... after this, Michael has a few thoughts and then open up to the audience to ask their questions. Welcome, Sanjay.
Thank you. Thank you.
First question on the test equipment forecast, and hoping to time you guys no longer forecast full year. But what would be really helpful for investors if you can walk us through what has changed from January this year, when you guys did provide a forecast and had to amend the day? So broad scope, what has changed so far this year in those decisions?
Sure. So yes, that's true that we have stopped providing our TAM forecast update publicly. But I will share a little bit of color on what's going on in the market from our perspective. So when I think about it, I'll cover. First, I'll talk about compute. You know, our compute, the compute segment has consistently grown and is very strong, really driven by AI. And from our perspective, what's driving our growth, and we'll have a strong second half, but what's driving our growth is really tied to AI, driven by custom ASICs at vertically integrated producers or VIPs, also known as hyperscalers. And you know, we talked about on our call that, you know, these shipments we expect to be lumpy 'cause they're really tied to project ramps.
You know, we have some projects toggling Q3, Q4, and Q4 and Q1, but overall, from a market perspective, compute continues to be the dominant market segment in semiconductor test. Moving to mobility. Mobility was really first half dominated in the market, tied to, I would say, supply chain transitions. In our call, we talked about this being transitory, and specifically that it wasn't end market driven, but really tied to supply chain shifts versus complexity or incremental volume in the smartphone market. Moving along to auto and industrial. I think the best way I'd describe it is, it's still in a lull.
While we haven't seen it weaken, we haven't seen it increase as well, so it's operating at a low level or, as I called it, a lull. Moving to memory. Memory is very exciting. You know, in 2024, there was a very strong memory TAM, and we called for a decline in January of the memory TAM in 2025. We still view that as reasonable. The reason is that there's a significant amount of HBM testers shipped in the marketplace that need to be digested. And in Q2, as we anticipated, there was a decline in shipments in Q2, but second half appears to be quite strong, again, driven by HBM memory testers.
Flash continues to be weak, really tied to the smartphone market and just looking forward, you know, I'm not necessarily going to talk about 2026, or maybe I'll have a comment at the end about 2026, but you know, looking forward from a market perspective, we really see, as we talked about in our Analyst Day, three key themes tied to enabling a continuing growing market. First is AI. It's going to continue to drive compute and memory. The second is electrification, and third is verticalization. We have a strong conviction that those are going to continue to drive the end markets up and to the right.
I would say about 2026, you know, if I reflect on our view of the second half of 2025, and we've only guided Q3, but in general terms, you know, our business is really driven by high performance compute and memory, or really AI. And when I think about looking forward in 2026 or even beyond, we see that continuing, and our business is no longer tied to mobile end market launches, which are typically in the second half of the year. And so if mobile comes back and, you know, there's tailwinds and headwinds, the tailwinds are obviously two nanometer Gate-All-A round, and that complexity of products being launched, we'll be ready to ship.
But our plans are such that, you know, our revenue is diversified now, that it really is tied to AI.
That's very helpful. Sanjay, let's start with the compute market first. It was great to hear on your last earnings call that Teradyne has been invited to the GPU game. Can you share Teradyne's strategy to break in, and how does Teradyne differentiate from the incumbent?
Sure. So first, let me reiterate what we said on the call. We haven't won anything yet. We don't have any orders. We haven't been qualified, what have you. So just clearly, I want to just comment, and I don't want to talk about any specific customer. But when we think about kind of merchant GPU customers. Yes, it's true. As we noted on the call, we have been invited to qualify and go through the qualification process at a merchant silicon provider. And you measure that process in quarters, and so the outcome of that has yet to be determined, but think about that if we're successful or not, as a 2026 outcome.
And so our UltraFLEX, our strategy, getting back to the strategy, is that we design the UltraFLEXp lus product tester specifically for high-performance compute, kind of bottoms up for this, as opposed to an evolutionary increment to an existing tester. So it was designed for this particular market, which we feel very good about. And, historically when we've been given a level playing field and you know, we continue, and that's why when we think about, you know, our success in VIPs, that we have a level playing field, and we usually have a very good success. And just given the opportunity, it provides us with that opportunity to demonstrate what our capability is.
Really, the last point I'll make, and we made this on the earnings call, is that we were given this opportunity. It was really tied to very large companies. You know, they want to have security of supply. They want to make sure they're not single-threaded from a supply chain perspective, and they want to have the ability to scale. You know, test is a derivative market, and when the wafers come out, depending on the yields or the product quality, to get more, you know, known good die out, you have to throw more, sometimes you have to throw more test seconds at it, more testers. I think companies are really starting to believe that and are executing against that strategy.
Sanjay, the pushback I get from investors is that the test equipment market is very sticky. You have lifetime of 10 to 11 years for these testers. So for Teradyne, and you guys have done historically very well in entering a market and taking share, and I would say LPDDR was one example, HBM has been another example. So when you put your R&D dollars to work, you know, and you create products that are differentiated, but so second sourcing, you know, might not be the most compelling reason if the competitor decides to, you know, lower pricing or something like that.
So, the question we get is, like, is there a big technology reason or inflection, whether it's, let's call it Rubin or whatever platform is being designed for GPU makers, that requires an entry point for you to come and differentiate substantially?
Yeah. So, so clearly, cost of test and quality are clear objectives of our customers. And when we come in and supply chain service, there's many factors that really drive a very good relationship. And at Teradyne, we pride ourselves on really demonstrating our capability and making our customers successful. So when we enter a customer situation, and you use the example of memory, where you know, for those that don't know, in 2004, Teradyne entered the memory space through acquisition. We had 4% market share. You know, roughly, you know, round numbers, we grew that to about 40%, 40% plus market share. And we did that by solving complex problems for our customers and driving the economics and the quality and the service that they really enjoyed and were passionate and rewarded us financially for it.
And so we believe that recipe is going to continue, going forward.
Right. And on the VIP side, and I have to say, you guys love your acronyms at Teradyne, on the vertically integrated ASIC suppliers, can you walk us through? You have a pole position on the networking side, but is the demand for compute a lot stronger than the networking piece? And just kind of help us understand the two pieces within compute.
Yeah. So, we have a very strong position in networking, and we have a growing revenue in our high-performance compute, really tied to the vertically integrated producers of the VIPs. And you know, not all of the VIPs are created equal. These are complex chips and, you know, my prior role at , you know. I spent 15 years at Qualcomm. These are complicated ASICs, and you need very deep technical teams that really understand the ecosystem and how to hit both the time to market and the cost performance of what you're trying to do. But, there have been successes with VIPs, and there have been, unfortunately, failures at some of these VIPs. We've been fortunate enough to be utilized in some of the more successful VIPs.
And our strategy, and I talked about it, you know, when we have a level playing field, we believe we can compete and win our fair share, and our target is to win 50% of these new participants, these VIPs, in the market. And so far, we've been hitting that target, and we track things from design wins, to tester loading, to, and obviously, you know, the amount of testers we ship, you know, relative to the competition, and we're roughly in the 50% level, and we feel pretty good about our position. The last thing I'd say about the VIPs is that, you know, they are growing in experience, and some with success, some not having so much success, but they're coming back really strong.
As they become more technically capable, they're starting to behave a little bit more and more like a merchant silicon provider, where in the ecosystem, typically test is defined by the design team that specifies the test, that will specify the tester. As the engineering capability and their experience gains, those decisions are starting to be made more and more by the designer of the chip and the VIPs. We feel that this opens up the level playing field, where we will continue to compete and feel very good about our position.
All right. Let's switch gears and talk about Universal Robots. Amazon could be a big opportunity into next year, you guys talked about on your call. Can you touch on why there is a lull near term? I think in the September quarter, you're expecting industrial automation to grow the slowest among the three segments, and you guys have been kind of working towards, you know, bigger customers in the last few years, and where are we in terms of that big adoption with the big warehouses?
Sure, so you mentioned the particular customer, so again, I don't want to talk about a particular customer. Now, I'll sit here and say: We're not preventing our customers to share their experience with our solutions, which I think is very clear in videos, publications, and pictures that I'm sure many in the audience have seen, but my comments won't discuss anything in particular, but I'll share a little bit about our strategy in robotics, and I think you hit what is our strategy, just to recalibrate, and we pivoted maybe a year and a half ago. One is to really drive large accounts. The second is to drive OEMs.
So OEMs are. They create a full solution. They purchase robotic arms or AMRs, and they go to market with a full solution and help enable customers' applications. And then third is new product introduction to really address different market segments. And on the large accounts, we have specifically put this focus in place, and really why? Large customers have technical teams that have the ability to ingest the technology and solve problems. They can then really scale, so the repeat business is at a very high level. Same thing with OEMs. They solve hard problems, we sell robots, they fan it out. So that's why we kind of made that strategic pivot. And you know, our large accounts, we're kind of talking about when the volumes would really take off.
You should think about it like a second half of 2026, really ramp period for you know kind of key large accounts, and then the industrial market, I think you noted, is in a very weak, consistent position. Capital budgets are tight, and you know even though the ROIs are great for automation, you know CFOs like me in companies are saying, "Gee, we don't have the cash flow to actually make the capital investment. We're going to continue on with the labor and go," but you know we're working through it. We're starting to see our strategic pivot take hold as some of the publications and videos you've seen.
And in terms of the profitability of this segment, you know, it still comes up in conversations. Are we still around break-even level, and are there significant investments required to chase that opportunity in the second half of next year?
Yeah. So, our expectation is we are not going to break even. So in the beginning of the year, we restructured. We took about 15% of our operating expense out, but we really did it thoughtfully and strategically. You know, when we looked at kind of like the overlap of the customer base, we had an individual product line of MiR and a product line of UR, that were servicing a global market in all these different territories, and many of them were calling upon the same customer. And so what we did was we combined the teams and changed our sales solution to not only be product focused, but platform focused, to enable kind of a more comprehensive dialogue with customers. That brought some efficiency, as well as some other things.
And we had an initial design point similar to the 2024 revenue levels. Unfortunately, you know, the market continues to be weak, and we don't believe that we are going to be at break even this year.
Great. Let's talk about the mobility segment. I know you guys have gone through a bit of a prolonged correction in the mobility market. During COVID, you saw very strong demand in mobility, and then your end customer started to repurpose some of the testers towards compute, away from mobility, but it was good to hear that on the last earnings call that you're starting to see more new system demand versus upgrades, so maybe some green shoots in the mobility market. The question is really around, you know, Apple announced generative AI search yesterday, and Gemini use and all that could drive excitement around AI and smartphones next year, and then there's two nanometer, and there's some interesting changes happening on the substrate level with Chip-on-W afers.
And so the big question is really about at what rate can the mobility demand recover? I remember the high point was maybe eight hundred million type sales with-
Mm-hmm.
with that customer a few years back. Can you walk us through how you're thinking about the recovery in the mobility market?
Sure. Again, not talking about any particular customer. You know what? I would say that I am encouraged by AI at the edge. And our hope is that this is going to drive a big upgrade cycle. But if you're like me and a cheap accountant at heart, you really don't spend a lot of money. You know, this is my first, this is actually the first smartphone I have purchased. You know, in my prior company, we helped enable them, and customers used to give them to us. And so but, you know, we do see that complexity, first of all, the market need for AI at the edge is really driving the requirement for the complexity of two nanometer and Gate-A ll-A round. So that's the tailwind.
The question is: What's the volume going to be on those? And so it's an uncertain outcome at this point. But as I noted earlier, you know, our business right now in the second half, which we expect to continue into 2026 , is really going to be AI-driven by compute and memory. Going back to the point you said, which is, I think, a really good one, talking to kind of what happened tail end of 2023 and 2024 , and what happened is that as the smartphone market really contracted, there was a bunch of underutilized testers at the OSATs or in the ecosystem. And what companies did was they repurposed those testers to not test mobile, because the demand wasn't there, but to test high-performance compute, so chips.
So they needed upgrades, they needed what have you. They weren't driving new tester requirements. We think that's pretty much behind us in the first half of 2025, as our view of utilization rates are much higher. And so going forward, as incremental demand ties to mobile, that should drive new testers, as well as incremental demand for compute should drive new testers. So that is behind us, and then as the incremental demand shows up in compute and mobile, new testers should be the tailwind. And we have a good belief that this complexity will be a tailwind. It's to what extent and will be measured by or counterbalanced by the volume. And you know, every year, customers get efficient, you know, 5%-10% efficiency that they drive in their own installed base of testers.
Will December, January, be the timeline where you'll see some early signals from your customers around the rebound?
From a mobile perspective, it's usually the tail end of Q1, Q2. Usually the market launches, you know, the significant shipments are predominantly in the back half of the year. So kind of March, April is when we'll really have good visibility into that.
Great. Let me pause here and see if there are questions in the audience. If you have a question, please raise your hand. We'll just keep going. Sanjay, for the September quarter outlook, you're expecting memory to come back to normal levels. It was down in the last quarter before a ramp in fourth quarter. And I remember you guys talking about the productivity of your tester really being kind of maybe limiting the TAM growth for the market. So help us understand what's going on in the memory market as we move into HBM4. Does the test intensity go higher, and how should we be thinking about the memory market outlook?
Yeah. So I think, first of all, the memory market is really driven by DRAM, and specifically HBM DRAM. You know, flash is really going to be driven by the smartphone market, and right now that's still very moderated. And so what we see right now is tremendous tailwinds in HBM, really tied to the AI boom and compute boom. And, you know, we feel our position is improving. You know, we-- I talked earlier about we had 4% memory market share, and that grew to 40%. The large part of the market that we didn't participate in was DRAM wafer sort, and we didn't participate in that because, frankly, it was a commoditized market with low profit pools.
And as soon as HBM kicked in, you know, with DRAM technology, and we pivoted our engineers 'cause it was very complex. And, you know, we believe the profit pools, and, you know, customers would reward us with, you know, helping to solve those problems. We pivoted our engineers, and, you know, in 2024, we had very good business with a leader, one of the OEMs, one of the leaders in the market. And in 2025, you know, in first quarter and second quarter, we obtained incremental design wins. So, you know, we're with a couple of the key customers. And so we view our position in the market as a tailwind, which we expect to gain share.
From a market perspective, I really believe that overall, we are in a very strong cycle, again, tied to AI.
Okay. And then you recently launched a memory tester, Magnum 7H, for HBM, and it's interesting. It talked about integrating HBM with GPU and accelerators. Can you just talk about what the part is, and how is it different from the incumbent?
Yeah. So the Magnum 7 tester that we've announced is a very unique, high-performing tester. It significantly improves or outperforms from a digital channel perspective, parallelism, and power. And an example of that is, we were qualifying it at an OEM, and they had a 60% throughput improvement, which really helped enable our design win. The second thing, and as you noted, it is a very unique tester that is unique because it can test both logic and memory. The other uniqueness in it is that it has a lot of headroom, so it can test HBM3E, and is forward compatible to HBM4. And as many of you know, as interface speeds change in memory, you need a new tester, typically.
And so when you can go from HBM3E to HBM4, the customer gets tremendous economic benefit, because the tester becomes both fungible of what it can test, but also they don't have to buy a new tester for HBM4, which makes it uniquely performance. And so, maybe the last point I'd say is that we have a significantly high-performing reference probe card that helps enable the performance, and it's and it's currently being licensed by our competitor. And so, we feel very good about our last point, maybe is I'll say it again, we believe in 2025, we are going to be gaining share in memory.
Great. Let's talk about financials. You have 59%-60% gross margins in your 2028 target model, but you're already not too far from that range at a $3 billion run rate. Any upside to gross margins, and as we think about, you know, potential recovery in mobility next year, which historically has kind of dragged your margins lower, should we be thinking about mix effects?
Yeah, I think, you know, our business model really calls for 59%-60% gross margins, and why? Our strategy is simple. We want to solve complex problems, and we want to be rewarded to solving those complex problems, and that's where we make our investment bets to hit the market window on a timely basis. And so in our portfolio of companies, we target 59%-60%. Now, in any given quarter or even year, those could fluctuate. So I'll give you the example, in 2023, we spent a lot of time in our supply chain resiliency. We de-risked our factory production from geopolitical risk, as well as hundreds of thousands of components go into testers.
So we wanted to make sure in high-risk countries, where we had a deep supply chain, that we had alternative country, either suppliers or parallel factories for the components to come up. That was money, and that was reflected in our gross margins, which we had a little bit lower gross margins that year, and in any given quarter, we could have higher gross margins, like in Q1, we were 60.6%. You know, in the first half of the year, our gross margins are 59%, but in any given quarter, the configurations of the testers we sell vary, so it could fluctuate. I would say that looking forward as a $3 billion run rate, your number, not mine, in 2025, we have tailwinds and we have headwinds.
Tailwinds are that we have, as revenue scales, we will have the ability to amortize our operations cost, and which should be a tailwind with higher volumes to lower gross margins. Now, we outsource about 80%-85% of our manufacturing, and so that shim layer of fixed cost that we have in operations has diminishing returns of benefit. It has tremendous protection on the downside, as we went from $3.7 billion to about $2.7 billion. You know, those factories, when they were underutilized, are outsourced contract manufacturers, factories. So we have a predominantly variable manufacturing strategy, which does have a degree of gross margin upside as the fixed costs get amortized, but that kind of is diminishing as you get to really significant scale.
I would say that we continue to plan to a 59-60% gross margin model, and expect... but any given quarter or even year, it might be slightly higher or slightly below, just given the customer mix, product mix circumstances.
All right. And then lastly, on the capital allocation, particularly M&A, you haven't done any, you know, major or tuck-in type acquisitions. Recently, you've done some JVs and I think you bought something on the photonics side, tester. But as we think about the automation market, you know, again, potentially taking off next year, will M&A kind of come back into the picture?
Yeah, I think, you know, what's our M&A strategy? It's clearly our first priority is to seek and acquire companies that are going to help us technically go to market to drive earnings power that's incremental versus our share buybacks. Then we prioritize our share buybacks, and we have a dividend. So we think about it like a balanced approach. Having said that, we have, you know, at the end of May, we closed on silicon photonics. You know, we really believe that co-packaged optics and silicon photonics is really going to is in a technical inflection point, and we really believe in acquiring assets like that.
I do believe that there is a key AI demand cycle that is occurring, and we're always on the lookout to add capability and content that will help solve customer problems that fits into our business model. You know, we're consistently looking, and if something comes up, we'll be happy to share it.
Great. Sanjay, thank you for coming to the Citi Conference.
All right, well, thanks for having me.