So the lead times are indeed stretching out for us and for everybody else, obviously, reflecting the strength of demand. And what we've been doing with our customers is working with them directly and with our channel partners to say, when do you really need product, right? So what we've done with the extended lead times is try to work with the customers to get the bookings through the year. And as Ford said, we're seeing them in through the middle of next year, past in some cases, so that we can stage our supply chain and meet the customers' end demand. So the way it's helping our business is in planning, so we can actually deliver the parts. The visibility that we now get from a tighter channel, right? We dropped the supply in the channel down to where we have much more direct visibility and are able to target supply to end customers that really need it. That's really helping our customer relationships. And then on the pricing side, we have a different approach than some other people in the industry. We have taken a very long-term view of our strategic customer relationships. We intend to be with our customers for a long time. And what we're working with them on is we have had a consistent mature product price increase type of program. And we're keeping a balance between potential cost upside in the supply chain with how we're approaching pricing. And so we're comfortable that lands us in the gross margin we've guided through the year. So we're definitely not doing a short-term pricing move to optimize margin in any quarter. We're looking all the way through this demand cycle and then through the product ramps that Ford described in the opportunities we see in front of us.

Can you talk about other puts and takes around gross margin? It seems like product mix should be favorable for you. You said the pricing isn't changing, but it seems like other factors may.

So it's a very dynamic situation. So I didn't say pricing wasn't changing. I just said we're taking a balanced approach and long-term approach to it. Obviously, we're on the lookout and constantly in communication with our supply chain on where we -- so that we are able to sense where price increases from them will come from. So we're trying to manage our cost. We took steps starting about 6 months ago actually on both the cost side and on the supply availability side, because our Head of Operations saw some of this coming. And so we're trying to get ahead of that. But we still have to be realistic that it's a factor in our industry, right, the supply chain constraints and possible cost increases. So we started to manage that. We do have the new products. We do intrinsically have a high-value product, right? And our customers do want to work with us and use our product because of the low power, high-performance characteristics of it and the suitability in all the applications for its data. So we are trying to manage that side of the equation on the pricing side to keep it in balance. But we do have great product, and we do have strong demand, but we're all very aware of the possible supply chain issues.

Great. And on that new product revenue, I think you're talking about a mid-20% of total revenue this year. Can you talk about the progress of that and then the Nexus versus Avant timing?

So we think that 20% will get to the mid-20% this year. So good progress and that growth could continue to grow. Last year, we grew 70% year-on-year. This year, we think we grow 60% or more year-on-year on the new product. Nexus is really 2026 with Avant really taking off in '27.

Okay.

There's a question there.

Yes. I'm sorry. I'm just sort of... If you were to explain somebody not very bright like myself, how does an FPGA fit into the broader hardware world? And I know you keep making the point that there's a broader role for it in this new AI, GPU, et cetera, universe. But how does it slot in? And why is the opportunity so much larger relative to where we were in the past?

Yes. No, it's a very good question. So at a high level, if I were to draw a continuum, right, from a -- the most flexibility is a processor, right? So you'd have microprocessor, microcontroller can give you the most flexibility programming in higher-level languages, C, TensorFlow, PyTorch, all the way to the other end of the scale in ASIC that is going to be fit for that purpose. So it's going to be a GPU, a switch, a NIC that is going to require hundreds of millions of dollars to produce, but the best power, the best performance for that application, right? So that's the continuum, right? We're in the middle. And think of an FPGA being hardware-like performance, right? But the issue we have is you have to program it at hardware-like level. So we don't have the same appeal today just because the microprocessor, microcontroller, you can program and see, you've got many more people that can program in this micro versus FPGA. But if you look at these applications such as the application of data center we're providing to the humanoids, I talked about the robotaxi, aerospace and defense, medical, all the applications require hardware-like performance. So they require the performance that we can provide. And you don't want to go spend hundreds of millions of dollars to develop an ASIC, which if you've got an application like that's really high volume, such as an AI accelerator, switch, NIC, board management controller, may as well go spend tens of millions, hundreds of million dollars to do an ASIC. We can provide you not quite the same performance as an ASIC, but quite a good performance, much better than a microcontroller. We provide you performance. We provide you latency. We provide you deterministic, we provide you accuracy. We provide you much lower power. We provide you many more benefits than you'd have in a microprocessor, right? So we're in this in between. Why is today FPGA becoming more important, okay? So the cost of developing these ASIC continues to grow. The time it takes and the size of the team it takes to develop this ASIC continues to grow. And you've got applications such as the one I just described that require that hardware performance, that require this unprecedented, you want to have this -- you need this performance, you need this power. And yet you don't have tens of millions, hundreds of millions. FPGA is a great solution, right? So as the cost of ASIC continues to grow, the cost of putting these functions on a 2-nanometer, 3-nanometer advanced node continues to grow. So I look at this AI chip and this big chip that takes over the whole reticle as much -- I've got as much logic I could put on this chip. And would it make sense to offload this power sequencing, this control, this boot to it? We're in 65-nanometer, 28-nanometer, we're in these back processes that make this function much more cost effective. So it makes more sense for me to offload these functions to this FPGA. There's more and more pressure on system time to market. So NVIDIA, AMD are driving to 1 year between different generation of GPUs, okay? Elon Musk comes in and says, I want to go 9 months between my xAI type of GPUs, right? So there's more and more pressure on going to market fast. And so a lot of these functions that typically you'd have the time to go put in an ASIC, you don't have the time anymore. You're going to have to say, okay, FPGA is good. I mean FPGA gives me the performance, gives me the cost, gives me time to market. And by the way, it gives me flexibility. So the security algorithm, as I said, keeps changing. What FPGA allows you to do, FPGA allows you to program in the field in the future. So today, you program for a certain algorithm for security. The bad actors are going to change those algorithms on you. They're going to come back tomorrow and say, okay, hey, I just changed this encryption. Now I can't know how to decrypt it. Okay, well, guess what, I'm going to be a step ahead of you. I can go change this in the field in my FPGA. So many, many factors. The ease of use AI will allow us to close the gap. So look at Cursor, look at Claude being able to program at a higher level. AI will allow us to close the gap and make FPGA broader and more available and more interesting. So we're seeing adoption across the board. But yet, you want adoption with low power, low cost, fast boot time, small area, and that's what we provide. So all of a sudden, we're seeing this thing mushroom. I mean people in the meetings today saying, "Hey, we've never seen FPGA grow at these rates. So this is all the factors that are helping us grow the company at this rate. Makes sense?

Yeah. We've another question there.

Could you elaborate on the humanoid opportunity in terms of how you provide value to that new market? And what you see as the time frame over which it will evolve into a very big market?

Yes, yes. So look, first, we start with vision. So humanoid obviously has to see where it's going. It needs vision, and you've got many cameras. Sometimes you've got a LiDAR or RADAR to guide that humanoid, right? And near vision, you need to have an image processor that allows you to do this image processing, to do the control, to link up all of these different vision pieces together, and we provide that in FPGA. Next, you've got motors. So there are many podcasts that talk about each of the hands having upward of 20 motor per hand. You've got motors all over the torso and the shoulder and the legs. So it could be anywhere from 45 to 70 motor controllers, and we could provide the motor controlling support control. You think about the humanoid having 70 motors. Well, the 70 motors better be synchronized and work together at the same rate. We provide connectivity to provide this EtherCAT that allows you to synchronize all these motors together. So a lot of value we provide and security, by the way.

Time frame. He asked about the time frame.

Time frame. Look, I mean, it depends who you hear. Some people have -- there's probably a 2x difference in projection of how this thing grows, but it starts next year. It starts in 2027.

Can you talk about -- when you see this kind of visibility out to mid next year in some of these markets, can you talk about the risk of double ordering, anything like that going on?

Yes. So I talked earlier about how we're utilizing the extended lead times to get more visibility and expand a little bit more on what I said about tightening the channel inventory. We are really applying rigor to overall demand we are seeing to make sure that customers are aligning their orders from us with what they see as demand and, in fact, trying to do our best to look at the rest of the kit that they have in their BOM, so that they don't have our parts sitting on the shelves and no memory, for instance. And we're then working with our overall sales force in different areas to see what they are hearing on the ground in terms of the end demand at that. And I think there's a lot more rigor that we're applying as we are managing through this with the priority being supplying our end customers with the parts they need and doing additional checks on it. But is there some double ordering? Maybe. But because we're allowing customers to book out in time and giving them confidence we will give them the supply, we think that tendency is diminished. And by tightening the channel, we try to decrease any buffer that's held there as well. So it's not going to be perfect, but we're doing our best to not be surprised by it.

Great. And then the other question we get, when FPGAs start growing into high-volume spaces like AI, is ASIC replacement risk, ASSP replacement risk, can you talk about that?

Yes. I mean, look, if we were a really expensive part that has a lot of power and very visible on these boards, we'd be at risk of being replaced. Today, we're into very reasonable prices, and we have so many applications and across so many vendors. If a hyperscaler wants to come in and design their own to replace us, they're going to miss -- I mean, we learn across all of the hyperscalers, the Neocloud, the enterprise, the server vendors, the ODMs and come up with these FPGAs that are optimized across all these use cases, across all these vendors, across all these functions. And so many use cases, it doesn't make sense to replace us. It's not like we've got one really expensive high-power, high runner that you can easily identify and say, I'm going to replace. I mean this is a widespread, many tens of use cases per rack, per industrial application. So...

Right. In the BOM of our customers, we are at the small tail of their cost. And as Ford said, it's a multitude of functions, so they would have to contemplate doing a multitude of ASICs, which doesn't seem to make economic sense.

Yes. Okay. Helpful. Obviously, a lot of growth on your plate here. But can you talk about M&A strategy? And is there anything that you would think about where you need to add capabilities?

Yes. So look, we've done actually a few small tuck-ins that we haven't discussed, because they're small, like a few either acqui-hire or IP licensing or small tuck-ins to mostly around IP software tools and solutions. And over time, I think we would like -- I think we feel very strongly about the organic growth that we have, it's very strong organic growth. So we don't have to do an M&A. Now if the right one shows up and aligns with what I just described on the vision that we just outlined, it would make sense for us to move forward. But we're going to keep a high bar and be disciplined on how we do it.

I mean do you need scale in this business given the size of your 2 bigger competitors?

Look, scale always helps. I mean, but we can get scale via acquisition or we can get scale via organically growing it. The cost per engineer from the time I joined to now, we drove our cost per engineer from $150,000 per head to $100,000 per head. And so we're growing very fast in India and Penang and Philippines. And so we're looking to grow in Taiwan as well. So we're growing in all these geographies and adding a lot of people organically. So ideally, yes, we can do an M&A, but that's not required.

Yes. The way we have thought about scale on the organic side is exactly as Ford described, Joe, which is we actually lowered OpEx between '24 and '25. We have more engineers. And we're helping the productivity of those engineers, not with just the IP acquisitions, like Ford said, but also we deliberately invested in infrastructure. It's an unnoticed unglamorous-type of investment, but it's really helped the productivity of the engineers. So we're getting to scale that way.

Very helpful. We have 2 minutes left. Anybody has any questions?

Yes. So the question is, can we give a rough breakdown of our Industrial segment. What I would tell you, our Industrial segment could go back to where it used to be. So I think if you think about it, we do break down the industrial and comms and compute. So last year, industrial was about $195 million of the $523 million. So we did $523 million, $195 million was industrial, $290 million was comms and compute and the rest was consumer, okay? And this year, we expect the comms and compute to become over 60% of our revenue from sort of 56% to over 60%, and industrial consumer to be roughly flat and industrial to be the remainder. So you could see what that growth is. And next year, we expect industrial to go back and increase faster. And what happened to us in industrial, we built this inventory, and we've been bleeding inventory at the rate of about 3 weeks per quarter for the past 5 quarters. So we're now through that, and we're down to inventory level that makes sense. So we're now going to be shipping our revenue at natural demand, which will help this year. And next year, we do expect an inflection in industrial because of all these new applications and products that I discussed. So at a high level, we are committed to be a 2-legged stool to start. So leg #1 would be this comms and compute that's very strong today. Leg #2, industrial that's going to get stronger and continue to get stronger next year. And we want to develop a third leg over time, and that could be around the solution angle. So eventually, we're going to have a more stable 3-legged stool to continue to grow the company.

Yes. I'm not sure underneath your question might be what's in industrial. So we actually call it industrial and auto, and you probably derived from the commentary, auto is not a very big part of our business. We don't have a lot of exposure there. But within auto, we have classic industrial factory automation type of applications. We have medical applications. We have defense as an emerging application set for us. We have industrial power management, a bunch of things like that. So it's a very broad category for us. We don't break it out any further. But despite the fact that's named industrial and auto, auto is not a significant part of our overall business.

Okay. We'll wrap it up there. Thank you so much.

Thank you. Good to see you.