Two of Citi's 2026 Global TMT Conference. My name is Elizabeth Sun. I am in the Citi's Semi and Semi Cap Research team here. With this session, we have FormFactor. We will have Mike Slessor, the CEO of FormFactor here on the stage. Welcome. We also have VP of IR, Stan, in the audience as well. Welcome both. Mike, on a high level, personally, I think it is a good time to invest in semi testing market. Maybe high level, can you talk about the trends we are seeing in, let us say, customers making more investment in the testing space overall, or maybe just probe cards specifically?
Yeah. First of all, thank you, Elizabeth, and thanks to Citi for having us. This is always a great conference coming out of summer to touch base with people and understand where the different data points in the industry that help inform our strategy as we go through the back part of the year. Thanks again for having us. It is a very good time to be in test in the semiconductor industry. Really, if I deconstruct it into the fundamental driver of that, it is a theme we have been sharing with people for the past several years that has been informing our strategy, and that is the adoption of advanced packaging or chiplets in constructing a lot of the advanced semiconductor products in the industry.
It turns out that breaking up a chip into chiplets puts a very high burden on testing, because before these chiplets then get reassembled into the final structure, our customers need very high certainty that each of those chiplets is good. Otherwise, they can end up with a scenario in something like an HBM stack where one of the 16 die actually causes the entire stack to be essentially scrapped and lost. So the burden for test has gone up, customer spending on test has gone up, and we see our markets and our business growing quite substantially, driven by that tailwind, which fundamentally has its roots in advanced packaging and chiplets.
Yeah, that makes sense. Before I get into each of the end markets, I want to ask about probe card's intensity overall. Historically, you talk about probe card intensity is about 0.4%. With semi sales growing rapidly, and partially I think that is because of pricing, but also there is a lot of new opportunities in what you call advanced probe cards in ASICs, GPUs, and other things. Are you expecting this so-called probe card intensity to grow over the next few years?
Yeah. It is a great question because, historically, what we have done is looked at probe card intensity and defined it as essentially the fraction of our customer's revenue that gets reinvested or spent on probe cards. As you note, although it has been growing, for the past couple of years, it has been around the 0.4% level. We see that distorting pretty substantially for two reasons. One is probe card spending, at least on a per unit basis, is going up, but it is not going up nearly as fast as some of the ASP increases in some of the end markets like DRAM. If you look at the ASPs for a DRAM chip, they have exploded. Although we are seeing increased spending on test, as we talked about, driven by advanced packaging, we are not seeing these spectacular increases in ASPs driven by some of the shortage like our customers are.
I would expect that definition to be pretty distorted here as we go through 2026 into 2027, given the industry pricing dynamic.
Yeah. No, that makes sense. I guess the more important part is per unit is growing.
Yeah.
That is important. I want to ask about each specific end markets. I want to start with custom ASICs, because I think that is kind of the newest into the advanced probe cards market. You talk about a couple of multi-million dollar projects, but it feels like more to come on that front.
Yeah.
How should we think about the opportunities for the ASICs for advanced probe cards overall? Then we can talk about computing.
Right. Well, ASICs, an interesting opportunity, and we are talking here about the custom ASICs, things like Google's TPU, AWS's Trainium. Essentially these very high-end compute chips or clusters of compute chips that are becoming one of the key elements of the overall high-performance silicon powering generative AI. We have highlighted for people some traction and some design wins over the past year or so. Those are really driven by the migration of these ASICs into a space that requires an advanced MEMS probe technology. This gets into the details of one of the ways FormFactor differentiates. We build each of the individual probes on a probe card. A probe card is this array of tens, if not hundreds of thousands of individual probes that contact the chip and transact electrical signals in and out of the chip.
We build each of those probes using a proprietary advanced MEMS technology, special metallurgy, special patterning, all vertically integrated inside our facility in Livermore, California, and our upcoming facility, which is beginning to ramp in Farmers Branch, Texas. The technologies we have invested in here really enable customers to drive higher current, higher speeds, operate at higher temperatures, and in some cases, lower temperatures. Custom ASICs are embarking in a region here where the power both consumed and produced by these chips is driving their performance envelope to an area where a MEMS probe with these advanced metallurgies and advanced structures is really required to do tests at a productive level, high uptimes, the kind of productivity our customers are used to. Up until now, legacy, essentially wire-based needle probe technologies have been sufficient.
We see custom ASICs really crossing this brink in 2027 and into 2028 of needing an advanced MEMS probe technology, again, to test in a productive, cost-effective way, much the same way GPUs did a few generations ago. Went through exactly the same transition for those interested in the history.
Yeah. Are we expecting all the new programs coming online in 2027 beyond to adopt this advanced MEMS probe cards, or it is going to be mixed?
In different measures, right? Any incumbent technology always is going to be able to be extended further than-.
Yeah.
You know, one might expect. If you look at the architectures associated with some of these ASICs, they are also built using these chiplet technologies, where perhaps some of the chiplets can still employ a legacy wire-based probe card technology and will not need the advanced MEMS technology. Whereas, maybe the advanced compute chiplets do require the MEMS probe technology. Having said all of that, we do see 2027 as the beginning of this pivot toward advanced MEMS probe technologies, where we see, if you like, the marker requirements moving towards our strengths, our competitive advantage. As we described for people at the May investor day, we see this as one of the real growth drivers on the way to our $1.6 billion, sorry, $1.6 billion target model in 2030. Got ahead of myself there.
Yeah. So if I may, how many customers or how many programs are you currently engaged in ASICs?
Yeah. I'm not going to enumerate them for you, but all the names you'd expect us to be engaged with and their foundry and OSAT partners we're engaged with. We have seen some initial revenue associated with some of the projects that we've shared with people. Again, I see as we move through the latter part of 2026, win some of those designs, get some of those qualifications, and begin to ramp in 2027, I think they'll become a more prominent piece of the FormFactor revenue makeup. Now you may not see them as, for example, 10% customers because we've continued to grow the overall business pretty significantly. We're essentially operating at $1 billion annual run rate now. So if you like, the threshold or the bar to become one of the named 10% customers continues to get higher.
It's all part of us building a broadly diversified set of growth vectors for FormFactor to execute to on the way to the $1.6 billion model.
That makes sense. Since you talk about 10% customer, I want to ask about one 10% customer, on the networking side. I want to ask on the other side, on the GPU side. So I think you are more engaged with them now. Can you update us, what is your current view of how big the GPU market can be for you guys?
Yeah. So qualifying for GPU applications, primarily as the world's largest foundry, has been a key initiative for us over the past couple of years. As we updated people on the last earnings call, we've now received qualification and are shipping pilot production volumes that are being used in production, for GPU test at the world-leading foundry. We're still sort of sizing that market, but I think it represents, if you look at both our primary competitor who serves most of that business, as well as some of the ATE vendors who supply the test equipment for that, it is a pretty significant market. We initially sized it maybe a little less than $100 million a year. That feels awfully conservative at this point.
Yes.
A big opportunity is in front of us. We have executed well on the initial qualification and beginning the ramp, and it is another element of these growth vectors that we will begin to be able to see in our revenue in 2027. An important part of delivering to these is expanding our capacity. We are pretty constrained on capacity right now, and our ops team has done a fantastic job in squeezing more out of our existing capacity footprint, but we are really eagerly looking forward to ramping our new Farmers Branch, Texas facility as we move into 2027.
Yeah. I will touch a bit on the capacity later. Just staying on this topic, how are you expecting your share progress at this GPU customer?
Yeah. Anywhere where you are displacing an incumbent, you are going to have to really show some unique value to significantly gain share and not just be a second sort. We think there are some elements to the way we have implemented our advanced MEMS probe technology that offer our customers higher uptime, better MTBF, better overall test cell productivity. We will see how those pan out as we go through the initial production ramp. Assuming those things hold, we will be able to gain on the next generation, maybe somewhere in the 20%, 25%-30% share range, I think. I think that is a useful goal for us to have in mind.
Within one year for the next-.
Yeah. In the next generation.
Got it. Then on HBM side, in my model for HBM, if I look at correctly, your HBM revenue was up maybe 70%, 60% year-over-year in first half. Can you decompose that? How much is driven by volume versus pricing and/or share gains? Yeah.
Yeah. HBM is an interesting counter example, if you will, where we are the incumbent. Very strong market share position, certainly with the leader in the HBM market. Our largest customer is the HBM leader, and that's no accident. But as we've updated people over the past several quarters, we've also made great strides in gaining share at a second HBM customer. Really where we're focused is on the high-speed test, where we really differentiate. When you talk about the 70% growth on a year-over-year basis, some of it's been due to those share gains, adding a second customer, some of it's just due to the growth of overall HBM bits as the industry went from HBM3 to HBM4. A lot of increase there.
We had some ASP uplift, not necessarily because we raised prices or increased prices, but because we were delivering a higher value configuration, able to test at the 11 Gb ps spec of HBM4 and test hundreds of die at once for the high productivity that our customers need. That provides a pretty unique market position for us.
Is gross margin for HBM better by each generation? Is it possible ever to get to the foundry/logical level of gross margin?
Well, gross margins, and I'll take this from the standpoint of overall DRAM probe cards and mix, because one of the admittedly Achilles' heel FormFactor historically had was if we were going to have a strong DRAM quarter, you can guarantee we'd have a weak gross margin quarter. We've done an awful lot to really mitigate that by improving our operational efficiency, by differentiating in applications like HBM, but really reducing the overall cost structure associated with producing DRAM probe cards. I'll give you an example. As we go from Q2 to Q3, we expect to again have record levels of DRAM probe card revenue with actually the mix shifting to DDR5 as our customers optimize their wafer start mix and take advantage of very strong spot market pricing. We're also guiding gross margins up.
Right? Up to 54%. Now, there's some one-time tariff benefits in that, but still a gross margin up in the 50s, as we show the progress towards the 55% target model. We'll continue to work on making sure that we're differentiating and getting compensated for that differentiation in areas like HBM high-speed test, which helps gross margin, but also on reducing the fundamental COGS of things so that we're able to deliver revenue from all the elements of our diversified revenue mix, while still delivering strong gross margin.
Understood. Mix is not a big factor in-.
Yeah. Simply put, trying to take mix and mix shifts out of the equation of determining what our gross margin, and therefore, op inc and EPS are going to be.
Understood. There's a lot of discussions around this packing in HBM.
Yep.
I just want to hear your thoughts on what it means for probe cards. On the one side, people argue there is less complexity because you are lower, not that many dies, high stacks. On the other side, you need more units, and you also need more networking, actually.
Right.
Which is your strength, actually.
That is right.
Net net, how does it work for FormFactor?
Yeah. Net net, I feel like it's much ado about nothing. There's a lot of noise associated with this HBM3e spec . This is a very common, at a higher level, kind of a systems level architecture optimization that our customers do. We've seen Apple do it for years with handset configurations in mobile and memory content and different pieces. I think this is more of the same. Certainly, there's sensitivity associated with HBM complexity and how it impacts our business and the whole supply chain business. But for right now, as you say, there are multiple offsets. For example, there will be more HBM units associated with that, and then the networking infrastructure where we have a strong position to transact the data across these units probably goes up as well.
It's one of the fundamental reasons why we've tried to build a diversified business that has exposure to all these pieces of the semiconductor industry. If we were just an HBM one-trick pony, then this might be cause for concern. Given the fact that we have nice exposure to some of the other areas that would compensate for the HBM3e spec or the squeezing of the balloon in other places, we feel pretty comfortable that net net, this is going to be neutral for them.
Understood. May I ask about the third HBM customer? How's your progress there?
Yeah.
Can you share?
Yeah. Well, historically, we've been a significant supplier to all three major DRAM manufacturers worldwide, and that's still true. Our major competitor in the DRAM probe card space does have a very strong incumbent market share position at that third customer. Whereas, with our number one customer, obviously we have a strong incumbent market share position. That's not an unusual situation given how those customers manage their supply chain. We do see some differences in test strategy between those customers, that maybe we don't see as strong an adoption of our high-speed stack- die configuration at that third customer, but we see some. Again, as we bring more capacity online, we're primarily in this capacity-constrained environment.
As we bring more capacity online, we do see an opportunity, and there is a very active conversation with that customer about what to plan for in 2027 and into 2028, as we have more capacity available to serve them.
Understood. Because you guys are very popular in the photonics space, I want to touch base on that. First thing is, recently there's a lot of discussions around NPO, and I think you guys are pretty strong across Silicon Photonics overall. I just want to understand, are you agnostic to NPO versus CPO and the people talk about four insertions for CPO, and how does it different in NPO?
Yeah. I think broadly, photonics is one of the key growth themes that I'm really excited about for FormFactor. We've been investing in this area for the better part of a decade. Although we are seeing the initial real growth impact associated with CPO here in 2026, we started the year with a view of maybe $10 million-$20 million in revenue. Have now successfully upgraded that to reaching $20 million by the end of the current third quarter. It's a leading indicator and maybe an early inning indicator of the broader photonics growth vector that we have and that we've been investing in, both organically, as I talked about. But the most recent acquisition we made of Keystone Photonics in late 2025, really an enabling technology for optical probing.
The way we are serving photonics is really the same way we serve the broader semiconductor market with electrical probes. Here we are building optical probes and optical probe cards, and optical probing systems. Going back to the CPO versus NPO and even extending to the pluggables question, we have got pretty nice exposure across each. What we have tried to do, and I think successfully done, is build a fundamental optical probing technology which is applicable across all of these. CPO, because of the investment some of our key customers are making, is really making a positive impact on our growth and revenue here in 2026. Much more broadly, I see photonics as one of the growth vectors for FormFactor longer term.
A little early to say whether we would be agnostic CPO to NPO, the pluggables, but again, we have got pretty good exposure across all of those with photonics probing, photonics test as a fundamental theme that we are going to continue to drive a leadership position and invest in.
Do you see some pull-ins from NPO testing?
I would not say from at the scale we are seeing CPO impact our 2026 revenue and that acceleration. We are not seeing the same kind of pull-in, but there is lots of activity for sure, across the customer base.
Understood. I know there's one big customer or big group in the CPO market that you generate a lot of your revenue from.
Yeah.
Are you engaging with the more broader, all the CPO players out in the market?
Yeah, we are engaged very broadly. If you go back, again, there's some details of this where we tried to share the history of both CPO and Silicon Photonics and our engagement in the industry. We've got hundreds of systems installed worldwide at close to as many customers, so a very broad customer engagement. All the names you'd expect, where we have systems in their labs, in some cases, in pilot production. One of the exciting things that we're now seeing in the second part of the year is broader production adoption of this TRITON high-volume manufacturing platform at multiple foundries with multiple fabless customers, so seeing it broaden out a little bit. As you said, right now, pretty concentrated, with one major foundry fabless partnership, but building backlog and building the breadth of the backlog where this diversifies a little bit beyond that one customer set.
Got it. Yesterday we had one of the photonics players here on the stage, and when asked about bottlenecks, they were like, "The biggest bottleneck is the testing space." It is probably not new to you, but I am just curious, a year ago versus now, are you seeing this testing getting harder and harder? I know you are more in the insertion one and probably something in two, but from your perspective, are you seeing there is more challenges than you previously expected when you took on this challenge?
Yeah. Well, being a bottleneck is a double-edged sword, right? It is a good place to be because it means you are relevant and you have chosen to solve a problem that the industry needs. However, it then does demand the resources to help resolve that bottleneck, and I think that is exactly where we are with CPO, having installed now tens of systems, ramping those in production, resolving some of both the technical and operational challenges in real time, partnering with the customer. You only work up that learning curve by being in the game and having systems installed. Is it more challenging? Probably not from a macro level. I think some of the details of the things we are learning, we couldn't have expected or anticipated without really participating in this production ramp. I think it is what goes to the incumbency and leadership that we are building.
That learning is invaluable, right? As you go through sort of the successive cycles of learning and make sure that we are solving these problems and resolving these bottlenecks associated with test in photonics production.
No, that makes sense. I am going to take a pause here to see if there are any questions in the audience.
Thank you. I'm curious on the gross margin kind of go forward, you've obviously had amazing success in expanding that. So I'm curious, looking forward, how much of it is expected to be price or finding efficiencies? Just on the go forward in terms of how you're viewing the margin profile.
Yeah. Well, I'll ground you in, again, what we shared with people at the Investor Day and the target model that produces 55% gross margin, $5 of earnings per share on $1.6 billion of revenue. Really, the trajectory from where we are today, we printed 53%, give or take, in the second quarter, guided the third quarter to 54%. There's some one-time tariff benefits there, so we're not as close to 55% as it might look at a high level. But it's really the same playbook as the gross margin improvement we've executed over the past year or so. Most of this is, some of the analysts have called it, internal self-help, us improving yields, reducing cycle times. I'll use this as an opportunity to give a shout-out to our operations and engineering team on really making FormFactor a much more efficient manufacturer of these specialized products.
Some great progress there. Very little of this is due to pricing. We're a value-based pricer, so something like HBM high-speed tests, where we're delivering unique value, we expect, and our customers do compensate us for that. But we don't price our products like it's a commodity, right? This is not a supply/demand commodity-based market, at least in our view. We think, in a cyclical industry, right now, a very strong upcycle, but there will be a downcycle or a pause at some point. We want to make sure our conversation about our prices continues to be about value and not supply and demand. So really, the gross margin improvement to date has all been COGS-focused. Much of it going forward is COGS-focused as well, and of course, delivering differentiated products that our customers compensate us for.
Thank you very much for sharing. A quick question. You mentioned your opportunities of the business, right? So from optics and GPU and any other new areas, do you think if CPU ratio to GPU goes up, would that be positive for a business? Also, you mentioned you don't think you're as a commodity for your products. So going forward, if we reach a more balanced state of supply and demand, what do you think will be some of the challenges? Even for the current upward environment, what could be some difficulties in terms of accelerating your growth?
Right.
Challenges you face.
Well, there was quite a few questions in that question. Let me start with the growth aspect to it. As I said, right now, the simplest way to think about our situation is we are capacity constrained. What's limiting our growth, if you will, or what's constraining our growth, is our capacity. That's one of the reasons why we're making significant investments in ramping the site in Farmers Branch, Texas. We're going to have a grand opening in late October. Customers, lots of different people. That's really a milestone in us qualifying and bringing that site online, a site which we expect to deliver twice the current California probe card capacity in the current phase. That really unlocks one of the key limiters to our growth.
If we look at some of the market-driven or demand vectors for that growth, yeah, we talked about photonics, we talked a little bit about GPU. One of the ones you touched on as well is CPUs, where we have a great position. One of our historic long-term customers, one of the major CPU suppliers to the industry, their business obviously getting stronger associated with some of the turnaround they've done and the CPU strength. We recently qualified and won some designs at a fabless CPU and GPU manufacturer. In addition to that, there's a third customer building their own CPUs for their overall high-performance compute products, where we've won some designs and expect to have that as a 2027 contributor.
Across the spectrum, largely what we shared with people at the May Investor Day, the growth vectors associated with this intersection of high-performance compute and advanced packaging or chiplets, and our ability to really efficiently manufacture, both in our existing footprint and then the exciting capacity and output unlocked by the Farmers Branch investments as we move into 2027. We put all those pieces together. We're looking forward to a strong end to 2026 and a strong year in 2027.
I'm just wondering about the recurring revenue aspect of your business, and as you're getting these new wins, how should we think about the recurring as a percentage of overall revenue?
Yeah.
How does that grow?
Yeah. Probe cards, they are a consumable, but when I got into the business about 15 years ago, they were consumable because they actually wore out. Essentially, there's cleaning cycles that wear these probes down, and they have to be replaced. That's still true in markets like automotive and industrial, where product life cycles are very long. It's no longer true in areas of the market like high-performance compute and mobile, where product life cycles are essentially less than a year. What makes probe cards still a consumable is that they're unique to each customer chip design. As our customers release a new product, a new mask set, as a customer like NVIDIA goes from Blackwell to Rubin to Rubin Ultra, those are each new probe cards, even though they use the same capital equipment and test equipment.
Because essentially the probe position and the way the connections and netlist works, they're different for each chip design. That's what drives the consumable nature now. Less recurring because of a wear out cycle, and more consumable because of the rapid increase in design change velocity in the industry today.
One more question on the capacity side, finally here. A lot of the feedback we get from investors is that you are being very cautious on ramping up the capacity, which I think I totally understand. You want to be cautious, and you do not want to sacrifice a gross margin. I am just curious if demand is going to the roof, and then how fast can you ramp up your capacity at Farmers Branch?
Yeah. Cautious is one way to put it. I think if I were to self-assess us, we are conservative as a management team, but I think we are approaching this thoughtfully. When you look at the purchase of the Farmers Branch site a little over one year ago, we purchased a site that had a ready-to-go clean room, had expansion potential. It is close to a 300,000 sq ft site. So the ability to have the infrastructure and factory shells, if you will, to expand further. That does not mean we are going to turn on all the capacity at once because, again, we want to make sure that capacity is matched to demand, so that we are not facing large utilization and therefore gross margin headwind. But right now, given the demand environment, we are trying to bring this site on-site as fast as we can.
Now, again, we are doing it in a thoughtful, measured way, where we are not putting our customer relationships or our customers' production at risk, doing careful, thoughtful, comprehensive qualifications of the core technology and making sure that we are properly qualified before we ship out of the site. Start to ship the initial qualification units in the fourth quarter, and again, ramping through 2027 to deliver a substantial increase, essentially 2x the existing capacity as we get through the back part of 2027.
If, say, demand is there, like $1.6 billion level ahead of your schedule, just assumably getting to that level early 2028 or late 2027, are you able to support that level?
Yeah. If you look at the $1.6 billion level in 2030, the assumptions associated with that, I think one that people have challenged us on being too conservative is really the market forecast to serve market growth. We deliberately chose to use an external market forecast, so everybody had visibility to those assumptions. Arguably, that high single-digit market growth rate was overly conservative. We're putting capacity again in place as fast as we can to meet the demand we have associated with all the opportunities that we've laid out for people today. I think the expandability, the flexibility of that site to add capacity in tranches as we see, get good visibility to the demand.
I think we're in a position if the market and our ability to then execute on these initiatives drives the demand side of the equation, we will be able to reach the target model faster than 2030. It really depends on the demand side of the equation.
Okay, that's good to hear. With that, we are out of time.
All right.
Thanks, Mike, for coming to our conference.
Thanks again for having us.