Teradyne - Earnings Call - Q2 2025

Executive Summary

• Q2 2025 revenue was $651.8M and non-GAAP EPS was $0.57, both above the midpoint of guidance; GAAP EPS was $0.49. Semiconductor Test (SoC for AI compute) drove outperformance, while Memory was lower sequentially and YoY due to shipment timing.
• Versus Wall Street consensus, non-GAAP EPS modestly beat ($0.57 vs $0.54*) and revenue was essentially in-line ($651.8M vs $651.8M*). Strength in AI compute was the key driver of the beat; Robotics remained weak and below breakeven.
• Q3 2025 guidance was raised sequentially: revenue $710–$770M, non-GAAP EPS $0.69–$0.87, GM 56.5–57.5%, with tax rate stepping up to 16.3% due to new legislation and a YTD catch-up; management’s tone turned more constructive on second-half demand driven by AI in both SoC and Memory.
• Stock-relevant catalysts: broadening AI compute opportunities (including potential merchant GPU dual-sourcing), new HBM4 test insertion points, and improving utilization that is shifting orders from upgrades to new systems; watch Q3/Q4 timing risk as ramps straddle quarters.

What Went Well and What Went Wrong

What Went Well

• “Semiconductor Test drove better than expected results in Q2,” led by SoC for AI compute; visibility improved with strengthening demand in compute, networking, and memory.
• SoC compute traction broadened, with significant UltraFLEXplus system orders and confidence that AI will be “the majority” of Semi Test revenue in 2H; management emphasized scalability and lower cost of test as differentiators.
• IST more than doubled YoY (HDD and mobile) and Product Test was up 7% YoY; Quantifi Photonics acquisition accelerates silicon photonics test leadership for AI compute.

What Went Wrong

• Memory revenue ($61M) was “considerably lower” sequentially and YoY, driven by timing of deliveries; snapback expected in 2H, heavily Q4.
• Robotics ($75M) stayed weak YoY and will not break even in 2025; macro headwinds persist despite 9% QoQ growth post reorganization and a large plan-of-record win that will impact 2026, not 2025.
• Tax rate will rise to 16.3% in Q3 on GAAP and non-GAAP due to new legislation and YTD catch-up, tempering EPS conversion despite higher volume.

Transcript

Speaker 3

Welcome to the Q2 2025 Teradyne earnings conference call. At this time, all participants are in listen-only mode. A question-and-answer session will follow the formal presentation. If anyone should require operator assistance during the call, please press star then zero on your telephone keypad. As a reminder, this conference is being recorded. I would now like to turn the conference over to your host, Traci Tsuchiguchi. Please go ahead.

Speaker 1

Thank you, Operator. Good morning, everyone, and welcome to our discussion of Teradyne's most recent financial results. I'm joined this morning by our CEO, Greg Smith, and our CFO, Sanjay Mehta. Following our opening remarks, we'll provide details on our performance for the second quarter of 2025 and our outlook for the third quarter of 2025. The press release containing our second quarter results was issued last evening. The slides, as well as a copy of this earnings script, are on the investor page of the Teradyne website. Replays of this call will be available via the same page after the call ends. The matters that we discuss today will include forward-looking statements that involve risks that could cause Teradyne's results to differ materially from management's current expectations. We caution listeners not to place undue reliance on any forward-looking statements included in this presentation.

We encourage you to review the Safe Harbor statement contained in the slides accompanying this presentation, as well as the risk factors described in our annual report on Form 10-K for the fiscal year ending December 31, 2024, on file with the SEC. Additionally, these forward-looking statements are made only as of today, and we do not undertake any obligation to update forward-looking statements to reflect subsequent events or circumstances except to the extent required by law. During today's call, we will refer to non-GAAP financial measures. We have posted additional information concerning these non-GAAP financial measures, including reconciliation to the most directly comparable GAAP financial measures, where available on the investor page of our website. Looking ahead between now and our next earnings call, Teradyne expects to participate in technology or industrial-focused investor conferences hosted by Evercore, KeyBank, Citigroup, and Goldman Sachs.

Our quiet period will begin at the close of business on September 19, 2025. Following Greg and Sanjay's comment this morning, we'll open up the call for questions. This call is scheduled for one hour. Greg?

Speaker 4

Thanks, Traci. Good morning, everyone, and thanks for joining us. Today, I'll discuss our second quarter results and provide an update on what we're seeing across our businesses. Sanjay will then provide more detail on our second quarter results and third quarter guidance. Through the second quarter, end market trends noted in prior quarters were generally consistent with a strengthening second half. Strength in AI compute is more than offsetting lower demand in auto and industrial end markets. Pockets of improvement in mobile are driven more by customer-specific dynamics than an uptick in the end market demand. Visibility is starting to improve. In terms of capacity utilization, we believe that we have turned the corner towards more new system sales rather than selling upgrades of existing idle mobile capacity for new compute and mobile applications.

Demand is strengthening in AI compute, and we're seeing a broadening of opportunities where Teradyne, and especially the UltraFLEXplus, is getting strong consideration in areas where we have not historically had a seat at the table. While new program ramps and new test insertions can drive a lumpy order pattern, we are optimistic about the opportunities on our horizon. In the second half of 2025, we expect AI compute to be the dominant driver of our SoC business. The long-term themes we've discussed—AI, verticalization, and electrification—remain intact, with AI and verticalization emerging as the primary growth drivers in the near term. Our Q2 results reflect the evolving composition of our business. In the past, the typical seasonality in our revenue was heavily driven by consumer mobile demand. This has now been superseded by the waves of demand driven by specific customer program ramps and AI compute.

These have no correlation to consumer holiday buying patterns. In the second quarter, we delivered revenue, gross margins, and earnings per share above the midpoint of our guidance ranges. Semitest, specifically SoC for AI compute, drove results above our expectations. End demand trends in mobile persist, but we saw pockets of customer-specific strength in RF and mobile power in the quarter. In the industrial and automotive end markets, demand has stabilized at a low level. As expected in Q2, memory revenue was lower quarter on quarter due to the timing of shipments and is expected to snap back in the second half. In the quarter, our memory business unit secured an important HBM4 post-stack singulated die win. HBM suppliers are adding test coverage to improve device quality. Some suppliers are adding a test insertion for HBM singulated stacks.

While this new insertion is not yet pervasive across the broader industry, we believe that it is an important growth driver for the memory TAM in the future. This win builds on momentum from the HBM4 post-stack wafer test win in Q1 for our memory business unit. As we discussed in our analyst day, there are four elements to growth in our IST business: accelerated bid growth in HDD, share growth and recovery in the mobile SLT market, emerging SLT for AI accelerators, and solid-state disk drives. In Q2, IST revenue more than doubled compared to the same period last year, mainly driven by HDD and mobile. All of the businesses within our product test group delivered second-quarter results generally in line with our expectations and up year-on-year.

In the quarter, we closed the acquisition of Quantify Photonics, accelerating an important element of our strategy to gain share in AI compute by establishing a leadership position in silicon photonics test. In robotics, recall that we executed a structural reorganization that consolidated the customer-facing sales, marketing, and service organizations of UR and MIR in the first quarter of 2025. In Q2, this new organization delivered 9% quarter-on-quarter growth despite persistent difficult market conditions, and we continue to optimize our OPEX envelope to respond. In the second quarter, as part of our pivot to large customers, we secured a plan of record decision from a large customer. This is not expected to have a material impact on robotics revenue in 2025, but is expected to be a significant growth driver later in 2026.

In support of this opportunity and others, the team plans to open a manufacturing operation in the United States to best serve customers in this region. Moving on to Q3. As we progress through the third quarter, we are gaining confidence in AI compute-related revenue inflecting in the second half of the year, driven by both SoC and memory. We are less certain of the quarterly timing of shipments between Q3 and Q4, and then between Q4 and Q1 due to customer schedules. That said, we expect the relative size of AI compute in our SoC and memory business to represent the majority of our semitest revenue in the second half. Our expectations for mobile are modest in the third quarter and the second half generally, expecting that the bulk of the demand we'd see for the year has been satisfied in the first half.

Growth in the mobile segment is coupled to the ramp of 2-nanometer gate-all-around and the expectation of more compelling AI applications in the generation of smartphones coming in the back half of 2026. In the auto and industrial end markets, our end customers remain cautious about significant capacity adds, but we do not expect test equipment order patterns to deteriorate further. There are areas within this end market that are showing strength, like the power semiconductors for data center buildouts, and we believe that the long-term trend towards electrification will drive growth beyond 2025. Overall, we feel good about where we're headed in the third quarter and the second half of the year. We are significantly more confident than we were 90 days ago. Demand trends in AI compute have strengthened and forecasts are materializing into orders. Utilization rates have improved considerably, leading to an increase in UltraFLEXplus system orders.

With the work that we have done to increase the resilience of our supply chain and dual-source our manufacturing, we are in a position to effectively scale volume with increased demand and provide timely delivery of our testers to fast-moving customers. I want to emphasize that we are opening these new opportunities because of the scalability of our newest systems, our capabilities in silicon photonics, our parallelism, and higher throughput that lowers the cost of test for our customers. With that, I'll turn the call over to Sanjay.

Speaker 0

Thank you, Greg. Good morning, everyone. Today, I'll cover the financial summary of Q2 and provide our Q3 outlook. Now to Q2. Second-quarter sales were $652 million, and non-GAAP EPS was $0.57, both above the midpoint of our guidance ranges. Non-GAAP gross margins were 57.3%, consistent with our guidance range. Non-GAAP operating expenses were $275 million, up year-over-year, as we have increased our R&D investments and targeted opportunities to drive longer-term growth. OPEX came in flattish sequentially as we continue to practice discipline spending controls. Non-GAAP operating profit was 15.1%. Turning to our revenue breakdown in Q2. Semitest revenue for the quarter was $492 million, with SoC revenue contributing $397 million, memory $61 million, and IST $34 million. Strength in SoC was driven by mobile upgrades. AI compute growth exceeded our plan.

Expected memory revenue was considerably lower sequentially and year-over-year due to the timing of customer deliveries, which is expected to be back halfway to this year. In the first half of this year, customers have been digesting the HBM test equipment delivered in 2024. We expect DRAM to dominate the memory mix in 2025, just as it did in 2024. IST revenue of $34 million was up both sequentially and year-over-year, driven by mobile SLT and HDD testers. In product test, Q2 revenue was $85 million, up 7% year-over-year, with all business units within product test up year-over-year. As Greg noted, we closed the acquisition of Quantify Photonics in the quarter as its results are included in this segment. Now to robotics. Revenue was $75 million a quarter over quarter, but down year-over-year. In the quarter, UR contributed $63 million and MIR contributed $12 million.

While the long-term drivers of AI and onshoring and advanced robotics remain intact, near-term macro factors continue to be a headwind. Our second-quarter operating results were better than our first quarter, and we expect the second half of the year to be better than the first half. Despite this, due to the weak end market, we had lower volumes yielding a lower gross margin. We expect the weak market to persist and do not expect robotics to break even this year. Some other financial information in Q2. We had one customer that directly or indirectly drove more than 10% of our revenue in the second quarter. The tax rate, excluding discrete items for the quarter, was 13.5% on a GAAP and non-GAAP basis. At a company level, our free cash flow was $132 million, primarily driven by improvements in network and capital in the quarter.

We repurchased $117 million of shares in the quarter and paid $19 million in dividends. In the first half, we returned $316 million, or 138% of our free cash flow, through dividends and buybacks to shareholders. We ended the quarter with $489 million in cash and marketable securities. Now turning to Q3. Q3 sales are expected to be between $710 million and $770 million. Third-quarter gross margins are expected to be at 56.5%-57.5%. Q3 OPEX is expected to run at 36.5%-38.5% of third-quarter sales. The non-GAAP operating profit rate at the midpoint of our third-quarter guidance is 19.5%. The Q3 tax rate is expected to be 16.3% on a GAAP and non-GAAP basis. The increase in rate is driven by the impact of the new tax legislation, which goes into effect in Q3. Hence, a year-to-date catch-up is included in Q3. A full-year tax rate is expected to be 14.5%.

Q3 non-GAAP EPS is expected to be in a range of $0.69-$0.87 on 158 million diluted shares. GAAP EPS is expected to be in the range of $0.62-$0.80. In terms of guidance, we will continue to guide one quarter at a time. Summing up, we delivered sales and earnings above the midpoint of our guide. We feel more confident in our near-term outlook than we did 90 days ago. Looking ahead, our visibility is improving, driven by demand and artificial intelligence in both SoC and memory. Second half of the year is shaping up to be stronger than the first half, as we expected earlier in the year. Our multi-year investments in testing artificial intelligence are beginning to deliver new opportunities and accelerating top-line growth. We are confident in the long-term growth drivers of AI, electrification, and verticalization trends that will drive our business in the coming years.

With that, I'll turn the call back to the operator to open the lineup for questions. Operator?

Speaker 3

Thank you, Zoe. At this time, we will be conducting a question-and-answer session. If you would like to ask a question, please press Star and then 1. A confirmation tone will indicate your line is in the question queue. You may press Star and then 2 if you would like to remove your question from the queue. Again, if you would like to ask a question, please press Star and then 1 now. The first question we have comes from CJ Muse of Council Fitzgerald. Please go ahead.

Yeah, good morning. Thank you for taking the question. I guess first question, your tone on your outlook clearly much more positive versus three months ago. Curious, is that a reflection of a pickup in business that you're seeing today in the second half? Is there any way to sort of frame whether growth continues into Q4? Perhaps more importantly, is it also related to new design wins that give you confidence beyond 2025?

Speaker 4

Hi, CJ. It's Greg. Yeah, so it is. The confidence is definitely related to an uptick in demand, primarily in AI compute, both in SoC and in memory. We expect that to—it's a—the major driver for our Q3 guidance, and it's also the reason for our optimism about the full year. As I said earlier in my remarks, a lot of the programs that were designed into, the ramps are occurring late in the year. The split of demand between Q4 and Q1 is a real X factor. We know the demand's there. We just do not know the exact timing of it. Your other question in terms of, for the second half of this year, is this related to new wins or existing programs?

I would say that many of the things that we're seeing in the second half of 2025 are actually wins that have occurred mainly in 2024. In memory and in SoC, there are customer wins that we achieved either in 2024 or early in 2025 that are ramping. There are more opportunities in the funnel that we're hoping to close before the end of the year, but we would expect those to have a more positive impact in 2026 than in 2025.

Very helpful. I guess the follow-up to that, as you look to 2026, is there a framework for thinking about AI contributions to you and segmenting between custom silicon, networking, HBM, and other?

Looking out into 2026, we haven't done that analysis in terms of how that demand is going to segment. I think in general, we would expect the same kind of a split in the memory market that we're seeing now, with kind of between 85-95% of the memory market being driven by DRAM and most of that being driven by cloud compute AI applications. In the SoC market, looking into 2026, we've had a lot of strength this year, both in VIP compute and also in the networking part of compute. We would expect that that's going to continue into 2026. We're hopeful that we're going to be able to add in additional compute revenue from the merchant suppliers, but that is an opportunity, not something that we have in the bank right now.

Thank you.

Speaker 3

The next question we have comes from Timothy Ouchi of UBS. Please go ahead.

Thanks a lot. Greg, for the robotics business, I think you said you talked about a plan of record opportunity, and you're opening up a manufacturing facility in the U.S., which, given how bad the business has been, you must feel pretty good about how big that opportunity can be. I believe this customer's put some stuff on their website. Can you just help us size how big that can be and sort of when it starts to impact your business? I think you said it's going to be a "significant growth driver" for next year. Can you just help us, give us any color on that?

Speaker 4

Yeah. From the statement that I made in terms of establishing manufacturing in the U.S., for us, that opportunity is at first primarily driven in the U.S., North America more generally. It has the opportunity to spread into other regions as well, but we're able to serve the European region from our Denmark manufacturing quite well. I do want to emphasize that one of the things that is a requirement for many of the larger customer opportunities that we're pursuing, including the one that we won, is supply chain resilience, that these manufacturers want to see that you have multiple production facilities so that they're going to be able to get their material no matter what. We believe that having geographic diversity in our operations is kind of a key element of this large customer strategy in general.

Now, specifically around this one opportunity, the solutions are largely developed, but there's a lot of kind of new product introduction work that has to happen in 2025. The demand associated with this opportunity in 2025 is going to be quite—it's not going to be—it's not going to have a material effect on our robotics results. When we get into 2026, I can't give you an exact number, but it will represent a sizable fraction of the UR business. UR is the majority of the robotics business. It's definitely a needle mover in 2026, but I can't give you an exact number.

Okay. Awesome. Thanks. And then just on the—I know you're not that excited about the opportunities in mobile this year, but it does seem that your big mobile customer, they're moving to a new package next year. And what is your assessment sort of on what impact that's going to have on test times? Obviously, you have—and two, you have transistor density going up, and test times seem like they're going to also go up. I know you saw the modem test orders this year, but does it make you optimistic about your mobile market and your large customer in particular next year? I know you haven't been that optimistic lately, but I wonder if you can comment on that.

Yeah. So, we've talked about the drivers for demand in the mobile space, that there's unit volume, that there's device complexity, and yield, the yield and utilization, sort of the efficiency of the process. On complexity, 2-nanometer gate-all-around is definitely going to enable significant increases in transistor count kind of complexity. The parts are going to be more complex. The other thing that drives complexity is the packaging methodology and things like the memory bandwidth. If, for instance, there are wider buses between the processor and memory, that means that you'll consume more tester resources per device than you would if the buses were more narrow. In addition to just the transistor complexity, if the device needs more tester resources per device, you can test fewer devices at the same time. That would be another way that it would—that would come through in our model as an increase in complexity.

Going into 2026, we think that there is a potential inflection in transistor complexity. We also believe that there's a potential inflection around memory technologies and packaging. That makes us optimistic that mobile will be better next year. The thing is that mobile is going to be a smaller part of our overall mix in 2026 because of the strength of compute. It's like, if mobile gets bigger, it's going to be under, as a part of a more balanced mix than it was in our past.

Got it. Okay. Thank you.

Speaker 3

Thank you. The next question we have comes from Mehdi Husseini of SIG. Please go ahead.

Thanks for taking my question. Greg, I just want to follow up to the last statement that you made. I'm looking at the trend, and to me, 2026 could be an inflection point in both mobile catalyzed by 2-nanometer and increased complexities, as well as Teradyne's ability in scaling AI. The robotics, which is a second derivative beneficiary of AI, from AI to IA. I'm not asking for a guide. We all understand that you have a 2028 EPS target. If all of these inflection points were to start happening in 2026, we could see the mix changing, and we could see finally a strategy coming together. Am I thinking about this right away? I have a follow-up.

Speaker 4

Yeah. No, you're thinking—by the way, I love your optimism. Our strategy is to drive growth across all of our businesses. In Semitest, all of the different parts of Semitest: compute, autoindustrial, mobile, memory. We believe that we are positioned for growth across the midterm for all of them, some of them because that end market is growing very, very strongly, and others because they're positioned for a cyclical recovery. Our IST business, we believe we're positioned for growth because of the growth in HDD bit growth, the rise of SSDs, the fact that SLT is going to be needed for AI accelerators, and the fact that we're gaining share in the mobile space. We believe that's going to grow.

Our product test group is positioned for growth through the midterm across all of the different units inside of the product test group, including the new addition, Quantify Photonics, which we think is positioned for strong growth. In robotics, we are pivoting to segments that have higher growth, and we are pivoting the organization to be able to effectively serve large customers and winning those customers. Our plan has a fair amount of resiliency in terms of the sum of the parts is greater than the whole. There is uncertainty in terms of how fast those things will go.

That's fair. In more of a near term, last night, one of your key HDD customers guided to almost a doubling in CapEx. How should we think about the system-level tests? If there is an uptick in SLT tests catalyzed by a new technology hammer, is this going to be an immediate impact, or is it going to take some time for you to actually see it?

I think it's going to take some time. First of all, in that market, the testers that we build actually have a relatively long lead time. They are built to order, and installation takes a little while. It's not like we're going to crank out 100 of these things in a quarter. We do believe that demand is going to be strengthening through 2026. You also have to remember that there were huge amounts of idle capacity in the HDD market that is being filled more by the increase in test time of drives than volume of drives. We're actually kind of hopeful about the transition to HAMR and similar technologies from other HDD manufacturers because those drives take longer to go through the test and configuration process.

I think, generally, strengthening through 2026, but you won't see a doubling like you saw in terms of the capital budget for the player.

Okay. Thank you.

Speaker 3

Thank you. The next question we have comes from Vivek Arya of Bank of America Securities. Please go ahead.

Thank you for taking my questions. For the first one, Greg, I was hoping you could help quantify how large your AI compute business in Q2. So if your SoC was about $400 million, how big is AI compute? And if you could help break it between compute and networking. You mentioned new opportunities. I assume you're referring to the potential to be part of the GPU testing side. What needs to happen for you to get more confidence? If it does happen, is that a 2026 or a 2027 contributor? Just some quantification and then GPU timeline. Is it 2026 or 2027?

Speaker 4

Why don't we take those in backwards order? I'll try to answer the GPU-related question, and I'll hand it off to Sanjay around sort of the Q2 breakdown. What has to happen for us to gain share in the merchant compute space? Basically, right now, we have to execute, and we have to prove that our test capability is equivalent, that we provide the equivalent quality, and we provide superior value. We believe that we are on a track to be able to accomplish that. In terms of timing, I would expect that if we are successful, this will have a modest positive impact in 2026, and it has the potential to build over time.

When you are working with a customer that is implementing a dual-vendor strategy, typically, they will need to continue to buy capacity from their prior sole source for a period of time as they increase the number of devices that are tested on a new platform. The good news is this puts us on a level playing field for new devices, and we believe that we have a differentiated product that we're playing with. We think that we're very optimistic about where things are going to go. I think it's important to remember that it takes time to shift test strategy at customers of this size.

Speaker 0

On the first question, in round numbers, the compute as part of SoC was roughly about 20%, and it was a key driver of growth in Q2 versus expectations. As Greg noted prior, we're seeing significant traction in the back half of the year where compute, and I'll add memory in there, is going to be the majority of the revenue in Semitest.

Got it. For my follow-up, Sanjay, how much did Quantify Photonics contribute to Q2, if any, and how much are you assuming in Q3? I assume you include this in AI compute. I know you're not giving a specific number for Q4, but based on everything you are suggesting, is it reasonable to assume that Q4 perhaps grows sequentially and year on year to give Teradyne at least some modest top-line growth overall for 2025? Thank you.

Sure. So just Quantify Photonics, which we closed May 30, is actually in the product test group. We're not breaking those numbers out, but we've had it for about a month of the results. Regarding the growth in compute, we're not providing a Q4, but generally speaking, yeah, we are seeing incremental growth. As Greg noted in his prepared remarks, we're seeing program launches toggle or straddle between Q3 and Q4, as well as Q4 and Q1. It's our expectation that there will be significant growth Q3 over Q2 and then Q4 over Q3 as well.

Speaker 4

Just to also, I think what you said in terms of significant growth for Q4 year over year, I think that's right. I think that definitely this Q4 is going to be stronger than 2024's Q4.

Sorry, I meant overall for the company, not just the compute side. Can overall Teradyne sales in Q4?

Yeah, that was what I meant as well.

I see. Understood. Thank you.

Speaker 3

Thank you. The next question we have comes from Jim Schneider of Goldman Sachs. Please go ahead.

Good morning, and thanks for taking my question. I was wondering if you could maybe first talk about the trends, Greg, you outlined before relative to mobile SoC and when that could recover. In terms of that recovery, do you think that's at earliest likely to happen in a meaningful way in sort of the middle of 2026 as you sort of prepare for those product ramps, or could it happen a little bit sooner than that?

Speaker 4

The typical mobile pattern is capacity adds in Q2 and Q3. That is not a hard and fast rule, and it has a lot to do with sort of the loading of testers through the year. If there is sufficient capacity, then we'd see the demand drifting towards the second half of the year versus being in the early half of the year. We'll be able to give you more color about that kind of early in 2026 as we get more firm forecasts about the demand. Right now, I think it's definitely more of a second-half thing than a first-half thing.

Thank you. As a follow-up, can we maybe just kind of talk about as you continue to ramp the compute SoC business in the back half of this year, is there any impact on the gross margin line, either positive or negative, in terms of the mixed-bed business contribution above and beyond just the absorption piece? Thank you.

Speaker 0

Yes. I think overall, as volume scales up, we'll generally get some efficiencies. I will remind you that, or as you recall, our product margins are really driven by tester configurations. We operate the business model at the overall portfolio of about 59-60%. We do expect that we'll continue to operate at that level.

Thank you.

Speaker 3

Thank you. The next question we have comes from Samik Chatterjee of JPMorgan. Please go ahead.

Yep. Hi. Thanks for taking my questions. I think maybe just going back to the opportunity on the GPU front that you're looking to execute on, and totally understand you're saying it is more modest in 2026 if you're successful and ramps up over time. In the past, you've quantified the VIP compute TAM for us, $300 million going to $800 million. How should we think about the size of the test equipment market when it comes to GPUs? I think you've historically excluded that because you didn't have a seat at the table. How should we think of the magnitude of that opportunity which can sort of translate over time? I have a follow-up. Thank you.

Speaker 4

In terms of the merchant AI accelerator space, GPUs in general, that's big. I mean, it is the majority of the compute time. In general terms, we're probably looking at back in 2024, there was about $2.3 billion of compute. Making progress into merchant compute taps us into a pretty big pool. Probably in the neighborhood of 2X the size of the VIP compute out in the later parts of the term, probably more than 2X the VIP compute. Getting a share of that is going to be a positive impact.

Got it. Thank you. Maybe for the follow-up, for the large customer that you have the plan of record on for the robotics business, I know you said this year robotics you expected to not be break-even, but is that embedding some of the cost relative to supporting sort of the product-related expenses for this new program? Does that give you better leverage on the operating line on the robotics segment as you go and sort of get that revenue next year? Just trying to understand when the expenses related to supporting that program are largely expected to hit. Is it this year or next year? Thank you.

As I said, the revenue impact is going to be positive in 2026. There are some expenses that we're taking in 2025, mostly associated with establishing manufacturing in the U.S. That is not specifically for one customer, but certainly we will see benefits from being able to do that manufacturing in the U.S. for that and other opportunities. The term leverage is good that as our robotics volumes increase, since we are internally manufacturing most of robotics, we will be able to absorb our fixed costs associated with operations as our volumes go up. We would expect to see that absorption having a positive effect on gross margins. We will have to see how that develops, but there's definitely a few of the expenses that are going to come in in the second half of this year.

Thank you. Thank you.

Speaker 3

Thank you. Ladies and gentlemen, just a reminder, if you would like to ask a question, please press star and then one now. The next question we have comes from Atif Malik of Citi. Please go ahead.

Hi. Thank you for taking my questions. Greg, you talked about HBM memory snapback in the second half. In the past, you guys have been of the view that the TAM is more flat this year because of high productivity of your testers. What has changed? Are you more optimistic on new HBM qualifications in Korea, or is it HBM4 that's driving the improved outlook?

Speaker 4

Just to be clear, the snapback that I'm—our view of sort of how the memory market is shaping up for 2025 has not changed that much. My comment about snapback was really in terms of the timing of quarterly revenue. We had a low quarter for memory this quarter at $63 million. We are expecting that the overall TAM for memory is kind of stable, and the demand that's coming in for HBM is going to be definitely primarily in Q4 of this year. There's some uncertainty in terms of the timing of ramps between Q4 of this year and Q1 of next year because, especially for HBM4, there are only certain devices that are designed for it. The HBM is not at the point like a DDR where it's a commodity and used across a number of a lot of different designs.

There's a high linkage between particular AI accelerators and particular generations of HBM. That is the thing that's going to be driving those ramps. I do think that the majority of capacity demand in 2025 is going to be adding capacity for HBM4.

Understand. In response to an earlier question, you mentioned that complexity is going to be your friend on the mobile side as the packaging moves from integrated fan out to wafer-level multi-chip modules. There is also discussion going on on substrate-less packaging for new AI chips in the next two to three years. I was hoping that you can kind of pull the curtain a little bit on your optimism around the merchant GPUs. Is that packaging technology the driver where you feel like you can insert your testers?

No. I think that, I mean, we're, in general, changes in packaging strategy generally are driving our customers to invest more in wafer-level test of their devices so that they have less fallout further down the line. The opportunities that we have to compete in new parts of compute have a lot more to do with those customers seeking to have more resilience in their supply chain and more ability to take advantage of the technology that each of the tester companies brings to bear. It's like the opportunity isn't hinging on a new technology or an introduction of a new device generation. It's definitely more around an operation strategy for these companies.

Got it. Thank you.

Speaker 3

Thank you. The next question we have comes from Krish Sankar of TD Cowen. Please go ahead.

Yeah. Thanks for taking my question. Greg, I have two of them. First one on the GPU test win. Did you having the networking business help with the GPU side, or was it more of a direct comparison and value proposition versus Advantest? Also, how to think about potential share gains on the ASIC side as well, where Advantest has been a legacy test supplier?

Speaker 4

First of all, it isn't a win yet. Just to make sure that we're staying on the same page, we have an opportunity to compete in parts of the market where we haven't had that opportunity before. In general, serving a customer in another segment allows you to establish reputation and allows you to compete in other parts of that customer's business. We feel like in customers that do business with Teradyne, that reputation helps us. We think that opens doors and allows us to certainly demonstrate the capabilities of our equipment and our team. I don't want to go into more detail around any specific customer. Now, in terms of share of ASIC, the market is becoming, for VIP compute, much more complex. When the VIPs or the hyperscalers were in their first generation of parts, they were definitely following the guidance of their design aggregators.

They were like, whoever their foundry or their design aggregator was saying, "You should test your part on this platform," they were pretty much doing that. As they have ramped, as they have increased the volumes of their parts and increased the expertise of their internal teams, they're taking more and more control over the test strategy and the operation strategy for these devices. We are definitely doing our best to increase share at the design aggregators, but most of our energy is going into kind of the two ends: the people who actually buy the testers, the foundries and the OSATs. Demonstrating that we have a better platform, a platform that will help them make better margins.

Then for the specifiers, the hyperscalers, the VIPs that are making these test strategy decisions, that they're going to be able to get their parts to market faster and more effectively on our platform than the competition's. The power started in the middle, and we believe that it's migrating towards the two ends.

Got it. Very helpful, Greg. Then just a follow-up on the robotics win. Is there a way to size it in terms of the number of units you'll be shipping either in 2026 or longer term? Also, are you just providing the arm or even some of the bells and whistles like the groupers, vision, etc., for this big robotics win?

I can't give you exact numbers around units. Over time, as we get further into that program, we may be able to provide more information. Right now, we're going to have to wait and see. A lot of things depend on not just us, but also how quickly and effectively the technology can be introduced across thousands of locations for this customer. I would hesitate to put a number on it. There are some interesting analyses that have been done that I think are using the right methodology, but there are pretty big error bands around it. To answer your other question, we are providing the arm. In terms of the deliverable hardware, it's really the arm and the, call it, the robotics software platform. Other software, other end-of-arm tooling is being added, vision, all of that is being done by the end customer.

Our business is primarily around the arms themselves and the support and service of those arms.

Thanks a lot, Greg. Very helpful.

Speaker 3

Thank you. The next question we have comes from Brian Chen of Stifel. Please go ahead.

Speaker 4

Hi. Good morning. Thanks for letting us ask a question or two. Greg, maybe just to kind of frame and contextualize some of this. Taking a step back, how are you sizing the overall compute TAM this year and also a subsegment of that, the VIP SAM, again, inclusive of those AI-related ASICs and TPUs? I think in the past, earlier this year, you referenced sort of a $2.3 billion for the compute market, $400 million for VIP. I know those numbers have probably upticked since then. Can you maybe sort of provide some color? Also, do you expect to perform at sort of a 50% share of that VIP SAM?

Hi, Brian. As I said, a lot of the VIP compute is kind of perched in Q4 and bridges Q4 and Q1. The VIP TAM is really tough to call for 2025. I think what you said is that there's probably upward pressure. I would agree with that, that there's probably some upside to last year's 400. In terms of referencing the compute TAM for last year, I believe that there's also significant up this year, especially in the merchant GPU space. It's pretty clear that the TAM is going to come in higher, but we're not providing an estimate for the TAM right now.

Okay. Got it. I guess on one hand, does that provide some sense that networking is a very big component in some of this improvement, as well as on the memory side in terms of some recovery or pickup on the HBM4 test and new test insertion activity?

Oh, yeah. Networking is an important part of our compute revenue in 2025. Certainly. It has strengthened from where we expected at the beginning of the year.

Got it. Okay. That's helpful. Lastly, for the follow-up, would you attribute any of the uptick in test utilization rates to any pull forward of supply chain activity?

No. The pull forward has shown up in some of our customers' results, but those customers were pulling forward demand in an environment where they essentially were able to push up utilization versus adding significant capacity. I don't think it's had a meaningful effect for us.

Okay. Great. Thank you.

Speaker 3

Thank you. The next question we have comes from Shane Bretch of Morgan Stanley. Please go ahead.

Speaker 4

Thank you for taking my question. I just want to clarify on the GPU customer question from earlier. You said you do not have the win yet, but the tone from the call has been you are very confident that you will. Could you just level set on whether these wins beyond networking are looking concrete, or are we still kind of in discussion with that customer? Thank you.

I don't know that I can put odds on it. I believe that we have strong sponsorship from the customers where we're competing, that they want to execute a strategy that gives them more resilience in their supply chain. When we set our minds to something, we tend to succeed at it. I don't want to—I certainly don't want to claim something is won when it isn't won. I don't know if that's helpful, but we are proceeding with every intention to be successful in these design ends. Until you actually have the part in production, you have to be careful about what adverse factors you can't predict. The other thing is that we've definitely seen over time that the amount of sort of how long it takes to go through the qualification process can vary by a lot. Trying to call the timing of it is also a challenge.

I think we are very happy that we have an opportunity to compete for stuff that we haven't had the opportunity to compete for for 20 years. We are—I can't guarantee that we have a win.

Thank you. That is very helpful. My follow-up is, you spoke about VIP driving compute revenue this year. Could you just talk about the kind of customer breadth you are seeing there and just visibility towards further market share gains with those customers? Thank you.

One of the key trends that we identified was verticalization. There aren't all that many gigantic hyperscalers in the world. Their importance in this space is kind of directly proportional to their capital spend on data centers. It also has a great deal to do with the number and the number of chips that they're designing, the different chips that they're designing, and the applications that they're going into. The real prizes in this space are the CPUs and the GPUs. The CPUs, the ones that are being developed by the hyperscalers, the VIPs in general, they're ARM-based, and they have a very high volume because they're pervasive across data centers. The AI accelerators are also very important to win because they have so much complexity, a lot of test intensity to them.

To answer your question directly, our VIP compute is driven by a few customers, not a single customer, and certainly it's like less than six, more than one.

Got it. Thank you very much.

Speaker 3

Thank you. The next question we have comes from Steve Bagger of KeyBank Capital Markets. Please go ahead.

Speaker 4

Thanks. Good morning. Greg, you talked about the consolidation of customer-facing functions in robotics. Was that 9% growth primarily a function of that change, or were those deals already in progress? I'm just trying to get a sense for how quickly that change can drive sustainable improvement in sales.

With the magnitude of the restructuring, we were really pleased that robotics was able to deliver quarter-on-quarter growth. That said, we typically see a stronger Q2 than Q1. The thing that I was trying to communicate was that the restructuring was executed successfully, and it did not significantly—it did not have a significant negative impact on our results in Q2. I think that we are going to—the positive impact of this restructuring in terms of combining sales, combining service, combining sales, we are going to be giving our partners a broader product line to be able to sell and our sellers a broader product line to be able to represent. We believe that that is going to be a top-line positive. By combining service, we are going to be able to offer more consistent and better service packages for our customers that buy both kinds of robots.

I would expect that the positives from this restructuring are going to play out in 2026, not in 2025. In 2025, we are really focusing on the new product introductions and the large customer wins that we have achieved and that we are trying to achieve in the back half of the year. It was a big change, and we were really pleased that it went as well as it did.

Got it. If this change really does cause robotics to turn the corner in 2026, along with the plan of record that you talked about, can you talk about how you plan to build internal manufacturing capacity? Are you going to build where you see future demand, or will some of this be fulfilled by EMS companies? What level of sales do you want to capacitize to?

We have a significant amount of production capacity. The reason that we're doing something in North America is around operations resilience and customer intimacy more than trying to really meet up to immediate demand. In answer to your question in terms of internal operations versus EMS, our strategy in Europe and the U.S. is primarily around internal. Our strategy in APAC, as we localize production, would probably be to utilize partners there because they're going to be able to localize the supply chains more effectively than we could. We'll have a blended strategy between internal and outsourced manufacturing for robotics.

Got it. Thank you.

Speaker 3

Thank you. The next question we have comes from Gus Richard of Northland Capital Markets. Please.

Speaker 4

Just in terms of China, as the domestic Chinese semiconductor industry grows and you have more fabless and IDMs there, how is that impacting your business, particularly as it pertains to industrial and auto?

Right now, the market in China. If you look at the ATE market in China, there is a portion of that market that Teradyne cannot serve because of trade regulations. If you leave that aside, the remaining market is, the biggest part is Chinese memory manufacturers. And our share in Chinese memory manufacturers is actually kind of higher than our average memory share worldwide. We compete very, very well there. The rest of the SOC market in China, there's, just like everywhere else, there's compute, there's mobile, and there's auto and industrial. We are competitive in all three of those spaces. Most of the local competition in terms of ATE in China is really focused at the auto and industrial space.

We have significant competition, and we tend to focus on the higher volume, higher quality suppliers in the space because they value the throughput and the accuracy that our equipment has.

Got it. Super helpful. In terms of robotics, one out of two robots on the planet is installed in China. The EV market is shifting to China and production. I'm just wondering, is your SAM shrinking because of that transition to manufacturing in China of autos, and a little bit of color on how competitive the Chinese manufacturers of robotics have or have not become?

There is a number of competitors in the cobot space and the AMR space in China. They have good products. We believe our products are better. They also have great customer intimacy in that market. You're right that from 2018 to now, the growth rate for the cobot market in China has been kind of twice the growth rate of the cobot market outside. The Chinese auto maker, and so we're very focused on trying to compete in China and to make sure that we maintain and grow our share in that market. It is highly competitive. I would say the Chinese auto manufacturers are definitely offering great products in EV. They are growing faster than auto makers in other parts of the world. I am not sure if that is sort of an inevitable trend or if we'll see changes in the auto industry in the U.S. and Europe.

In terms of localization, that sort of confuses that situation going into the next couple of years.

Got it. Thank you.

Speaker 3

Thank you. The final question we have comes from David Dalley of Steelhead Securities. Please go ahead.

Yeah. Thank you for squeezing me in. Much appreciated. I also have a question on the GPU space. After the analyst day, it seemed like your opportunities to break into GPU tests were based on a combination tester with silicon photonics, and you closed your Quantify Photonics acquisition. So I guess the first part of my question is, when can we expect a kind of a combo product that can do electrical and optical test? The second part of the question is, it sounds like you also have another cut-in opportunity with the big GPU guy. I'm wondering, the key reason why that is, competitive-wise. Is it because you have twice the throughput, which would obviously make your economics much better than your competitor? Or why is there an earlier cut-in point besides silicon photonics, I guess, is really the question?

Speaker 4

Okay. So in answer to your first question, Teradyne is already delivering an electro-optical test solution. Right now and over the next 24 months, that solution is going to incorporate more and more of the Quantify Photonics instrumentation. And we expect that that part of the market is going to grow. And we expect that having a superior solution in that space will allow us to grow share in the AI accelerator space, both merchant and VIP. Now, in terms of why is there an opportunity in GPUs before that. I think it's mostly to do there are two things going on. The reason that the opportunity is there is because of customers' desire to have more resilience in their supply chain. That's why we were invited to the dance.

Now that we've been invited to the dance, now it is our level of success is going to be dictated by the differentiation of our product offering. And that is we believe that we have advantages in throughput, in reliability, and in time to market. But that's exactly what we have to prove. And the degree that we prove it will have an influence over how quickly we are able to penetrate those accounts.

Okay. Great. My second question is, could you just elaborate a bit more? You talked about another HBM win at the stack level, and it sounded like you're describing another insertion point. Could you just kind of take a step back and help us understand the number of insertion points for HBM currently and how you play in that space? Just to kind of.

Sure. So.

Thank you.

Thanks for unleashing my inner Professor Greg here. Let me just sort of take a step back. An HBM memory stack has multiple DRAM die, right now eight going to twelve, and then a base die. Each one of those memory die and the base die go through a wafer-level test. Once that wafer-level test happens, all of the DRAM die are diced up, and they are assembled down on top of the wafer of base die. You have a wafer that has a bunch of HBM stacks on it. After those stacks go through, the first test is at the wafer level. The second test is after the HBM die gets stacked on the base die. We call that post-stack wafer test.

The current production methodology is you chop up that wafer of stack die, you pack it up, and you send it off, and then you assemble it into an AI accelerator. What is happening is that after those AI accelerators have been assembled, they're actually seeing significant yield fallout due to HBM problems. What they're trying to do is decrease the number of defective HBM stacks that get installed in these accelerators. The thought is, if you do a test step after you cut that HBM stack die wafer into individual HBM stacks, you're going to be able to catch a set of faults that you did not see before. There are suppliers in this space that are experimenting with adding that step to see if it improves quality.

Thank you.

Speaker 3

Thank you. Ladies and gentlemen, we have reached the end of our question and answer session. I would like to turn the call back to Greg Smith for closing remarks. Please go ahead, sir.

Speaker 4

I just want to add a final thought. AI, as we talked about in our analyst day, we believe that AI was going to have a profound and positive impact on Teradyne's business. The investments that we've made in our SOC, memory, and IST businesses to align with AI trends are now delivering new opportunities, socket wins, and financial returns. We expect that the majority of our semi-test revenue in the second half will be driven by AI applications, and it's clear that our momentum is building. I'd like to thank you very much for joining today.

Speaker 3

Thank you, sir. Ladies and gentlemen, that then concludes today's conference. Thank you for joining us. You may now disconnect your lines.