Hello. Okay.
We're working.
All right. Welcome back, everyone. My name is Mark Garcia with JP Morgan, very happy to be hosting the management team here of Lattice Semiconductor. We have Ford Tamer, CEO, Lorenzo Flores, CFO, special guest, Sanjoy Maity. Yes. CEO of AMI. Yes. Great. For those who have questions on AMI, this is a great opportunity. Ford, it's been over 18 months for you now. Maybe you can just sort of catch us up, you know, what did you inherit there at Lattice? Really importantly, what's changed the most operationally today?
Thank you. When I joined, the vision of the company was to do a full portfolio of FPGA all the way from low-end to high-end, and there was a roadmap to do chips with, you know, millions of gate and including SoC, which is integrating the processor inside the FPGA. When I joined, we refocused the company on the small and mid-range, no SoC. Meaning we refocused the company on what's called the small FPGA is called our Nexus product line, N-E-X-U-S, and our mid-range is Avant. We have three product line, what we call the pre-Nexus, Nexus and Avant. Nexus, it goes up to about 200,000 logic cell, Avant to about, you know, 600,000, a nd that's the focus of the company.
We did this because our belief is that this is a sweet spot for FPGA. This is where, you know, FPGA make a lot more sense being deterministic, being low power, being low latency, being fast boot time, as a companionship to our silicon partner on the CPU, on the GPU, on the board management controller, on the MPU and MCU, on the networking switch and NIC, on the variety of different sensors. On the sensor, we sit near a image sensor. Let's say we've got partners like Sony on the radar. For example, we announced a partnership with TI.
We've got a variety of sensor work with ADI, image, LiDAR, radar, infrared, ultraviolet, as well as, you know, all kind of industrial sensor, like temperature and then a variety of motor control, so different actuators. We feel this is the sweet spot for FPGA. We're a high volume supplier, this year we'll do over 200 million units, on our way to over 250 million. This is our focus. What's different is we don't try to go after these large FPGA or the millions of gate, you know, where in my background when I was at Broadcom and eventually Inphi on board of Marvell.
I saw these large FPGA going in for 18 months and being replaced with ASIC because they're very high power, very expensive. What we do is the small FPGA. We're literally on a rack of server today. We went from tens of FPGAs per rack to hundreds of FPGAs per rack. Very low cost, very cost effective, very fast boot time. On the SoC, we go partner with folks like NXP or ST or Renesas, where we instead of trying to put the processor inside the FPGA, and then you're stuck with designs our competitors like, you know, Altera and Xilinx have these SoC that are 10, 12, 15 years old, with really old interfaces and et cetera. We go partner with the latest and greatest from NXP and ST and Renesas, where we go jointly to market.
What we show, like, for example, we go partner on industrial control and humanoids, where together we're gonna have the best of both worlds. You end up with the best deterministic low latency from an FPGA and the best, you know, latest tool set, the latest DDR, the latest PCI from a NXP. Together, we're actually gonna bring a solution that's lower latency, lower power. It's counterintuitive. You put two of us together, we're together lower power because deterministic tasks get done on FPGA versus the high-level stuff in a NXP. We refocus on data center. Data center has been growing very fast for us. The server market for us, been growing 85% year-on-year. Demand is strong for the foreseeable future, well into second half 2027 and up into the right. That's a quick summary.
Great. Actually, well, let's stick with data center then, talk about your segments. Compute and communications, as you said, has been growing more rapidly than the very high growth CapEx market. Can you kinda take us through that? What are the drivers there in terms of attach rates and ASPs for you, especially on the AI side within the compute segment?
If you look at these, we, it's a multiplication of various factors, right? Five factors, if you wish. Number one, we grow along with obviously, we grow along with CapEx, and CapEx, you could think of CapEx as a number of servers. We grow along with number of server, and that's been the CPU, and the CPU as recently has been growing faster because of agentic. That's number one. Number two factor that's multiplicative is AI as a percent of the total server, because we do have higher content on AI server than a traditional server. Number one, number of server. Number two, AI content. Number three is attach rate. Attach rate, we've got a number of applications that are continuing to grow.
When I joined Lattice, we had tens of FPGAs per rack. Today, we have hundreds of FPGA per rack. The number of application we're finding and quite a few we can talk about are growing. Number four, the needs continue to grow. Security has been, for example, a big application for us. Over time, the security needs have been growing higher and higher, driving a bigger FPGA per board, and hence a higher ASP. Finally, number five recently was agentic. It's not only CPU, but also storage has been growing. That's number five. The second part of number five is the architecture have been changing.
If you look at the disaggregated architecture have taken over, disaggregated drives more FPGA because like for example, right now, you go from one board being the main processor board to two boards, HPM, where you've got the processor and where we have I/O expansion, and the second board, SCM, which is a satellite board where we do security with root of trust. Also, the unit of compute has migrated to where before it was one rack with everything in it to now four racks, a compute rack, a communication rack, a power rack, a cooling rack. With that, you could see we end up having a variety more application, variety more needs for FPGA. You really take maybe it's really six factors together, and you could see that's why we're growing faster than CapEx.
The latest numbers that we've seen for CapEx for just the four hyperscaler for next year in 2027 is gonna be over $1 trillion. Over $1 trillion. This number continues to grow, and this is just the hyperscalers. This doesn't include the Neocloud, does not include the sovereign, and doesn't include the enterprise. This number continue to grow is very fast, and we grow faster than the CapEx.
It's interesting, you know, Harlan was just up here before and talking about the growth in custom in the data center space. What's really different this time, because it seems like custom and FPGAs are really going to coexist here now, going forward. Even with the growth of custom, the FPGA business is gonna continue to grow at a more rapid rate.
Actually, that's a very good comment. I mean, this is probably number seven, the growth of custom. That's another factor. We are actually totally agnostic to whether it's custom, whether it's standard product. We do wanna stay Switzerland, and that's why this AMI acquisition was so powerful is we are Switzerland for companionship for silicon. Whether it's standard or custom, AMI comes in as Switzerland agnostic for firmware. Together, we're bringing these solutions to customer and continue to support the wide ecosystem. You're right, it's very interesting today to see, you know, the custom going against the standard product. We support all of them. We don't take sides. We support all of them, and we're friends with everyone.
Mm-hmm. Great. Okay. Moving on to the Industrial and Embedded segment, this is potentially the next big wave of growth from physical AI. Big opportunity, maybe you could take us through that opportunity.
That's a excellent question. Yes, you're gonna see these factors superimpose on when you get to 2027. Not only do we have growth in a data center and see demand for foreseeable future, on the Industrial and Embedded, which is what we call now our segment for industrial, for medical, aerospace, defense, consumer, we put everything in this Industrial Embedded segment, automotive. That segment had been dragged down by an inventory situation. When I joined Lattice about 18 months ago, we had six months of inventory in the channel. We said we were gonna bring it down to 2.x, and we are 2.2 as of last quarter. Our goal is to go under two this quarter, and we're on our way to go under two this quarter.
The last time we were at 1.x, we had 10 quarters of strong demand ahead. We do believe we're entering this phase now where industrial embedded should see a really strong number of years ahead of strong industrial embedded. Number one, we are growing now to natural demand as opposed to undershipping demand because inventory in the channel has been normalized. Even more importantly, we've got some phenomenal design win across all these application, industrial, humanoid, drones, autonomous vehicle, aerospace, defense, medical, consumer, that are all going to market this year and will inflect into 2027. Content of some of these, like humanoid, is very significant. We do feel like industrial embedded is gonna be another growth vector for us.
If you go back to the Lattice in the 2022, 2023 timeframe, the growth was all from industrial embedded. Now you put these two on top of each other and should be very strong. Three growth vectors, if you think of us in the future, data center being one, Industrial Embedded being two, and AMI coming on top of it being three.
That's great. Let's go to AMI as the next catalyst of growth here. I think a lot of folks in the audience maybe weren't familiar with AMI, which has this incredible history and is very close to the server CPU, the GPU customers. Can you take us through what does AMI add for Lattice and how does it change what you can offer your customers?
Yeah. I'll give a high level, and I'm going to turn to Sanjoy because he's the expert. AMI started 43 years ago and a very rich heritage in the space. When we started like last week, we did couple calls with the sell side because on our earnings call, it's not clear what it is. Then we're going to, today is one of the many calls hopefully we'll have on the buy side and we'll be organizing some of these calls as we move forward. When the first question when I ask folks on the call, "Hey, what do you think?" after we explain the acquisition, people said, "Oh, let me go get my software analyst." I said, "No, wait a minute. This is not software, okay?"
That's why I think we wanna explain this from the basic level. There's two businesses at AMI. It's a simple business. Two businesses. Number one, 55% of the business come from what's called boot firmware, and the second one, 45% of the business come from manageability, okay? That's it. Very simple. Number one, boot firmware. What's boot firmware? Without AMI, there is no server. There's no data center. AMI basically run on a flash near the CPU. It sits near Intel, AMD, Arm, NVIDIA, custom processors, you know, and it's fundamentally that firmware brings up the CPU. It boots the CPU, and then it looks on the server, looks what components are on the server, recognize that the components on the server, and then it boots the operating system.
The software run on top of the operating system. Think of this AMI as a fundamental layer that's even underneath the hardware. You should think of AMI as hardware. That's how we think about it, okay? This firmware is like hardware. That's about 55% of the business, and we had some great calls with all the different partners on the CPU side. By the way, that AMI business is 75% server when it comes to boot, okay? On the manageability side, think about it. This is firmware that runs on with the board management, the BMC, the Board Management Controller.
These are Baseboard Management Controller from, it could be ASPEED, it could be Nuvoton, it could be an NXP, it could be new guys that are coming to market and not announced yet. A variety of different BMC. Think about the extreme example of having data center into space. I think, you know, whether you think it's gonna happen or not 10 years from now, it will definitely drive the right behavior, which is if this data center in space, you can't really send a human over there to go change some stuff. You want this data center to be totally self-managed, self-healing. Number one, you wanna collect a whole bunch of telemetry.
We actually have customers on the telecom side of the world that are coming to us and saying, "Look, we wanna collect the same telemetry on voltage, temperature, et cetera, to be able to do instead of doing predictive maintenance, we want to do predictive failures." The failure rate of some of these components is actually extremely high, higher than you would think. Collecting all of this telemetry and be able to do predictive failure on AI model based on this data, which by the way, does not exist in the comm-- This data on 43 years of predicting all of this failures attached to voltage temperature resides within the four walls at AMI. It does not exist in the open domain. Even if you wanted to train Claude, you don't have the data. The data is AMI.
We are training the models to be able to do this kind of stuff. It's Number one, it's phenomenal thing on collecting all this thing. Number two, you wanna be able to operate even if the OS is down, and that's what they do. AMI operates even if the OS is down, and you wanna recover from failure, they would be able to do it. They have a view of all the different components in the data center. The high level you could think about the main cockpit for the data center, where you control the whole data center, data center management, could be enabled by the AMI firmware. Very fundamental, very strategic. Finally, the value prop is that they do it across all of these vendors.
You have different people that Intel may have their own boot, AMD may have their own boot, Arm may have their own boot, NVIDIA may have their own boot, but AMI works across all this. Again, different BMC folks may have their own BMC firmware, but AMI works across all this. Being this agnostic Switzerland firmware across all of these different component is a humongous value proposition that we bring. If you want, for example, to do plug and play, so today you're working on a certain component, tomorrow you want to bring a different component. That's what in and by the way, that FPGA companionship is tremendous because we have a low latency. Let's say you want to do a leak detection for cooling.
We'd be FPGA sitting down with a very low latency, being able to shut down that leak faster than any component you ever see because our latency is very low, okay? Think about if you wanna optimize the performance of the GPU. You wanna be able to direct cooling to the GPU that's actually more, that's being more taxed and doing more performance than the one that's idle. Well, guess what? AMI allow you to do this kind of stuff. Very powerful together longer term we think will bring solutions to market. There's a presentation on the website that talks about some of this. That's a high-level view, and Sanjoy, I'll turn it over to you for more detail.
Thank you. As Ford mentioned, right, I will just give another analogy why it is foundational for every server or every data center. Without AMI firmware, I would say that the servers are basically metal and electronics. It's not operational in that state. Its analogy goes like a car, right? You see the ignition system. When you turn on, there is an engine check, there is a transmission check, there is airbag check. All these checks happens, it starts the car. That's our boot firmware. If you think of the continuous, when you are running and driving the car, there is a system continuously checking the temperature, checking the, you know, everything is engine, transmission, everything is running optimally.
If it is electric car, you have a battery management, everything is going. That's our manageability. Without these two, car will be not functional. Same way, our boot firmware and the manageability firmware, the servers will not be functional. As we talk about software, operating system and software. That is the software like infotainment system in a car or a navigation system in a car or something other than playing some apps or CarPlay. Those are the software. We are foundational in this data center. As he mentioned, I will get into little detail that there are, you know, inflection points in the industry today based on three things: scalability, security, and sustainability. Nobody can ignore these three today. That is driving and changing the technology landscape tremendously.
We see a totally different way of making servers and the racks and powering up, and the data centers are built that way. There are tremendous amount of opportunity to accommodate that because it needs a runtime programmable hardware at the fleet. It cannot be determined at the beginning, and FPGAs are very critical to do that. It's the same analogy I would give if you connect to a device to your USB port. It could be camera, it could be storage device, it could be printer, it could be anything else. The software and operating system recognize and, you know, do this automatically. In case of a hardware, there is no programmability. Only programmability you have is FPGA. These technologies are coming because of the scalability reasons, and that is why the growth of this kind of programmable hardware is growing tremendously.
AMI is working with the large hyperscalers as well as the Neoclouds, as well as the enterprise-level customers and ODMs, and try to make sure that this kind of technology shift which is happening in the industry are well pre-POCs level, it is done, and it is delivered by these companies together ahead of time.
Great. Ford, take us through the how you will expand Lattice solutions through or across that AMI footprint, which as you just explained, is these very, very deep relationships.
The one thing I'll tell you is, and I said on the earnings call, and I've done this twice before, I've done at Broadcom. When I joined Broadcom, we were number two on the switch. Marvell was number one. We acquired a company called Level 7 that brought us protocol firmware that we used to come up with this reference design. We had reference design hardware and software that were turnkey tested that we provided to our ODM partners in Taiwan, and very early on, partnered with ODMs in Taiwan to create the white box server and the white box switch. Pre-Broadcom, these were all OEMs. Broadcom invented this white box switch, white box server, and Level 7 was an important piece of the puzzle to bring this reference design to market. The second time I did at Inphi.
At Inphi, we were selling PAM DSP inside the data center, and we had a very difficult decision to decide to go into the module space to do a DCI for Microsoft between data center 80 km. Very different business because we went from a component vendor to be a subsystem vendor. We're selling these modules. The modules ended up being 20% of the Inphi business at the time. That knowledge that we had of having silicon photonics, having these system being designed and tested were tremendous for the component side of the business because when the component side of the business had a question on how to, you know, test, on how to take the stuff to market, the knowledge was in-house. These are two example in my career where adding this capability was tremendous.
When we made those acquisitions at Inphi, people did not understand them. We acquired a company called Cortina, and people thought we acquiring Cortina to go into telecom framer and mapper business. Little did they know that we acquired them because they had high-speed analog people that we wanted. When we acquired ClariPhy, which is a coherent DSP at Inphi, again, people thought we're getting in telecom DSP business. Well, guess what? Nobody knew we're doing this for Microsoft. When Microsoft announced this, it was a shock in industry going to, you know, to this DCI business. These are two examples in my career where I've done this before. We believe AMI would open up a similar opportunity for us.
We've got a slide on our website that shows some of these solutions that we are working on together with customers. Sanjoy talked about POC. Things like rack boot, so people don't want us to boot a server at a time. They want to be able to boot the whole rack. We're working on solutions together. Things like power and cooling, things like retrofit, so people want to go back to the old fleet and be able to retrofit the old fleet with solutions for security. Things like plug and play, you know, where you want to be able, for example, to plug and play new components inside of your existing server. All of these are gonna be enabled by a combination of AMI and Lattice. Where does this take us?
Well, number one, it opens up a short-term opportunity to build a multi-billion dollar company. We said we'll exit this year at a $1 billion run rate in Q4, we're confident this will happen at very good gross margin, very strong cash flow and very strong, you know, EBITDA. We were saying it'd be at least, you know, 40% free cash flow. Next year, we grow even much faster. You know, taking this to the longer term vision, this could be the first one of our many acquisitions that move the company from just a component supplier to a solution supplier.
Today, we're into all these markets where we're seeing, you know, the various market. Today, this is just compute. There's comms, there's power and cooling, there's a whole bunch of industrial embedded type of market along vision and motion control, et cetera, that we can get into. That's where we take the company.
Great. How about on timing? Timing of closing and then thinking ahead maybe to the integration and how to think about success going forward?
Integration is very important. I've done it like 25 times. At Broadcom I had 18 different companies. Inphi we had like six. Even Groq, we acquired a very key piece of Groq that enabled the $20 billion acquisition by NVIDIA. Integration is very key. Sanjoy and I are going down to Atlanta together. It'll be my second trip there. We're gonna spend a few days with the team. We're gonna go to China, not to China, to Taipei, for COMPUTEX. They've got 600 people in Taiwan. We're gonna go to Chennai and Pune, where they've got 700 engineers.
Here's a very interesting data point. There's only 5,000, I'm not sure if this data is correct, but if that's correct, it's like crazy. There's only 5,000 firmware engineers in the whole world. 5,000, that's it. 5,000. Over 1,000 of them are at AMI. Over 20% of the firmware engineers in the whole world are at AMI. If this is correct, this is like totally crazy. Is that correct?
Yes. Yeah.
Okay, he says yes. Integration is going to be important. We're on top of it. As far as delever, we think we delever very fast. The regulatory is just U.S. We think we get this done in 45 days, and we've got an inside date of July 6th. We think we'll close by then. Delever, let me at least ask Lorenzo. He has at least one question.
Yeah.
Delever, Lorenzo.
He's up next.
Yeah, yeah. While Ford brought it up, Ford said, you know, we expect to have combined very strong free cash flow. The debt we are taking on will be a little bit less than $1 billion from this acquisition. We are going to delever rapidly and, you know, be at or under two by the end of next calendar year. We'll be well-positioned. You know, we've always had a strong balance sheet, and we intend to maintain that.
Great. There is one question here. Do you see? In the back first, sorry.
I just have a very simple question, which is, how is it that your return on invested capital over the past few years has been low, or am I wrong? Lorenzo?
It's fundamentally the depressed profitability we've had given the decline in revenue from in the mid-$700s, two years ago down to the five, low $500s where we were. That's what's driving the lower ROIC. We get a lot of leverage out of our business model as we grow revenue. Both the, you know, the OpEx will not grow nearly as rapidly as revenue. This year we're investing. That rate of growth in investment will slow down. Our fixed capital investments have been really high, relatively high in the past couple of years because we've been doing some catch-up investments in infrastructure. That should tail off also. The ROIC should improve-
Yeah.
-significantly.
Think about Q1. In Q1 we delivered 40% year-on-year revenue growth, over 80% EPS growth. EPS is growing at twice the rate of revenue and we should be able to get much better as revenues scale.
Thank you for the excellent insights, really appreciate it. Two questions, one for Ford, one for Sanjoy. The question about FPGAs, obviously you guys have placed a lot of investment into UltraLite. Can you talk a little bit about obviously the AI edge use case? Obviously, there's a lot of questions around new devices going online, localized AI processing. Then for Sanjoy, in terms of AMI, do you have any plans to open source some of the AMI frameworks for OS development or custom OS development? Obviously, with a lot of AI going up online, looking to run the orchestration layer beyond what AMI is doing.
Two great questions. Let me start with the first one and turn it over to Sanjoy for the second one. On the first one, we distinguish between near Edge AI, which is really the heavy processing, and there we do believe that's the domain of NVIDIA, Qualcomm, our partners. We don't wanna go into this near Edge AI. The other FPGA co-competitors, they do, and that's fine, more power to them. But we are very focused on what's called Far Edge AI, which is near sensor. We call it contextual intelligence, near sensor intelligence. There, our number one role is to be this companionship where we ingest all of the, all of the, whether vision, motion control, and then do the pre-processing. Our role, we think, is adding AI as a pre-processor to make the near Edge AI being NVIDIA, Qualcomm more efficient and more productive. Sanjoy?
Yes. For the open source and open architecture perspective, it's not a plan. We already did that. We are the platinum member of OCP. We are the only firmware company in the world who has contributed the boot and the manageability firmware, the open source. As a matter of fact, today we are actually doing a major event jointly with Meta at Meta's campus. Across the industry, including Meta, Microsoft, Google, all the speakers are there, and it is happening today about our open source strategy and what we are doing. We are doing it for last three, four years. We are a big contributor to Linux Foundation OpenBMC. We are the, as I mentioned, the most secure OCP S.A.F.E. is the security compliance stack. At OCP, we are the only one who contributed that and maintaining it f or the industry.
Lattice is also very committed to open source. You know, we've also been a big contributor on OCP, and we do believe that together, we wanna be the Red Hat for the space. You know, with open source, Red Hat grew a great company by contributing to Linux. We'll do the same on manageability and control.
Thank you.
Great. Okay. Clearly, demand is very strong. Maybe Lorenzo, we can talk a little bit about the supply side. What actions have you had to take or are you taking in order to ensure that you have the supply, which is very tight out there, in order to meet the demand?
Sure. I should give some credit to our head of operations, who started to sense the tightness in the fourth quarter of last year and laid the groundwork for getting supplier commitments through the year, starting on the wafer side. Which, you know, we fortunate we use older nodes. They tend to be less constrained, so our wafer supply got locked in. We did actually buy in advance of the demand we saw so that we have wafer supply. We have a, you know, fundamental inventory strategy of building inventory in-house, and we've begun to execute that, like I said, as soon as the Q4 last year.
The other side of the assembly of the overall process is the assembly and test process, and that's where we've seen and the industry has seen a little bit more constraint. When we started to identify that, we went out and we were on a program of diversifying our supplier base, multiple sites from the same vendor and more vendors. We're still seeing everybody else is seeing in the industry and some pressure on cost, price increases from our suppliers.
We've been able to manage that, both with negotiations as well as obviously passing it through where we can to our customers. We feel we're in a good position from ability to support the demand. We have the fortunate problem, good news problem of demand keeps increasing, we have to keep working at it. So far, we've been able to manage the cost side and sustain our gross margins, and we expect to be able to do that for the rest of the year. Thank you.
Great. We're out of time.