This conference call transcript was computer generated and almost certianly contains errors. This transcript is provided for information purposes only.EarningsCall, LLC makes no representation about the accuracy of the aforementioned transcript, and you are cautioned not to place undue reliance on the information provided by the transcript.
spk06: Good day ladies and gentlemen and welcome to U2's 234th Terrordent Inc. 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 conference, please press the star key followed by the number 0 on your telephone keypad. Please note that this conference is being recorded. I'd like to turn the conference over to Tracy Tuglici. Please go ahead ma'am.
spk01: Thank you. Good morning everyone and welcome to our discussion of Terrordent'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 of our performance for the second quarter of 2024 and our outlook for the third quarter of 2024. The press release containing our second quarter results was issued last evening. We are providing slides as well as a copy of this earnings script on the investor page of the Terrordyne website that may be helpful in following the discussion. Replays of this call will be available via the same page after the call ends. The matters we discussed today will include forward-looking statements that involve risk factors that could cause Terrordyne'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 10K filed with the SEC. Additionally, these forward-looking statements are made only as of today. During today's call we 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, Terrordyne expects to participate in technology and industrial focused investor conferences hosted by KeyBank, Evercore, Jefferies, Citigroup, and Goldman Sachs. Following Greg and Sanjay's comments this morning, we'll open up the call for questions. This call is scheduled for one hour. Greg?
spk18: Thank you, Tracy. Good morning, everyone, and thanks for joining us. Today I will summarize our second quarter results and discuss the trends we are seeing in the semiconductor and advanced robotics industries. Then Sanjay will go into more depth about our second quarter results and forward-looking guidance. At a high level, the market dynamics that we identified in our April earnings call have continued through the second quarter. Cloud AI is driving strong demand across the SOC and memory test markets. We have accelerated engineering and sales investments to continue to improve our market position, drive share, and increase our ability to deliver long-term sustainable growth. Outside of compute and memory, all other major test markets, including mobile, continue to be soft. Robotics has delivered on-plan results in a weak macro environment, and we continue to expect incremental growth each quarter of this year. Focusing in on Q2, we delivered second quarter financial results above our revenue, gross margin, and earnings guidance ranges. Memory and SOC delivered above our plan and showed strong performance in the quarter, primarily driven by AI applications. Continuing the trend we noted in the first quarter, Cloud AI demand drove compute revenue with considerable strength in networking devices. AI-enabled data centers have a very high number of network connection points to sustain, to support training large language models, which is changing the mix in this segment to include more networking silicon. Our historic strength in networking combined with shipments to support a vertically integrated producer, or VIP, resulted in compute revenue in the first half of 2024, exceeding all of our compute revenue in 2023. We currently expect that Teradine's SOC revenue from the compute end market will be on par with mobile revenue this year. In memory, AI-driven HBM DRAM demand remains strong. We are now seeing AI-driven servers pulling demand across a broader range of memory, including enterprise SSD NAND flash. We are also seeing memory demand for the mobile market with strength in LPDDR and continuing retooling to support the latest protocol-based mobile flash memory technologies. As a result, our memory business has grown nearly 30% in the first half of 2024 compared to the first half of 2023. While we are not changing our estimated memory TAM expectation for the year, we expect that the market is trending towards the upper end of our $1.2 to $1.3 billion forecast. Moving on to Q3, the positive impact of AI on test is expected to continue into the third quarter. However, a meaningful uptick in other end markets, including legacy auto and industrial, may not occur until the 2025 timeframe. As a result, at the company level, for the full year, we continue to expect low single-digit revenue growth from 2023. As a reminder, excluding the impact of the sale of DIS to Techniprobe, our 2024 revenue growth would have been a couple of percentage points higher. Now turning to robotics. Despite a weak macro environment, our advanced robotics business grew sequentially from Q1 to Q2. Looking at the first half of 2024, we grew 11% compared to the first half of 2023. We estimate that traditional players in the automation space have seen sales actually decline in the range of 5% to 7% over the same period. We are executing a three-pronged growth strategy for our advanced robotics business, SAM expansion, channel transformation, and recurring services and software. In the second quarter, we have made progress in all three areas. First, SAM expansion. Our new high payload cobots, the UR20 and UR30, began shipping late last year. In Q2 of 2024, these products represented over 20% of UR sales. The new AI-powered MIR-1200 pallet check was announced in the first quarter and will begin shipments in Q4 of this year. This product includes a market-leading pallet detection solution developed in collaboration with NVIDIA and has been well received by target manufacturing and logistics customers, resulting in significant backlog for Q4 shipment. Our highest priority in our robotics -to-market transformation is the development of an OEM solutions channel for UR. We have seen that purchasing cobot-based solutions from these partners get into production more quickly and have fewer problems than customers that build their own solutions or rely upon an integration partner. There are two aspects to the OEM channel strategy. The first is signing up new OEM solution partners. In the first half of 2024, we have increased the total number of OEM solutions partners by 8%. Second, we work with these partners to get them to scale, which we define as having an annual revenue run rate above $1 million. Midway through the year, we have nearly as many OEM solution partners that have reached that revenue level as we had in all of 2023. One of our largest revenue OEM partners in the first half of 2024 uses our cobots in an AI-based logistics solution. Overall, the OEM solutions channel has shown over 70% growth from the first half of 2023 to the first half of 2024. In the second quarter, the OEM channel represented over 30% of UR's revenue. Finally, because of the criticality of the processes that our robotics are being used to automate, we saw an opportunity to build a strong service business. In the first half of 2024, we launched managed service offerings at UR and Mir and are beginning to see customer uptake. On balance, the positive effect of these growth vectors and the challenging demand environment, we are expecting growth towards the low end of this year's target 10% to 20% range. Last quarter, I shared the idea that AI would be a transformational secular growth driver across Teradine's businesses. In the first half of 2024, we saw the considerable AI-driven growth in memory, networking, and ramps of vertically integrated producers. But we think that this is just the beginning. We believe that a larger opportunity lies ahead as inference applications and edge AI begins to permeate the mobile and industrial end markets, markets where Teradine is traditionally strong. We are also already seeing the impact of AI on our robotics business with AI-powered OEM solutions for UR and a strong backlog for our AI-enabled pallet jack. We believe that we are well positioned as a leading platform for the development of AI-based solutions for manufacturing and logistics. We are seeing AI-driven growth now and we expect AI to be an overarching growth driver for years to come in test and in robotics. With that, I'll turn the call over to Sanjay. Sanjay?
spk14: Thank you, Greg. Good morning, everyone. Today, I'll cover the financial summary of Q2, provide our Q3 Outlook, and planning assumptions for the full year. Now to Q2. Second quarter sales were $730 million, which was $5 million above the high end of our guidance with non-GAAP EPS of 86 cents, which was above our high end guide of 84 cents. Non-GAAP gross margins were 58.3%. This was above our guidance due to higher volumes in product backs. Non-GAAP operating profit was approximately 22%. Turning to our revenue breakdown in Q2, semi-test revenue for the quarter was $543 million, with SOC contributing $414 million, and memory $129 million. Strength and SOC was driven by both compute and mobile. Memory test shipments were driven by technology tooling for new UFS 4.0 standard and mobility. While the broader auto industry remains sluggish, we benefited from a VIP with a large purchase in this segment reflecting Edge AI's impact on the auto market. In memory, we continue to expect DRAM to dominate the memory of X. We have significant backlog for HBM, enabling strength in the memory market driven by AI. We closed the sale of our device interface solutions business, or DIS, to Technoprobe on May 27. DIS contributed $16 million to our revenue in the quarter, consistent with our expectations. In system test group, Q2 revenue was $61 million, with $17 million in storage tests on low SLT and HDD demand. Recall SLT has high exposure to the smartphone market, and even as HDD and markets begin to recover, tester utilization remains low. In wireless test, revenue was $36 million in Q2, improving as expected due to gaming and the We executed to our revenue plan in robotics. Revenue is $90 million up sequentially and increased 26% year over year. In the quarter, UR contributed $75 million and MIR contributed $16 million. Given the potential changes in the regulatory environment involving China, we thought it would be helpful to provide some insight into our revenue exposure in that region. Year to date, approximately 10% of our total company sales were shipped to China. This includes shipments to indigenous and multinational customers. Total sales to indigenous customers was less than 5% in the first half of 2024. This is consistent with the full year of 2023. Our team continues to service our customers in this market while complying with all regulations. Shifting to some cash metrics. At a company level, our free cash flow in the quarter was primarily driven by earnings and networking capital improvements. We repurchased $8 million of shares in the quarter and paid $19 million in dividends. We ended the quarter with $584 million in cash and marketable securities. The completion of the Technoprobe transactions resulted in a net cash outlay of $434 million and resulted in a 10% equity stake in Technoprobe. Some other financial information in Q2. We had two 10% customers in the quarter. The tax rate, excluding discrete items for the quarter, was .25% on a gap basis and 15% on a non-gap basis. Now to our outlook for Q3. Q3 sales are expected to be between $680 and $740 million with non-gap EPS in the range of 66 to 86 cents on 164 million diluted shares. Gap EPS is expected to be in the range of 62 to 82 cents. Some color on our Q3 revenue expectation. In our April call, our mid-guide for Q2 was $695 million. And we noted that the third quarter would be flattish. Our second quarter revenue came in higher than we forecasted on the heels of strong demand in our semi-test business. Our third quarter revenue forecast of $710 million at the midpoint is now higher than it was 90 days ago. Third quarter gross margins are estimated at 58.5 to 59.5%. And OpEx is expected to run at 38 to 40% of third quarter sales up from Q2. As Greg discussed, we are acting on opportunities to accelerate investments that we believe will drive share and long-term sustainable growth. The non-gap operating profit rate at the midpoint of our third quarter guidance is 20%. Our total semiconductor ATE TAM estimates remain unchanged from our view in April. However, we made some slight adjustments within the segments. We have included a slide in the appendix of our earnings deck with this information. Recall our SOC TAM range is $3.6 to $4.2 billion with a midpoint of $3.9 billion. This is comprised of compute, which we now estimate to be $1.6 billion, up $100 million from our prior estimate. The increase in compute is offset by a reduction in our estimate for mobile, which is down $100 million to $800 million. We estimate auto MCU at $500 million. Industrial at $300 million and services at $700 million each at the midpoint of our range. Our estimated memory TAM range of $1.2 to $1.3 billion appears to be tracking towards the high end. Back to revenue. With our outperformance in the first half of the year, our expectation for revenue distribution for the full year is now less back half-weighted than our view in April. We currently expect around 48% of the company's revenue to be in the first half and 52% in the second half. We expect full year revenue to grow in the low single-digit range compared to 2023. Note that excluding the impact of the GIS divestiture, our full year revenue growth expectation would be nearly three points higher. Now to gross margins. Gross margins have improved the course of the year and are expected to be at our full target gross margin model by the fourth quarter. Full year gross margins will likely be in the 58 to 59% range unchanged from our prior outlook. Regarding OPEX for the full year. We expect full year 2024 OPEX to grow approximately 8%, which is above our prior guidance of 5 to 7% as we accelerate investment and opportunities continue to strengthen our position and gain share. Turning to robotics profitability. As Greg noted, we expect to grow revenue towards the low end of our 10 to 20% range. We expect robotics will be roughly break even in 2024. Our gap and non-gap tax rate excluding discrete items are forecasted to be .25% and 15% respectively in 2024. With regard to capital allocation, we will continue to target our share buybacks in 2024 to an amount necessary to offset dilution from equity compensation and our employee share purchase program in order to build cash back up to $800 million. Summing up, we delivered sales and above the high end of our guidance range as memory and compute revenue exceeded our plan and semi-test. The mobile, industrial and legacy auto markets remain soft. Our robotics team delivered sequential and year over year growth as we continue to execute our new product development and go to market strategies. Our company's first half performance gives us confidence that we are on track for the year. Our midterm fundamentals remain strong and we are investing capture the opportunities beyond 2024. With that, I will turn the call back to the operator to open the line up for
spk02: questions. Operator?
spk06: Thank you, sir. Ladies and gentlemen, we will be conducting the question and answer session. If you would like to ask a question, please press the star key followed by the number one on your touchtone phone. If you would like to withdraw the question, you may press star and the key followed by the number two. If you are using speaker equipment, please lift the handset before making your selections. To give everyone the opportunity to participate, please limit yourself to one question and one follow up. If time allows, you are welcome to rejoin the question queue for further questions. Our first question comes from CJ Moose of Canterford-Sherald. Please come ahead.
spk05: Good morning. Thank you for taking the question. I was hoping maybe you could speak to visibility and outlook into 2025, particularly around semi-test. We would love to hear your thoughts around how you are seeing recovery and mobility as well as on the memory side progress with securing that second potential customer in HBM.
spk18: Hey, CJ. This is Greg. Basically, we are pretty bullish on 2025. The semi-test market in 2024, we're seeing strength in computing and we're also seeing a fair amount of upgrade business where people are converting testers that were underutilized in mobile to use for compute and VIP applications. That's soaking up some of the excess in that market. We expect that to really sort of accelerate business as mobile returns in 2025. It won't have that capacity to grow into. On HBM, I think we continue to expect that we are going to make progress in two directions on HBM. One is that our share has primarily been in the pre-stack wafer test of HBM. We believe that we're going to be gaining share in performance tests of the HBM memory after it's been stacked. We're hoping that we can expand both our market share within the account where we are participating in HBM and then also break into additional accounts for HBM. That would be more of a 2025 thing. I think overall, our outlook for 2025 and into 2026 is in line with what we talked about in January.
spk05: Excellent. I guess as a follow-up, could you update us where you are with VIP? I think you talked about one large auto customer in June. Can you kind of summarize that for the industry for this year? How do you see that growing next year and what additional kind of customers or end markets, whatever color you can give us in terms of how to think about the trajectory to that business into 2025?
spk18: Sure. I think the way that we're looking at VIPs right now because that growth has been happening at a time when there is a fair amount of capacity in OSATs to soak up. We're actually trying to keep track of the number of testers that are being used for these VIP customers. Right now, it's hundreds of testers. There are currently hundreds of testers that are being used for VIP. We expect that to end. We have I guess we should probably call it multiple VIP customers that are loading significant numbers of testers. We have additional VIPs where they've made plan of record decisions to put their parts on our testers. Those parts are going to be released later in this year and into 2025. We think that we'll probably exit this year with twice as many testers being used for these vertically integrated producers than we
spk03: have right now. Thank you. Our next question
spk06: comes from Tim Okuri of UBS. Please go ahead.
spk17: Thanks a lot. Sanjay, can you help us handicap the guidance for the DIS sale? I know you said it contributed 16 million in June. If I run rate that out, it's probably a 20 to 25 million dollar number that we should effectively adjust the September guidance for. Is that correct on a rate basis?
spk14: Yeah, if I go back to 2023, it was about a hundred million dollar business. Twenty to twenty-five million dollars is roughly the number. This quarter, with the sale occurring into the second month on May 27th, it was 16 million dollars. I think that estimate you have is reasonable. As I noted in my prepared remarks, we expect growth year over year to be in the low digits. Without DIS, that's a shortfall of roughly three points. You would increase that by three points if there was no DIS in 2023 and 2024 from a growth perspective.
spk17: Okay, got it. Greg, talking about next year, you guys have always talked about N2 being a big driver for the SSCTM. We're seeing clear evidence that there's going to be a lot more N2. We're seeing upside pretty much everywhere from N2, wave for demand next year, and the speed of the ramp. I would think you're pretty optimistic about, I mean, sure, there's the mobility piece, but I would think you're pretty optimistic about next year because of N2. Can you speak specifically to the, you've always talked about that as a big driver, and here it is. Can you talk about that?
spk18: Sure, yeah. We tend to think of the process nodes not as, from our business's perspective, not as a driver, but as an enabler. So the two nanometer or gate all around, they're going to be enablers for higher device complexity, but the thing that really drives our business is whether the end market is pulling for that complexity. And we're really optimistic that between cloud AI, and especially for edge AI, like in mobile, that that is going to be a major driver for companies to get into N2 even at a faster rate than they got into N3. So we think that that's, that it's a good thing that the capacity is being put in place for N2, because we think that it's the deployment of edge AI, it's really going to increase complexity of the application processors. It's also going to be pulling harder on the mobile part of the memory market. So we think that there's sort of a virtuous circle in terms of the capacity being put in place, the end market demand that's going to use that capacity, and then the pull effect on other parts of the market enabled by the features that are going into these products. So we think basically this is going to be a good demand driver really through the rest of this midterm, 25, 26, 27.
spk03: Got it. Okay, thank you.
spk19: The next question comes from Mitty Hussini of SIG.
spk06: Please go ahead.
spk12: Yes, thanks for taking my question. I want to follow up on your commentary that mobile SOC revenue in 2024 will be comparable to compute. I understand the diversification of customers and additional data center, but your assumption also is impacted by a weak mobile market. So assuming that N2 and smartphone refresh in 2025 would help you with the mobile SOC, should we also assume that the compute would also grow at the same rate? In other words, this equilibrium with its sustainment to next year?
spk18: So that's a good question, Mitty, because there's a little bit of subtlety to this. So you're right that one of the reasons that our mobile revenue and our compute revenue are going to be comparable in 2024 is because the mobile market is weak. But the other reason is, and I talked about this in my prepared remarks, that our compute revenue for the first half is bigger than all of our compute revenue for 2023, and we see good strength for compute in the second half of year as well. So things are going great in compute. Next year, we think that compute is going to remain strong, but the TAM sizes for compute are likely to be in the same range, maybe a little bit stronger than where they were. The mobile market really should come back pretty strongly, and so our mobile revenue will come back with that. So we anticipate that we'll have good revenue growth in compute, but it's very possible that mobile will outweigh it next year because of the strength of that recovery.
spk12: Okay, thank you. And then one follow-up for Sanjay. If robotics business is going to be breakeven in 2024, then profitable next year, is the underlying assumption baked into your 26 target, or could there be an acceleration of profitability? I just want to see how the breakeven in 2024 for robotics fit into your 26 target. Thank you.
spk14: Sure, so in 2024, we started the year with thinking that we'd be just a little bit above breakeven, low single digits. As we come, as we look to the year and we see the range of 10 to 20 percent this year of growth, and we see that coming in more to the lower end of the range, we think that number is about breakeven. And in our earnings model, we had 20 to 30 percent growth. So with that incremental scale, we did build in a level of profitability in the outer years, and we still believe that that trajectory is going to occur as we move into the future.
spk03: Okay, great. Thank you.
spk19: The next question comes from Vivek Arya of Bank
spk06: of America. Please go ahead.
spk10: Thanks Vivek, my question. Greg, for next year, as you suggested, the turnaround in mobile seems to be the biggest catalyst for Teradyne. So when we look at the time of about 800 million for this year, could you help us give some range for what recovery looks like? Are we talking 10 percent, 20 percent, 30, 40? Can you help us have some broad range, and how much of this you think comes from mobile industry with faster unit growth? How much of this comes from more testing complexity? And then a kind of a small nuance question. Your large customer could also have a change in their modem suppliers for at least some SKUs. Is that net positive, negative, neutral to your business?
spk18: So let me take the second part of your question first, and then actually I'm going to tag team the first part with Sanjay because I want to work on sort of a history lesson perspective of the mobile TAM to try and bracket what we would see in 2025. But in terms of the modem supplier for our historically largest customer, the thing to remember is that whoever is supplying that modem is also likely supplying all of the ancillary devices associated with that modem. So there's a modem, there's power and there's a lot of pluses and minuses. The thing that I will tell you is that they're all relatively small relative to the magnitude of the overall AP business. So if that was to change, I would expect it to be a slight positive for us, but I don't think it's a game changer kind of a thing. And so on the mobile TAM, so Sanjay, could you just give us sort of the history of the mobile TAM like over the past few years?
spk14: Sure. So this year at the midpoint, we as we've noted 800 million and last year we noted approximately 900 million and in 22 it was roughly 1.6 billion. It peaked our view in 2021 at approximately 2 billion.
spk18: Okay, so take what I'm going to say with a grain of salt because we are entering into our strategic planning process and haven't completed sort of an updated view of 2025. But the thing that I will tell you is that the mobile TAMs are unlikely to go as high as we saw during the COVID peak, but they're likely to be significantly stronger than what we're seeing in 23 and 24. So going in the middle of that range might not be a bad idea, but it's a pretty big range. Vivek, does that help at all?
spk10: Yes, extremely helpful. Thank you so much. And for my follow-up question on the robotics business, you know, you've taken the growth rate this year to towards the lower end of the 10 to 20 percent. I think you mentioned some, you know, macro factors. If I look at again, you know, a historical view, that segment continues to disappoint every year in terms of its growth rate. It's not really profitable. So I'm curious at what point do you think it actually realizes the promise and anticipation, right? Is it a matter of scale? Is it a matter of channel? What will it take for it to actually achieve the growth rate that you anticipate for that business?
spk18: That's a great question. And so I think you are, in terms of your hypothesis, you're hitting the nail on the head that part of it is scale, part of it is channel. The thing that we are really, we're actually really optimistic about robotics business right now that the macro environment is really weak. Our automation peers are struggling with declining sales. We posted an 11 percent gain over the first half. And the thing that is encouraging to us is that that 11 percent gain is on the back of specific aspects of the strategy that we're executing. You know, we are, we were trying to expand our SAM by adding a high payload, high payload options for universal robots. And those robots are accounting for 20 percent of our sales in the quarter. We are trying to build additional channels and our OEM channel grew by 70 percent year on year for the first half. We have additional, additional irons in the fire to drive additional growth that have not yet paid off. And there are primarily other aspects of market expansion. So our pallet jack for the AMR business, additional products that are coming to market for you are as well. We have additional channels that we are building. We've been working for the last year and a half on building out a large accounts channel. And we have significant pilot pipeline for that, but we don't have significant revenue growth to see from it yet. And then finally, we are, we think that there's a very significant revenue opportunity in service and recurring software. Right now, that's a vanishingly small part of our revenue in robotics. But every business that has machinery in critical processes definitely needs to have a way to keep that stuff running. And we have an example in our semiconductor business where service represents a significant portion of our overall revenue. And we are trying to apply that same model of managed service plans in robotics. And we're actually seeing good customer uptake, but not a lot of revenue yet. So we think that those three engines are going to be kicking in in 2025 on top of growth that we're going to still get from the things that we've done in 2023 and 2024. So we're pretty optimistic that the end market is terrible, but we are, you know, we're actually doing what we said we were going to do in terms of the strategy.
spk06: Okay. Our
spk03: next question
spk06: comes from Krish Sankal of TD Cohen. Please go ahead.
spk08: Yeah, thanks for taking my question. I have two of them. Greg, just first on the clarification, you maintain your full year revenue guide of low single digits, but I don't want to compare the last quarter, you kind of increase the test revenue up modestly. So I'm just kind of curious, is this just a little bit of nuance that is within the margin of error or anything else happening over there?
spk14: Hi, Krish, it's Sanjay. At a top level, you're right, we did guide the full year consistently with the first half a little bit stronger than we anticipated. And what's happening under the covers is semi-test is up, but our other test businesses are down a little bit in the way of, you know, in storage, it's tied to the global market and SLT and the HDD as that market returns, they're still under utilized capacity. In our defense and aerospace business, we had some projects push out and our production board business, we had, it's tied to the automotive slowdown. And so, you know, overall, semi-test is stronger and the other test is lower. And then we did move to the lower end of our robotics range. So at a top level, we're calling the same year, but, you know, semi-test stronger, other tests a little weaker, and robotics is at the low end of the range.
spk08: Got it, got it. Very helpful, Sanjay. And then just a quick follow-up, you kind of mentioned strength networking test. You know, if you look at some of the GPU providers, clearly, you know, Verigee Advantage has a very strong position. Do you think this networking test opportunity can manifest itself into some share gains on the GPU side, or you think that AI opportunity is more to the VIPs, not to the semi-device makers?
spk18: I'm really, really glad you asked that question. So there are a couple of things going on in the compute space around this topic of cloud AI and building large training, large capabilities to do training of large language models. So the first is obviously huge demand for GPUs. You know, that's why people have seen, you know, Nvidia run up as much. They are providing an incredibly valuable engine for all of this. But if you look at an AI enabled server, there is a much larger number of high speed data links within a server rack than there is for a traditional server. And so we, you know, in you know, more traditional, traditional accelerated computing, we were seeing a ratio of, you know, like within a specific account, it might be 90% GPUs, 10% networking, that's trending more towards 20% or even higher. So there's, you know, there's definitely a tailwind there that is beneficial. The other is as these models become more sophisticated and the players in this space are trying to differentiate between these large language models, they're each trying to gain an advantage through architecture. And that's driving this VIP design trend that we're seeing. So we think that there's enough end market demand to drive growth for both sort of traditional GPU based AI models and also drive even faster growth for the VIP or, you know, home grown silicon that is being built into these large server farms.
spk03: Got it. Thank you. Very helpful.
spk06: Our next question comes from Tisha Harry of Goldman Sachs. Please go ahead.
spk15: Hi, good morning. Thank you so much. Greg, you talked about strength in compute, you know, soaking up some of the idle capacity across the mobile space. I'm curious where test utilization rates are in mobile across, you know, OSATs and your IDM customers to the extent you've got some visibility. And then for auto and industrial as well, where are utilization rates generally today? And when you talk about a recovery in 2025, you know, I know it's hard to predict, but at what point do you think utilization rates get to a point where customers, you know, resume purchases of testers?
spk18: So I think to answer the last part of your question first, I think we're very close to the point where the balance is going to tip back away from soaking up capacity to driving new system sales. The trends that we're seeing, third party data around test utilization are showing a good uptick, you know, a little bit above sort of typical seasonality. Our own data is showing mixed utilization right now. We don't see a clear trend there, but we do have a fair amount of both systems that have already been converted and also upgrades, you know, that would be used as conversion kits. And so that is going to essentially occupy hundreds of testers that are underutilized now. So we are expecting utilization to take up pretty strongly in the second half of this year and for that to tip over to the point where it turns into real system demand. When you get to the IDM customers, especially in auto and industrial, there's an interesting, you know, many of those customers are down significantly year over year from 23 to 24, but a lot of them are persisting in relatively aggressive capacity expansion plans. They, you know, they see the same thing that we've been talking about in the automotive space, that there are going to be short-term fluctuations, but the crossover to EVs and hybrids is something that is a question of like when, not if. That is going to happen and the amount of compute power in cars and other electronics in cars is going to continue to increase the attach rate of, you know, more chips per car shift. So they have long-term plans to drive capacity and they have long-term plans for purchasing our equipment that we think is going to be important. So we're really, you know, we are, you know, cautious about the current downturn that we're seeing in automotive, but we're really confident about the long-term potential in that space, especially parts of it where we have a ton of good share and great customer relationships like battery management, discrete and power, and more and more importantly ADAS stuff. So I think, you know, right now I think the utilization in auto industrial is down, but what the signals that we're hearing from those customers is that they expect that to be relatively short-lived.
spk15: Got it. That's really helpful. Thank you. And as a follow-up on the robotics side, you know, your point about AI, you know, helping your UR business and the mirror business makes intuitive sense to me. I just have a hard time sort of, you know, thinking about the timing and sort of the magnitude of that kicking in. You know, you mentioned backlog for your new mirror product being pretty significant into Q4. Maybe we start to see benefits there, but on the UR side, what kind of applications should we be looking out for as it pertains to AI helping that business? And again, at what point, you know, does sort of the contribution of AI start to really show up in numbers? Thank you.
spk18: Okay. So there's a lot to unpack there. So let's get down to it. So the story for Mirror is pretty simple that take a robot, add AI capability, and suddenly you have a robot that works better. You know, that's our basic hypothesis behind the pallet jack is that we're going to be able to have a higher success rate picking pallets than competing solutions. And so far, the competition customer reception is that they think that those claims have some merit. So we have a pretty strong backlog for a product that we're going to start shipping in Q4 of this year. So that's pretty simple. For UR, the story is a lot more complex because what we're really doing is creating a platform for our solution providers to build AI-based solutions on. And there's actually like four chunks of an AI-based solution on UR. At the bottom, you have our robot and the software that controls our robot. And then on top of that, there's AI toolkits that can come from NVIDIA or can come from other providers that really give the primitives that allow people to use AI capabilities to solve problems. Then on top of that, there's usually a solution builder. And so what that solution builder will do is they'll take our robot and that toolkit and they will build a solution to do something like heterogeneous bin picking or vision-based inspection or welding. And then they will take that solution and they will market it to their end customers. So the key thing that differentiates the UR platform from other robots is its features as a platform. So the APIs that people can use, the features that they need in order to build these solutions for, and that's what makes us sticky, is that once a solution provider builds a solution around a specific toolkit and our robot, it's tough for them to move off. The thing about that model for UR is that it's a slower burn. It's an indirect, AI drives our business indirectly because we're not building the AI solution, our solution providers are. So I'll tell you that we have AI-based solutions in the market. Right now, our estimate is that a high single digits of UR sales are going into AI-based applications that are in the market right now. So that's sort of our baseline and we intend to grow from there. These solution providers, it can take a while from when they begin building their solution to when they see a commercial inflection. And that's often in excess of a year. So we have a great pipeline of partners that are using AI right now to try and build solutions. And we would expect that to deliver material revenue in 2025, but not in 2024. So sorry for going deep, but you kind of hit a nerve.
spk05: Thank you for all the details.
spk06: The next question comes from Joe Moore of Morgan Stanley. Please come ahead.
spk09: Great. Thank you. I wonder if you could talk about the HBM test. In particular, I'm just curious, as we move from HBM 3 to HBM 3E and then later to HBM 4, can you talk about how much test reuse there's going to be versus upgrade requirements?
spk18: Okay. So on HBM, the primary difference from a test perspective as you go from HBM 3 to 3E to 4 is the data rates that are required in the performance test of the final HBM product. And the capacity that's in place right now for HBM 3, most of it, and essentially all of it, is incapable of testing at the data rates required for HBM 3E and HBM 4. So the people that are building HBM 3E are seeking additional capacity. They're essentially retooling to be able to test the HBM 3E. Now, this is one of the areas where we believe we have an advantage that will allow us to gain share in HBM performance test. That we have a tester that is going to be effective for both HBM 3E and HBM 4. And so that essentially will give customers that adopt that solution longer asset life than if they were to adopt a competing solution. So we are pretty hopeful about our opportunity to gain share in HBM performance test. And we're working on essentially benchmark competitions in multiple customers to try to prove that.
spk19: Great. Thank you very much. The next question comes from Samik Chatterjee
spk06: of JP Morgan Chess & Co. Please go ahead.
spk07: Hi. Thank you for taking my questions. I guess for the first one, Greg, if I can go back to one of the comments that you made about the compute time for next year when you were comparing compute versus mobility and your expectations for next year, I'm just wondering, you said compute to be robust, but remain at these sort of strong levels. We've seen the compute time go from, I think, 1.2 to 1.6 this year. Are you implying that we sort of, 1.4 to 1.6, I'm sorry. And are you implying you sort of hold around this 1.6 level or a similar increase that you're seeing this year? And what should we think about what's incorporated in that for the portion of the market that's sort of associated with the VIPs and growth outlook for them in 2025 in that number?
spk18: So I think it'd probably be a good idea to do our TAM history lesson in compute as well. So, you know, really strong this year. Why don't we go back a couple of years and set that as context and then we'll bracket kind of what we think we'll do in 2025. So Sanjay, could you do that?
spk14: Sure.
spk18: So let me give some
spk14: more context. So 2019, the compute TAM, our view was about 600 million. You know, fast forward, we went to nearly double that in 2021 between 1.1 and 1.2 billion. And then in 2022, it went to 1.3. And as you noted, 23, at approximately 1.4 billion. And this year, at the midpoint, our view was 1.6 billion.
spk18: So as you look into 2025, you know, we don't think that we would see mammoth increases to the compute TAM. It's, you know, it is, you know, between 2x and 3x larger than it was in 2019 at this point in time. And for sure, people are trying to add capacity all through the, you know, all through the production chain. But it's coming off of a really high base. So looking forward into next year, I would probably be looking at kind of where we are this year, plus or minus something versus a much, you know, a lot of increase in the compute space. Now you asked about VIPs. Right now, you know, VIPs are kind of in the, you know, are towards the low end of a, you know, a 100 to $200 million range. And, you know, it could trend towards the high end of that by the end of this year. We think that by the time we get out to 2026, VIPs will be about $500 million of the overall compute spend. So that's a, you know, I think that's a pretty exciting chunk. There could be, you know, a quarter to a third of the compute TAM could be driven by VIPs.
spk07: Got it. Got it. And for my follow up, I know you're reiterating your 2026 plan today, which still sort of implies a -25% KGOR for the next couple of years. And I'm just trying to compare that with the exit run rate this year that you have in terms of growth. Even when I add back Technoprobe and the impact of that, it's sort of in, is in that sort of 10 to 15% range. Can you help me just think about what, how to think about the impact that the underutilization mobility is having on your revenue just to think whether, I mean, how do you bridge the gap from this 10 to 15 in the back half to maintaining a -25% KGOR for the next couple of years?
spk02: So I guess
spk18: the best thing to say is that at the point in time, there are two things that are important. One is that the capacity was larger than we thought at the point that we built our plan. The other thing that's happened since we built our plan is that even in 2024, compute has strengthened by a couple hundred million dollars. That the, you know, sort of the underlying demand drivers in the market are stronger than we were, than we were forecasting. So when you balance the, the sort of deeper hole of utilization and the stronger end market demand, then I think our hypothesis for 2026 holds.
spk14: Maybe I'll just add a little bit. You know, when we take a look at our model, you know, as Greg noted earlier, mobility comes back. But let me remind also what he said, it doesn't come back to the peak levels. It comes back to kind of a regular level. But mobility comes back. The trends in automotive with more silicon going in, combustible engines, the conversion to EV and battery management systems is another tailwind. And industrial comes back coupled with the compute. And I'll remind you that, you know, our robotics business had a 20 to 30% growth in there as well. So that's kind of how you get there.
spk07: Yep. Thank you. Thanks for taking the questions.
spk06: The next question comes from Brian Chin of Stifle. Please go ahead.
spk16: Hi there. Good morning. Thanks for letting us ask a few quick questions. Greg, maybe just curious, in Q2, in compute and memory, which obviously drove a lot of that sequential strength, was there any shipment pull-in from second half of 3Q?
spk18: I'll actually let Sanjay answer that one.
spk14: Yeah, Brian. You know, we did see acceleration from Q3 into Q2. We, you know, customers are pulling. And that's, you know, in my view, we've kind of, well, we have a full year view and semi-test is a little bit stronger, as I noted earlier. Customers are pulling and the demand is there. And we see a lot of that going into Q3 with kind of low turns. So we're in a, I think, from a compute and memory perspective, yes, we did see some acceleration. Okay, great. That's
spk16: helpful color. And then, yeah, maybe just thinking about the service and software opportunity in robotics small today. But just curious, can you sort of size up the active installed base of UR and mobile robots in the field and what their typical useful life is? And how big could service and software be as a percent of revenue by 2026?
spk18: Yeah, so rough numbers here, but we have on the order of 80,000 UR cobots out in the field, and probably on the order of 10,000 or more AMRs. The useful life of them is, you know, we have cobots that were delivered, you know, before Teradyne bought universal robots that are still being used. But most of our customers will depreciate the asset over about a five-year period. So I think it's, if you're trying to model sort of installed base decay, it's not a bad idea to look at maybe a five to seven-year life. So I think, and you're going at this exactly the same way that we're thinking about it, that we look at the entire installed base, not just new equipment sales as our total available market for service agreements. And we're also doing this service agreement build in alignment with the build out of our large account channel. So large accounts are the ones that are most likely to have large fleets and are most likely to see advantage from a managed service plan. So that's another sort of aspect to the strategy that we're building out.
spk03: Okay, great. Thank you.
spk19: The next question comes from Atif Malik of Citi. Please
spk06: go ahead.
spk11: Hi, thank you for taking my questions and thank you for providing the history lesson on the TAM on slide 13. A quick one, the mobility weakness, fair to assume that is on the Android side?
spk02: The mobility weakness is really
spk18: both Android and iOS. If you look across the major suppliers for silicon for all of those ecosystems, it's generally pretty weak. Now, if you peel it back a little bit under the covers, there are some areas of the mobile space that have done better during this downturn than others. So one aspect of it is in image sensor that the transition to like 48 megapixel image sensors really put some pressure on the suppliers in that space and they've added capacity both in iOS and Android land. The other is the improvements in wireless charging and other charging aspects that even now when we've seen significant weakness in power and linear, we're still seeing some incremental demand as people always want their phones to charge faster. They want them to charge faster on wireless charging. So we're seeing increased silicon content there. So it's been a little bit of a bright spot within that market. So I think where we are, we've seen broad-based weakness in 2023 and continuing into 2024, and we expect to see relatively broad-based recovery in 2025.
spk11: Great. As my follow-up for Sanjay, you've talked about a full year gross margin 58 to 59 percent. And if you were to assume that the mobility and compute becomes larger at your large customer next year, all else being equal like auto industrial, how will that impact the gross margin profile for next year?
spk14: Yeah, so we're happy with this year's gross margin performance kind of playing out as we expected a little bit better though in the first half, but 58 to 59 percent with continuous improvement. And I would say that regarding next year, obviously we'll provide a little bit more color in January. But the way I think about it is we do expect revenues to grow. So we'll have a bit of a tailwind tied to volume. And it really depends in those segments you referred to as well, specifically the customer is ordering. It depends on the configurations of the instruments. And so I think in my view, I think that we're exiting the year at our model. That's what we expect. And look forward to giving you more of an update in January for 2025. But I would leave you with, you know, with a higher volume level, it should help as tailwind.
spk19: Thank you. Our next question comes from Steve Parker of KeyBank Capital Markets.
spk06: Please go ahead.
spk13: Hey, good morning. Thanks for stretching the call. Greg, you talked about all the growth opportunities for both test and robotics for 25 and 26. And I wanted to tie that back to the accelerated OPEX investments. Is that increased spend supporting everything we've talked about today or are there also new products you're planning on that add to the existing lineup?
spk18: So I will tell you that our OPEX acceleration is in both R&D and in our sales and marketing expenses that we are, you know, so you can infer from that that that means that we're working on new products and we're working on building relationships with new customers and trying to capture, you know, the key thing in semiconductors is that once customers make a platform decision, then it becomes very, very difficult to change, change their direction. And so we see a fleeting opportunity with all of the new VIPs to get in the door and influence that initial platform decision. So we're going to take those opportunities and we're going to make the investments to try to capture them while that share is available versus trying to move it to the next level. So I'll just say that we're confident in our strategy and robotics and we're confident for the OPEX plan that we put together. We saw no need to pour on more OPEX in our robotics business. We're going to execute the plan that we set. But when in the next year, we're going to be testing, you know, in the system level test business, we saw an opportunity by accelerating our spend that we would be able to drive a better outcome long-term. So that's where that money's going.
spk13: Understood, and just one last one, you said there's good customer uptake in industrials for services but not a lot of revenue yet. Is that because the new customers are not taking all of what you offer or why is revenue not tracking uptake?
spk18: So it is mainly because the rollout of those programs, you know, we're trying to make sure that we can walk before we run when it comes to these service programs. So we have essentially piloted the programs in a relatively small region, and we're seeing good results in that region, but we have not deployed the program worldwide because we have to build out the infrastructure to support it. So that's why I say that we haven't seen material uptake is just because it isn't globally deployed at this point.
spk13: Does that happen in the back half of this year or more 25? Oh,
spk18: definitely 25. You know, so in the back half of this year, we're going to be, you know, and also the sales cycle time for the service business is it's not terribly long, but it's probably a couple of quarters because you have to establish the value for the managed service plan. So we have a lot of interest. We are talking to a lot of customers, but we haven't closed the deals yet because I think some of it is that it needs to be incorporated into their 2025 budgets. You know, so I'm very hopeful that we'll see an inflection in service revenue in 2025. I don't think we'll see much in terms of revenue in 2024.
spk13: Understood.
spk03: Thanks.
spk06: The next question comes from Gus Richard of Northland Capital. Please come ahead.
spk04: Yes, thanks for taking the question. I'm just curious, you know, in terms of the jump falls with the vertically integrated producers, you know, where's your win rate? How much of that business do you guys think you're picking up?
spk18: So our win rate on sockets, when we sort of started describing our plans in VIPs, we said that we were trying to make sure that we got at least sort of an even split of sockets. And right now in 2024, that's kind of how it's playing out, that we're getting our fair share of wins in terms of the other way that we're looking at this is in terms of testers loaded. So we're not looking at it in terms of tester purchases because of the underutilization, but we are trying to count noses in terms of how many testers are being loaded by our customers versus testers being loaded by our competition's customers. And right now I would say that we are kind of 60-40 in favor of Teradyne in terms of testers loaded for VIP customers. Got
spk04: it. That's helpful. And then the follow on is, you know, thinking about your investment in Technoprobe, you know, how is that helping, you know, your effort to expand your market share in these jump balls?
spk18: So right now, right now it's not having a significant effect. The key thing that we are trying to do in the joint development projects that we have with Technoprobe is to take on relatively hard problems that required collaboration between us and Technoprobe that, you know, like unless we had a firm partnership, we'd be unwilling to take that risk on speculation. We have initiated those projects, but they do require technology development on both sides. So, you know, right now what we said at the point in time when we announced the deal is that we thought this would have a low single digit long-term effect on market share in the semi-test space. We still think that that's true and we think that we're on track to achieve those gains.
spk04: Yeah, and just to be clear, are these the VIPs or, you know, the, you know, existing wins and losses that have already happened? Do you think it can increase your penetration, you know, in areas where decisions have already been made?
spk18: So we haven't publicly announced sort of what segments of the market we are doing these joint development projects because we think that we need to keep that relatively proprietary until we have demonstrated the success. But I will say that, you know, the key things that we believe are enabled by the projects that we're working on with Technoprobe are around devices that have very high power requirements and also devices that have very high parallelism. You know, so the TPI probe technology is unlocking sort of the high-end scalability of our test platforms in both SoC and memory and we think that that's going to result in share gain by sort of opening up a place where both TPI and Teradyne are differentiated from their competition.
spk02: Got it. Thank you. Very helpful.
spk06: Thank you. Ladies and gentlemen, we have reached the end of the question and answer session. I will now hand over to Tracy Tuchiguchi for closing remarks. Thank you.
spk01: Thank you all. Thank you all for joining us this morning. We look forward to seeing many of you through the course of the quarter and until then, have a great morning. Bye.
spk06: Thank you. Ladies and gentlemen, that concludes today's event. Thank you for attending and you may now disconnect your lines.
Disclaimer