All right. Good morning, everyone. Welcome to the 27th Annual Needham Growth Conference. Joining me here right now is FormFactor. Very, very pleased to have Mr. Mike Slessor, our CEO, Mr. Shai Shahar, the CFO, on stage, and of course, we have Stan Finkelstein, the IR, and, well, you have other titles. I kind of forgot always about that in the audience. The management is going to make a very brief presentation, and then we'll transition into a fireside chat. Mike, Shai, stage is yours.
Thanks very much, Charles. And thanks to Needham for hosting us again this year. We were just joking with Laura that this has become an annual staple of our travel schedule, but always a great conference. I want to share with you a couple of slides just to level- set everybody before we get into the details of the fireside chat with Charles. And I don't seem to be able to work the slides. So let me ad- lib here. FormFactor, we're a company in semiconductor test and measurement, about $750 million annual revenue, trailing 12-month revenue, operating in two major segments. About 75%-80% of our business is something called a probe card, which is the device-specific interface between the testers built by manufacturers like Advantest and Teradyne and the customer wafer.
One of the really unique demand drivers in this business, demand characteristics of this business, is you get new demand every time a customer changes a chip design, changes a mask, that. And it's a unique character of this business that shows up and differentiates us from WFE, wafer fabrication equipment cycles, as you've seen throughout history. A thing to keep in mind, the demand for probe cards is driven by new designs, and they are, because of this, a consumable, just like a photo mask is a consumable. The other part of the business is what we call our system segment, about 25% of revenue, where we work with customers, again, test and measurement, both electrical and optical, in developing brand new technologies.
So we're working with customers right now and have been for several years on things like silicon photonics and co-packaged optics, which is poised to really revolutionize chip-to-chip communication in the data center. And these are fairly complex R&D tools and processes, and that's why a lot of engagement early on in the R&D cycle with our customers. Another one we've made some investments in in the Systems business over the last several years is in enabling quantum computing, building test and measurement Systems for the developers of quantum computers to be able to test their devices. And the unique feature here is all of these things operate at very close to absolute zero, which has all its own challenges associated with it. There we are. We're caught up on the slides. We are a global company.
About 2,000 people are arrayed worldwide, although we are headquartered in California, but big presence in regions like Taiwan, Korea, Japan, Singapore. I think we'll get into the China dynamics in the fireside chat. That's an area where we actually divested our direct operations in China about a year ago in the face of some pretty significant geopolitical headwinds, and we'll talk about some of the details of that. One of the core company values is focus on the customer, and you can see some of our recent 10% customers listed down here. We really do take pride in being a leader in the semiconductor industry. We are the leader in market share across our served markets, and we think it's strategically important to be paired up with names like you see on this list to drive our roadmap forward and have high market share at these leading customers.
I won't spend much time on this, but for those of you trying to orient, what are these products? What are these things? The thing on the left is a DRAM probe card. It's about 18 in in diameter overall and tests every one of the chips on the DRAM wafer at once. And so we offer huge productivity to our customers with the ability to test many, many chips in parallel at high frequencies with high power densities. So there's a lot of technology in here. A probe card is really an assembly of, in this case, over 100,000 individual probes that all have to make contact with a very, very small contact area, usually a pad on the wafer. You're talking about scales of 50 microns or so, which, to calibrate for you, is about half the diameter of a human hair.
So some pretty amazing technology, and it's one of the really fun parts of the business for me. Okay, so now maybe what you're most interested in, recent results and our outlook. This is available in our investor deck. We had a very strong second half to 2024, primarily driven by something called high bandwidth memory, a key component of these generative AI modules that NVIDIA is selling to everybody. HBM is a specialized type of DRAM that really allows the GPUs and custom ASICs in AI hardware to operate at the speeds with the compute capability it needs. And so a key enabling component of all of this investment and adoption of AI. In Q3, we had all-time record revenue for the company. Now you see gross margin, and I think we'll talk about the model a little bit, shifting down surprisingly at higher revenues.
This is a function of our product mix. DRAM products for us generally tend to be lower gross margin. And with a lot of the growth being driven by DRAM, you see that putting pressure on the gross margin. Overall profitability remains strong. Our Q4 outlook, we're not updating it or reaffirming it, but just letting you know the way we guided in our October call. We'll provide results. I believe it's scheduled for February 5th and provide Q4 actual results in our Q1 outlook. Themes continuing to drive the business really are around high bandwidth memory. And as we get into the fireside chat, I'm sure Charles will explore that a little bit. Okay.
Thanks, Mike.
Maybe first, yeah, let's dive into HBM. I think since this is a conference, assuming not everybody knows about your story, but let me recap some of the HBM-related business dynamics over the past one and a half years. You first started to report HBM probe card revenue, I believe that was in second quarter 2023. Back then, I believe it's like $10 million-ish revenue per quarter run rate. And then that part of the business steadily growing, then took a big step up earlier in 2024, I believe in first quarter, and then second quarter took another step up. Then since second half 2024, it appears to us that your HBM probe card business may be entering a softer patch. As you said, it's probably your customers are digesting what they have bought and built.
Maybe a little bit looking ahead, are you bullish on HBM given all the dynamics here? Because one thing we know you're not quite an equipment supplier, but on the equipment side, there are a good amount of mixed messaging from different semi-cap equipment companies. Some think HBM CapEx up, some think it's down in 2025. Pretty mixed. From your perspective, your HBM business, how do you view 2025?
Yeah. I'll answer the question, then we'll go back and retrace some background. We do expect our HBM business to grow in 2025 year on year. Now, if we take a step back, it quadrupled. Our HBM probe card business quadrupled going 2023 to 2024. We don't expect that kind of growth but if you look at the fundamental underlying drivers, again, go back to what drives probe card demand, it's when customers release new designs, and so this strong growth in the first part of 2024 was really associated with what's now our number one customer, the leader in HBM, ramping HBM3 and HBM3E designs, and those are chips they're producing now using the probe cards that we shipped to them earlier in 2024. We expect a similar transition to next generation HBM3E designs that we're shipping now in volume.
And of course, HBM4, which becomes even more complex, higher test speeds, higher layer counts, all these things driving up the complexity of the test that needs to happen for the chip, which amplifies our competitive advantage. So we do expect this business to grow sequentially in 2025, again, based on the release of new designs, both HBM3E and 4, and those ramping in significant volumes. Of course, all of this is connected to hyperscaler generative AI investment. And whatever your assumptions are based on that, I think you can look at the pull-through of HBM and GPUs and consequently our HBM business as well.
Got it. We recently watched your keynote at the International Test Conference or ITC. You were on the stage and you cited, you gave a really good overview presentation about the industry, about tests, about probe. But one specific number really I found interesting. You said probe card intensity, basically the probe card market TAM divided by overall semiconductor revenue has been about 0.4%, right? And you did provide another number. Of course, that number you cited, the other broker, my competitor, 0.9% in HBM. That's the probe card intensity in HBM. It's slightly more than twice the intensity. So when it comes to intensity, any reason to believe that intensity will stay there? Obviously, I know you guys are pretty conservative. I'm not expecting you to say that intensity is probably going up. I don't think you will say that.
But any reason to believe it's going to maintain there instead of probably going down, maybe getting back closer to that 0.4%?
Yeah. Yeah, well, thanks for the plug on the ITC keynote. That was a fun experience. The test intensity for HBM and more generally, all of these advanced packaging and chiplet architectures does tend to be quite a bit higher than for monolithic chips, and I think HBM is a great example. High bandwidth memory or HBM are these memory cubes that are built from stacks of individual memory die. For HBM3 right now, typically eight high. Each of those chiplets in the stack needs to be tested pretty comprehensively before it's committed to the stack to make sure it's going to work. Because if it doesn't, you end up in a situation where one chiplet can kill its seven neighbors, right, and you end up with a very, very large cost event associated with this yield loss.
It's one of the reasons customers for all advanced packaging applications, we see this in our PC business as well with one of our customers' adoption of advanced packaging and chiplet technologies. We see these increases in test intensity compared to the industry standard 0.4-ish%. Given the dynamics in HBM, we're going to higher and higher stacks, 12 high, 16 high, higher test speeds, larger temperature ranges. We do expect this higher test intensity for HBM to persist certainly as we go through HBM4 and into HBM5. Customers will get better at yielding these parts. And so that'll be somewhat of an offset. But the fundamental dynamic of trying to build chips through stacks of chiplets is driving up test intensity and probe card intensity.
So sticking with the HBM topic, right? We see that you talked about that as well, right? Most of your HBM probe card revenue currently is tied to one customer, actually the leader in HBM. But we see a little bit more spread out, like revenue mix from the equipment side, not as concentrated as what you are seeing. So why is that? Any sort of explanation you can provide us? And what is the prospect for maybe more revenue opportunities with the other two HBM customers?
Yeah. As we've reported HBM throughout 2024, I think we've been fairly clear with people that we do have contributions from all three major DRAM manufacturers. However, as you note, there's one leading in HBM that really is dominating our revenue. I think that's a function of two things. First of all, that customer who was for the first time in about a decade our number one customer in Q3, that's a very long-term relationship for FormFactor. And because of that long-term relationship and some of the co-development on the test methods, the probe cards, the technologies required for the specialties of HBM test, we've got a very high market share position there. So I think we're over-indexed to them. That turns out to be a good thing as they lead the HBM market.
Having said that, some of the special technologies that we have, the differentiated technologies we have, things like the ability to test at very high clock speeds, we think this is going to result in share gains and adoption at the other DRAM manufacturers as well. And as they inevitably grow share in HBM over the coming years, we think we're in a relatively strong competitive position to gain share there as well. So I think we've got the right constituents. It's probably consistent with FormFactor's overall strategy to try and be a key supplier to all of the leading customers, let them duke it out on their market share battles, but make sure we're in a position to enable all of them as they move forward on their technology roadmaps.
Yep. One of the things about HBM, right? I think going back maybe one year, you said that because if anyone look at that, look into that GPU package, look at the silicon area of the actual GPU chip and the six or eight modules of stacked HBM, the HBM silicon area is way bigger, multiple times higher, right? I forgot the exact number.
It's about an order of magnitude, about 10 times.
Yeah, about 10 times. 10 times the wafer area, silicon area. But still, I would like to talk about. That's a good place to be, HBM. Make no mistake about it. But let's still talk about AI chips, the GPUs, the custom ASICs. So those chips historically did not require what you provide as MEMS-type based probe cards based on our understanding. While we certainly still doubt ASICs will get there at least in the short term, but it looks like there may be some transition going on the GPU side. Do you see any upside opportunity for FormFactor to really get into GPU maybe a little bit down the road to custom ASICs?
Yeah. We do, and there's some good engagement going on. It's not big chunks of revenue right now. But to go back into some of the drivers that you alluded to, we and our major competitor in logic use a technology called MEMS probes, right? And you can think of this as essentially using semiconductor-inspired technologies, lithography etch, plating, planarization to build each of these individual probes that go into the 100,000-probe probe card. You don't really need this technology if you're a customer until your pitches that you're trying to probe, the distance between the probing pads gets below about 100 microns-120 microns, right? And so legacy technologies are much lower cost, right? And there's a wide array of suppliers around the world who can provide these legacy technologies.
Up until essentially now, major GPU manufacturers have been able to use these legacy technologies that are much lower cost. We believe lower value as well. As they adopt advanced packaging technologies like TSMC's CoWoS to take the GPU and package it together with these HBM stacks, that all of a sudden drives a reduction in pitch that absolutely requires a MEMS probe card. These legacy technologies no longer work at these smaller pitches. We're at an interesting juncture here where a wide swath of the market, the GPUs, are going to need to adopt the MEMS probe card. We're in the midst of qualification there, as are our primary competitor and leader in logic MEMS probe cards.
And I think it's an interesting, exciting area to gain some market share overall, probably not against our primary competitor, but against the remaining field of legacy probe card providers. We've seen this happen in CPUs a decade, a decade and a half ago. We saw it happen in mobile about a decade ago. So the steady progression as chips get higher and higher interconnect densities, to probe them, you need higher and higher probing densities. And that really does require these MEMS probes.
Yep. Thanks for the update. I mean, but if we look at your business, right, the microprocessor and mobile appear to be still probably the majority of your revenue today. Those two end market segments, they don't seem like they have seen much growth recently. AI has been strong, but the non-AI portion of the semiconductor industry, it's still in a pretty weak or maybe lukewarm kind of state. What are your expectations going forward for those two end markets that your business going into those two end markets? And I think specifically I want to ask because microprocessor for FormFactor largely means Intel, right? What do you think what's going on at Intel could benefit or impact your business going forward? I believe we've had this discussion around 2021, right, when they had that IDM 2.0 strategy.
And there was some discussion maybe they will outsource a little bit more wafers to TSMC, what that means. Back then, I think you said because Intel under that IDM 2.0, they still want to control packaging test. That's still done in-house. That's probably not going to be any impact for you guys. Turns out that's right, right? If you look at your Intel business in 2021, 2022, of course, after second half 2022, that's a cyclical driver overtook that dynamics there. So at this particular juncture, right, when that particular customer is kind of facing possibly another strategic pivot, what do you see right now? And what do you think is likely what will be going on for the next year?
There's a lot to touch on in that question. Let's stay maybe at the industry level because it's true for us, and I think it's true for most broad-based suppliers in the industry, that PC and mobile are very important end markets. They drive a ton of wafer volume, a ton of customer investment, whether it's an IDM like Intel or the foundries, right? And those markets are weak and I think continue to be weak. Our outlook through 2025, we do not see any real signs of growth in either mobile handsets or PCs. Now, one of the interesting features of our business is it operates on very short lead times, right? Our typical lead times are six, seven, eight weeks, well within a quarter. And so our real visibility into any kind of handset growth, PC growth is always going to be a little bit limited.
But I think consensus views of both of those high unit volume consumer-driven markets are not all that optimistic for 2025, and we don't see it any differently. And that's important because those are key drivers of our business. And I think they're key drivers for, as I said, almost any broad-based supplier in the semiconductor industry. If we deep dive on PCs for a second, you're right. Intel has been a major customer for many, many years, almost always a 10% customer for us as we report, and a very strong, again, just like SK hynix, technology partner in developing our roadmap. They're obviously going through some challenges, right? And we'll have to see how they respond to bringing in a new CEO, what strategic changes, if any, that drives. But I think it points to the example you gave, right?
We want to be a supplier to all the leading customers in the industry. So irrespective of any one customer's production strategy, if they want to do it in-house, great. We've got a position there. We can serve that. If they want to do it at the foundries, great. We've got a position there. We can serve that. It's one of the really important pieces of our engagement strategy. There's a few holes to fill in, but from a directional standpoint, we want to do whatever we can to make FormFactor's opportunity set and revenue stream as neutral to some of these big changes in customer strategies as we can.
So moving on, I think there are some very interesting dynamics going on in the test and probe card industry, right? Just earlier this month, you guys announced an expanded partnership with Advantest. Can you tell us more about this partnership or what that means? But I want to abstract what's happening between you guys and Advantest a little bit. So in November 2023, there was a similar or maybe more I mean, a more deeper partnership between Teradyne and Technoprobe that got announced. And Teradyne made an equity investment about 10%, right, in Technoprobe. And there are some transactions involved. But historically, right, our understanding is ATE companies, probe card companies, that's an open ecosystem. ATE companies stay neutral in between the probe card companies. Probe card companies stay neutral in between ATE companies.
But some of the dynamics, at least from our observation, seems to be changing. Seems like Teradyne is inching a little bit closer to Technoprobe. You guys are inching a little bit closer to Advantest. What is the fundamental driver of what's happening here? I mean, I don't mean the business. I mean, is there any technical reason that's driving this? Because presumably integration, vertical integration, there must be a reason if you're moving that direction. Or maybe you're not, but want to leave it to you and provide us some comments.
I think so. If you want to read the press release and that sort of thing, they were last Tuesday, I think. There are two fundamental drivers behind closer coupling between the consumables manufacturers like us and Technoprobe and MJC and the tester manufacturers, people like Advantest and Teradyne, and it's got to do with both the technical complexity we're having to deal with in testing devices like HBM, as we talked about, chiplet-based microprocessors, and making sure each of those chiplets is tested and known good before they go into these advanced packages. That's really hard to do, and as you've probably noticed in compute, NVIDIA has decided they want compute to now run on a one-year roadmap, so the things we're doing are hard, and we now have to deliver them on an accelerating roadmap.
That really demands closer collaboration between the various elements, the various subcomponents of the test ecosystem, and the fundamental driver or drivers, my view, and I think some of the other management teams in the industry share that view, is we need to make sure that our engineering execution is essentially flawless in delivering these capabilities on these accelerated timelines, and so that's really the driver behind this. I firmly believe the industry is still going to operate with this open ecosystem best of breed model. There are advantages for us. Our customers certainly like it from a supply chain hardening, a supply chain business continuity and diversification perspective, and I think it's a model that served the industry well over the years.
Having said that, right, the technical requirements and the speed at which those technical requirements are flying at us demand a little bit tighter collaboration model between all of us.
Great. Thanks for the additional color on that. Maybe two more questions. The first one, I would like to touch upon the recent round of U.S. export controls announced in early December. And this is probably a broader China question for you. And we've seen a couple of things, right, that in Q3, you guys did have a little bit of a jump of the DRAM probe card revenue in Q3 coming from China. And you guys said, well, presumably they think that's their last time buy, but they ended up not really getting on any of the Entity List. But the other part, right, this round of export control restrictions does seem to be a little bit tighter than people would have expected other than the piece that the DRAM company didn't really get put on the Entity List.
But a little bit going back a little bit earlier, I believe you guys have, like you said, divested your China operation and sold it to a company called Grand Junction. And what's the strategic thinking around what's the status now, what's the impact of new export control, and what's the rationale behind that sales of the China asset?
Yeah, sure. I'll take this one. Let Mike rest for a minute. So we've been very open in the last three, four years that we expect our China, local China revenue to go down to zero. We saw these restrictions getting more and more strict, let's say. And it wasn't a surprise that more and more companies were added to the Entity List. In anticipation of that, we sold our China operations to Grand Junction, which now operates as our distributor in the area. But if you look at our China revenue in Q2 and Q3, it was in the 12% and 15%-16% of revenue. A big part of it was international companies like Intel, Samsung, and SK hynix operating in the region. And only a small portion was local China.
Most of it is the local DRAM manufacturer that even though it was not added to the Entity List, some aspect of the restrictions made it very difficult, at least for us, to ship to them, and that's why we talked about in Q3 mainly, second half of 2024, some sort of last buys or almost last buys from that customer. That drove a lot of DRAM increase in our business. We talked about mix in the beginning. That drove the margin down in the expectation in Q4, and we expect this to have negative impact on 2025 and going forward in line with what we said we've been saying in the last three, four years. This revenue will go down to zero at some point.
Probably given the latest round of export controls and the trajectory, going to zero quickly.
Thank you. I believe I still have time for one more of my question. So one of the things about the target model, but I'm not going to go there about target model. I'm more interested in the CapEx trajectory over the past few years. I think you have a range, a target range for CapEx spending per year, somewhere in the $30 million per year. But over the last three, four years, you've been spending well above that range. Obviously, we understand how this industry works, right? You got to have the capacity before customer is willing to give the business to you. But it does seem like you have put in a lot of a good amount of capacity there.
So, mind if we ask how much of the revenue run rate your current capacity can actually support, because your revenue run rate just got back to the prior peak, right? Maybe. But is there any more headroom to go even higher?
Sure. So it's true. We've been investing a lot in adding capacity, probably beginning of 2020, when we saw our customers adding capacity and they wanted to make sure we are ready. Most of the investment was in adding a building in our Livermore campus in the Bay Area, building a clean room, but in some other locations as well. We said there's going to be a transition period, this investment that we're going to invest $60 million-$70 million a year to increase the capacity and then go back once we reach the target model of $850 million. CapEx is expected to go down to 3.5%-4% of revenue. We are on our way there. We invested less in CapEx in 2024 than we did in 2023.
And we have enough capacity to achieve our target model of $850 and beyond, probably $1 billion or even more. It really depends on the mix of products coming into the factory. But we have enough, maybe one more thing because it impacts gross margin as well. When you think about the three components of capacity, the facilities, the tools, and the labor, labor is adjusted to the level of revenue that we need now or really next quarter. Tools, we have a little bit more than that. You cannot turn on a tool and then turn it off when it's not used. So we have a little bit more agility there. And facilities, I mentioned the building, we only built half of it as clean rooms. So we have more room to grow there. So we have what we need to achieve our target model and beyond.
Go beyond that. Yeah. Okay. All right. I think it's time for us to open up the questions Q&A to the audience. Yes, please. Yeah. So sorry, I'm not so sure on the microphone is able to capture that. So let me repeat the question. Can you talk about market share? I think it's pretty generic. Maybe you can go segment by segment and specifically about HBM revenue in that digestion period. But still, can you provide some comment on HBM?
Yeah. I think if you look at our served segments, if we start at the top, there are really two supplier markets, right? Us and the competitor. And I think it's interesting that in each of our served segments, Foundry and Logic, Probe Cards, DRAM, and Systems, they're different competitors. We've tried to take a strategy where we weave together the common technology and customer pieces in those and try and create leverage from them. Each of our competitors in the segments tried to be more focused in the segment. And you can decide which you think is a better strategy. That we think allows us to scale more efficiently. So in each of those markets, we're either number one or number two. If you aggregate them all together, we're number one because we're the only one that creates scale in all of them.
I think in HBM, we're fortunate, although it is part of the strategy to have very high market share at the leading HBM manufacturer, right, and so that drives significant HBM overall share, but also significant DRAM share growth for us as well. We do have good relationships with the other DRAM manufacturers as well, but not as strong a share position inside the other two DRAM manufacturers we currently have with the leader in HBM. I do think, again, going back to the theme of advanced packaging, this is an area where we've focused an awful lot of our technology investments and R&D roadmap on things like higher test speeds so that these chips can be tested at speed and are known good to go into the stack. We believe this is going to drive some structural market share gains as we move forward, right, in the coming years.
We have time for one more question.