Good afternoon, and thank you all for coming to Teradyne's first Analyst Day in 20 years. I'm Traci Tsuchiguchi, our Vice President of Investor Relations. As Jason and Travis Kelce would say, we've had a lot of new news. All the press releases and the materials associated with this event are all available on our Investor Relations website. Today, through the course of the presentations, we'll be making a number of forward-looking statements. Those statements inherently have risks and uncertainties that may cause results to differ materially from what are our current expectations. We ask you to review the Safe Harbor statement available on our materials, as well as our risk factors in our 10-K. We will also be making reference to non-GAAP financial measures. Reconciliation of those non-GAAP financial measures is available on our Investor Relations website as well.
Now on to the good stuff. The agenda for today will be we'll kick it off with Greg Smith, our CEO, who will talk to you about the secular trends that we're seeing in the industry and how our business strategy is aligned to take advantage of these secular trends. He'll then hand it off to Rick Burns, the President of our Semiconductor Test Business, who will talk to you about the beauty of our Semiconductor Test Business and why we believe we're competitively positioned for sustainable growth there. He'll then hand it off to Ujjwal Kumar, President of our Robotics Division, who will talk to you about why now is the right time to be in the robotics business and how AI is transforming our served-available market.
Our newly minted President of the Product Test Division, Regan Mills, will talk to you about the businesses within our Product Test Division and how they seamlessly align to our Semiconductor Test Group. He'll also give you a bit of a deep dive in terms of silicon photonics and what we're doing there. We will take a short break, 10 minutes, and then come back to hear Sanjay Mehta, our CFO, tie everything you've heard together in the presentation materials to sort of tie it into our 2028 model. We will end it with a Q&A session. Following the Q&A session, the webcast will end, and I'll give more explicit instructions. For those of you who are in the room and who've made the trek to glamorous North Reading, we're going to treat you to some demonstrations.
We will break up into three groups and allow you to rotate through three different demonstration sessions that will allow you to better understand what we are doing in terms of our technology and the core of our products. Without further ado, I would like to invite Greg Smith, our CEO.
Thanks, Traci. Good afternoon. As Traci said, I appreciate all of you making the trek here to North Reading. When we were planning this first Analyst Day in 20 years, one of the things that was really important to us is that we really think that the key to Teradyne's future are our products and the people that are making them. By doing the Analyst Day here, we thought we'd be able to give you an exposure to a much broader range of people and also to get a firsthand understanding of the technology and our products. Today, we're trying to come up with a balanced approach. I think we're trying to do something that's a little bit challenging.
What we're trying to do is we're trying to communicate sort of a sense of responsible caution around the volatility in the short term and still communicate the elements of our long-term plan. We think that that long-term plan is going to drive strong earnings growth, and we want to make sure that you have an understanding of how that fits together. The majority of the time is going to be spent on these long-term trends. Before we go into that, I'd like to give you a quick update on our current business conditions. We are seeing some short-term volatility, and it's mainly affecting the semi-test business, almost exclusively affecting the semi-test business. We're seeing some uncertainty around tariffs and new trade restrictions.
That news is causing some of our customers to either reschedule deliveries that they have booked or they're reviewing their capital plans in terms of when they'll be placing orders. We haven't seen any cancellations, but we are seeing these reschedules, and we wanted to try and share that news with you as soon as it became clear to us. We do not anticipate any impact to our Q1 results. Our Q1 guidance is still intact. At the time of our January earnings call, we expected that our Q2 revenue would be up 5%-10% from Q1. As of right now, we believe that our revenue in Q2 will be flat to 10% down. Just to be clear, that's the change that we've experienced from January. We think it is related to this short-term volatility.
We don't have enough information yet to know whether it is a trend or not. Now, when we talked about 2025 in full in the January timeframe, one of the things that was a clear trend in our plan was that the second half of the year was stronger than the first half of the year. We still believe that to be true. Most of the weakness in the year we have taken out of Q2, we have moderated the second half slightly. At the time of the earnings call, we believed that we would have growth year on year, roughly 15%. At this point in time, we believe that growth is going to be between 5-10% versus 2024. Also, at the time of the earnings call, we went through the process of updating from a 2026-based earnings model to a 2028-based earnings model.
We said at that time that we expected our 2026 revenue and EPS to be in the previously defined range for that 2026 earnings model. We want to make sure that you understand that right now, given the weakness in 2025 and the amount of growth into 2026, we believe that we're trending towards the low end of that earnings model. In the range, but towards the low end. We will have an update on all of this when we get to our April earnings call. We wanted to share this information because we felt it was material in terms of how the year was going to pan out. Now I want to shift back to the longer-term perspective. That longer-term perspective is really, when you're talking about Teradyne, the long term is long term. Teradyne is a 64-year-old company.
It depends on whether you measure it from when our founders had the idea or when they had a product. We have been through a lot, and what we have been through has really helped to shape the way we look at the market now. The way we look at it right now is looking out from 2024 to 2028, we believe that we are positioned to more than double our earnings per share from 2024 out to 2028. I would like to sort of explain the elements to how we think we are going to get there. Back in 1962, the founders of Teradyne saw an opportunity. At that point in time, the process to manufacture semiconductors had been invented, but it was quite an immature process. You could build a lot of transistors. A lot of those transistors would not work.
The opportunity that they saw was a high-reliability, production-oriented tester that helped companies sort out the good ones from the bad ones. You could actually use those transistors and diodes in high-reliability or high-volume products. That really enabled the growth of the entire electronics industry around solid state. Now, as you go through the years from then all the way up to today, that same dynamic exists in the semiconductor industry. New technologies, new processes, new packaging technologies are continually developed, and test is used to enable those processes to be used in products before they are perfected. Today, many of the characteristics that define today's ATE were actually innovated by Teradyne along that way. The trend that started early on and is continuing today is very apparent in the compute market, where it's putting pressure on which process to use.
Advanced packaging techniques, chiplets, HBM, and soon a transition to co-packaged optics are pushing the need for test. Without test, the yield of these processes would essentially be zero. There's a continual process to try to catch all of the defects during that production process, and that's where test comes in. At the edge of what's possible, test is vital, and that's where we fit in. Teradyne was founded as a semi-test company, and we grew and we evolved. We parlayed our abilities in semi-test into other parts of the test market, and then we parlayed that into advanced robotics, a segment that we believe could drive secular growth for the next couple of decades. At its core, Teradyne remains a semiconductor test company. In 2024, 75% of our revenue came from semi-test. In 2028, it's going to be 74%.
It is at its core a semi-test company, and that's the way it's going to be continuing. Now, I'd like to continue the history lesson here and talk a little about the changing dynamics of the semiconductor test market. If you look back into the 1990s, on this graph, the squiggly line at the bottom is the size of the ATE TAM market, and the lines going across are companies participating in that market. You can see over the period of time from 1990 out through 2010, there's a considerable consolidation in the semiconductor test space. What happened over that period of time was that stronger competitors acquired weaker competitors, other people left the market, and what ended up is essentially a duopoly between Teradyne and Adventest.
With that structure, there's sufficient revenue for the participants to fund adequate R&D and to develop good earnings and for us to maintain growth in that space. It has turned from this unhealthy state to a much healthier state now that we think is going to persist over time. Rick will talk a little bit more about that dynamics when he gets up here. What we see in semiconductor test, we think, is playing out again in the world of advanced robotics. Back in the early 2000s, like before 2010, when the concept of cobots was invented, there were not very many players. Very rapidly, there were new entrants coming into that space, many different competitors. We have seen this picture before. This is an early-stage industry. Early-stage industries have tons of potential, but frankly, they are cyclical and they are nasty. There is undisciplined price competition.
There's people that are trying to compete basically on price versus differentiation. There is a lot of product experimentation. People are trying to figure out what solutions work best. Finally, as the installed base grows and these robots are put into critical processes, the support and distribution required for them changes. We have seen this movie before in semi-test. Having witnessed that movie before is really guiding the way we look at the robotics business now. It has helped us to set some principles that we're using to manage the robotics business. I'd like to share them with you now. The first principle is that we're in a marathon, not a sprint. Even revolutionary manufacturing technologies take time to be absorbed. The process of change in the manufacturing space tends to be slower. There are capital approval cycles and everything else.
In order to succeed in that world, there are two things that you really need. One is you need to set yourself up so that you can make consistent R&D investments across the cycle. You need to be able to maintain momentum in your new product engine so that you're able to keep at the edge of what people are looking for. At the same time, you also have to set up your cost structure so that you are going to be generating positive operating margin. The only way you can sustain as the market develops and is cyclical is if you set your cost basis at a level that you can sustain and doesn't become a drag on your earnings. That's the key to how we have restructured our robotics business. In our January call, we talked about resetting our cost basis there.
We moved our break-even revenue from $440 million to $365 million. In doing that, what we're trying to do is make sure that we're going to be able to deliver positive operating margin, break-even or better, even if the market does not deliver growth because of end market conditions. We are trying to set ourselves up so that we are going to be above the break-even level as we go forward. In addition to that, the second principle that we're using is that it's all about customer success, not just leading technology. Even if you have fantastic technology, customers are not going to incorporate you into their critical processes unless they can count on the equipment 24/7, 365. Our support model for our robotics business has to be able to deliver that level of success.
The good news is we've been doing that for decades. We know exactly how to do that from our semiconductor test business, and we're going to be leveraging that to both develop the model and also to help cost-effectively scale the infrastructure to be able to do it. Now, the third thing is differentiation. There are a ton of companies that are building plain old cobots and AMRs now. We have consistently delivered higher gross margins than any of our industrial automation peers. The reason we have been able to do that is really around product differentiation, the ecosystem that we've built, and the partner network that we've built.
We have a platform that people are building on top of, and that platform is helping us to command a premium that allows us to capture these higher gross margins, and we expect it to in the future. We are working very, very hard to establish a number one position in terms of the application of AI in robotics. And if we're able to do that, and we have a lot of confidence that we're on the right track, we're going to be able to maintain that moat and maintain that differentiation and gross margin. When you look at those three principles, I think it's a fair question to say, like, how are you going to judge success? Like, how are you going to know if this is all working?
Here are the two things that we're going to be looking at going forward to try to judge whether this is working or not. If you look at our history over the nine years that Teradyne has been in the robotics business, we have outgrown the industrial automation peers in eight of those nine years. Sometimes by a little, sometimes by a lot. In one year, we didn't. We're going to judge success by being able to outpace the growth of those peers, and we want to see that growth accelerating relative to those peers over this midterm. That's criteria number one. Criteria number two is that we're here to make money. Having this growth happen without it translating into operating margin is not particularly interesting.
The principle that we're doing is, as we achieve growth, 50% of that growth needs to come to the bottom line. If we grow 10%, we want to constrain our OpEx growth to 5% so that we're continually increasing our operating margin. I want to be clear. Our goal is to maximize long-term shareholder value. We see that there are synergies between our test business and robotics, and we believe the changes that we've made will help us achieve that goal. As a company, we regularly review our portfolio, and we believe that the revenue growth and the operating margin growth that we are looking at in this midterm from robotics is going to multiply the value of that business. It is going to be significantly more valuable at the end of this midterm than it is right now.
We are going to continually operate this business efficiently, and we're going to be looking at these success metrics. If we are off track against those success metrics, we're going to make the needed decisions to maximize enterprise value. With that out of the way, I'd like to go into some long-term trends. Before I get there, as Traci said, we have a lot of new news. I want to just touch on some of the new news that has come out over the past couple of days to give you my take on it. First, after close of business yesterday, we announced that we are planning to acquire Quantifi Photonics. CEO Iannick Monfils is over there. We expect to close this transaction in Q2, and this is a sign of how seriously Teradyne is taking the silicon photonics and co-packaged optics space.
Quantifi has strong customer relations in this space. They also have leading products, and their leading products are very focused on the key thing that's going to be happening in the future of this market, which is high channel counts and production-oriented testing. With Quantifi, we believe we'll be able to accelerate our roadmap of those solutions, and it will help us gain share in the high-performance compute space. Moving right along, in Q1, Teradyne has created the product test division. The product test division is collecting a set of businesses that were reporting in a different way before, but are now increasingly synergistic. It includes LitePoint, our defense and aerospace business, our production board test business, and when we close the transaction with Quantifi, it will include Quantifi Photonics as well. Those businesses have a significant amount of customer synergy.
The same ODMs and CMs that are building the wireless modules that LitePoint tests, the printed circuit boards that our production board test group is testing, they're also going to be testing optical connections, and they're going to need the things that Quantifi does. We want to make sure that all of those sales efforts are well coordinated. The other thing is that we want to make sure that we have a high degree of synergy between the product test division and the semi-test division. We already use LitePoint products in many of our semi-test solutions, most notably for like the ultra-wide band market. That's all LitePoint technology that we use for the instrumentation. We're going to be doing exactly the same thing in the optical space. We want to make sure that there is a great relationship between the product test division and the semi-test division.
That guided our choice of leadership for that division. You'll be hearing from him later. His name is Regan Mills. Regan is, right now, before this change, Regan was the manager of the SOC business unit inside of semi-test. Regan and his team helped to engineer the pivot that resulted in the three and a half X increase in our compute revenue in 2024 over 2023. Regan has a great understanding of the whole compute market, the silicon photonics market, and he'll bring a great attitude in terms of managing these teams together in the future. Now, the last part of this is the news that I hoped I would never have to give. One of my favorite people inside of Teradyne has decided to retire.
Rick Burns has been at Teradyne for one year less than I have, and he was, in many ways, the he ran the engineering organization when the engineering organization both came up with the UltraFLEXplus and also developed the strategy to work with Next Test to embed ASICs into that architecture that gave us significant differentiation. Rick has added a ton of value to the organization over time, and it's kind of with a heavy heart that I understand that he wants to move on. He's leaving big shoes to fill. Looking at what we need in terms of leadership in semi-test, this is our flagship business. We needed someone that was going to check a lot of boxes.
We wanted someone that understands the direction of process technology, somebody who had an in-depth understanding of the future of packaging technology, and somebody who understood how customer requirements are going to change in automotive, in compute, and in mobile. We found someone in Shannon Poland who checks all of those boxes and more. Shannon's in the back of the room there, and he joins us. He had a long career at Intel. He was involved in client compute there, and he was the Chief Operating Officer for Altera. He's been in the room for all of the product decisions, all of the manufacturing strategy decisions that really guided how those businesses were going to run. Basically, the future of semi-test is really in AI and compute and how compute is going to pervade both the compute space, the mobile space, and the automotive space.
We wanted someone who knew compute inside and out, and that's what Shannon is delivering. The transition of leadership for semi-test is going to occur in Q2, and we wish Rick well on his retirement. Rick has made the plans, and now Shannon has to deliver them. All right. If you look at sort of the post-news Teradyne, we have a semi-test division that's about 75% of our revenue in 2024. We have a product test division that's about 12% of our revenue, and we have an advanced robotics that's about 15% of our revenue, all in 2024. These three divisions can have goals and objectives on their own to achieve their results, but we see a future where they're increasingly synergistic with each other, and we want to make sure that we're set up for that. All right. Trends driving growth.
We have an ambitious plan in terms of earnings growth through this midterm. There are three trends that we think are going to reshape the world. Reshape the world in a way that is as fundamental as the smartphone changed the world in the 2010s. I want to try to make the case that Teradyne is positioned to benefit from all of these trends. The trends are verticalization, electrification, and artificial intelligence. Now, when I say these trends can reshape the world, they're going to redefine our business. They're going to redefine many of our customers' businesses. I would like to tell a little rowing story, if you don't mind. Back a few years ago, before I had this job, I was an avid sculler, and I raced a single scull. In rowing, there are two kinds of races.
There are sprint races where you go in a straight line as fast as you can, and then there are head races, and those are usually over a twisty-turny course, and they're a much longer distance, like four miles or something like that. My favorite was the Head of the Charles. There are people who make a study, like a PhD, in exactly the line to take when you race the Head of the Charles, exactly where you should be at every one of the turns. As a rower, I probably brought more enthusiasm than talent. I'm a short, stocky guy, and if you look at rowers, they're generally a little taller than I am. I put a lot of faith in that idea of trying to be in the right place at the time.
There is a saying that us old fat guys say, which is that races are won in the turns. At Teradyne, we believe the same thing. Change is a constant in this business, and we think that these changes play to our strengths. In the 1990s, Teradyne emerged from a crowded field as one of two winners in semiconductor tests. In 2009, we emerged from the brutal downturn with a 1,000 basis point improvement in our operating model. In the 2010s, Teradyne pivoted to capture the first opportunity of vertically integrated platforms in the mobile space. In the early 2020s, we have pivoted now to capture the transition to vertical integrated platforms in the compute space. In all of these trends, I want to emphasize that this is the thing that we're trying to do.
We're trying to pivot the company to make sure that we are going to be advantaged by the changes that are coming. First, verticalization. I had suspected that there was a concentration of enterprise value at the very top of the S&P 500, but until I started asking all of the AI agents about it, I did not really realize exactly how large that concentration was. This is all data as of the end of 2024, but at the end of 2024, the whole S&P 500 was worth about $50 trillion in enterprise value. If you look at the top 12 companies, that is $23 trillion of the $50 trillion. The other 488 is the other $27 trillion. So 46% of the enterprise value in the S&P 500 is in the top 12.
Now, of those top 12, nine of them are involved in the global deployment of AI. Let's call them the AI nine, just because that's fun. The AI nine, so six of them are vertically integrated platforms, and three of them are delivering the foundry, the design services, or the components to enable the other vertically integrated platforms. These vertically integrated platforms are in different markets: mobile electronics, automotive, search, cloud services, e-commerce. They are in different spaces, social media, but they all see that the development of AI is the pivotal thing to their future. They are racing to try to establish leadership and establish a position as a leader in the AI space, and they are spending crazy money to try and do that. If you look at the AI nine, Teradyne has a pretty powerful position.
We are a supplier to eight of those nine from semi-test today. In some cases, it's a 100% share. In some cases, it's split share, but we have a supplier relationship with eight of those nine. Our product test group has a relationship with four of the nine. In robotics, it's a little bit more complex, but we are either a partner of or a supplier to three of those nine. Now, if you look at just the name of vertically integrated platform, these companies are building chips, they're building boards, they're building products, they're building systems, and in order to do that, they need a lot more than just semiconductor test. They need to be able to test those products. They also need to be able to deal with extremely short product life cycles, which is an entry for advanced robotics.
We believe that we have opportunities in all of these verticalized companies for Teradyne to be a trusted partner and deliver services across our whole portfolio. There is no other company in the world that can offer this portfolio of products to this group of customers. Moving on to electrification. Electrification is something that has been going on for a while. It has been going on for years. There is an effort to build a more flexible grid that can incorporate more types of energy sources, deal with intermittent energy sources, more renewables, and also there has been a growth in electric vehicles and hybrids. The thing that is really igniting this space is the exploding power demand for data centers. I would like to give you a quick overview of how we see the future playing out here.
In data centers, if you look at the data, this is McKinsey data. In 2023, about 50 GW of power was going into data centers worldwide. By the time you get out to 2030, it's estimated that it's going to be more than 200 GW, so a 4X increase in the amount of power. What that is doing is creating a desire for power efficiency from the people that operate the data centers, the people who are designing the equipment going into them, and the utility operators. That is creating an opportunity for highly efficient power semiconductors, especially ones based on gallium nitride and silicon carbide. Moving over to automotive, automotive is in a doldrum right now overall. And we looked at the data in terms of how have the forecasts changed since 2022.
Now, this is from Omdia, and if you look at the 2022 forecast, which is the dashed lines, and the 2024 forecast, which is the solid lines, you can see that there's definitely a downward revision. The blue line is for internal combustion vehicles, and the green line is for electric vehicles. The internal combustion line is definitely down, the changes for EVs is much, much more subtle. If anything, it's accelerating the time when the crossover happens between EVs and gas-powered cars. If you think about it, there are three things that are impeding the uptake of EVs and hybrids. The first is price, the second is range, and the third for EVs is charge time. In terms of price, we know that all major automakers have mid-priced models in the pipeline that are going to be coming to market during this midterm.
The other thing that we know is that EVs and hybrids have about twice as much semiconductor content as regular cars. If you look at the forecast and the change, you can see that we have an expectation for growth in this space despite the changes in the end vehicle counts. One more point around range and charge time. The most powerful way to improve range and decrease charging time is through improvements in the efficiency of the semiconductors. Again, advanced materials like gallium nitride, silicon carbide are key to that. Recently, Teradyne announced the acquisition of a design team in Regensburg, Germany, from Infineon. That group is specialized in high-power discrete testing, and by adding that team into Teradyne, we believe we're going to be able to accelerate our plans in this space. What does that mean?
In 2024, Teradyne had about 46% of a $415 million TAM in power linear and discrete. Now, with the addition of the team in Regensburg, we believe that we're going to be able to gain share because you see the other wedges, see the big 24% wedge down there? That's others. That's tons of small companies. We believe there's an opportunity for share gain in this space. By the time we get to the end of this midterm, we think our share here is going to go up to about 55%, and we think the size of the market is going to grow to almost $700 million to $675 million. All right. Moving right along into artificial intelligence. Now, AI is going to affect every company. As analysts, I imagine you're like, "Oh, great.
Here's the AI section. I want to sort of put this into some context. AI is going to affect the way every company works in terms of their operations and everything else. For some companies, AI is going to make a pivotal difference in their products. It's going to change their products inherently. Other companies are going to benefit from the demand that's generated by AI in the cloud, and other companies will be benefited by the deployment of edge AI. Teradyne is one of the only companies that is going to be able to see benefits in all of those areas. Let me just sort of talk about how that works. In test, AI is primarily a demand driver. Today, it's in the compute space. We have seen some in automotive, but there's a lot more to come. It's also driving our HBM business.
Tomorrow, that's going to move to automotive, mobile, and storage. In robotics, this is going to change our products, and it's going to expand the SAM. AI lets us build smarter robots. It allows us to deploy more flexible solutions, and it enables a whole class of products that you'll see on the tour around mobile robotics. At the enterprise, like other companies, AI is revolutionizing the way that we do product development. It's already changing the way we do service delivery to improve customer satisfaction, and it's going to be used to leverage improved internal process efficiency. Cloud AI, I think that's today's story. There's a ton of investment going in over the next few years. Some of the numbers here are just mind-boggling: $500 billion from Apple, $100 billion in 2025 from Amazon, $80 billion from Microsoft in 2025, $500 billion for Stargate.
These are stunning numbers. What we think that is going to drive is a continued increase in the already large TAM for high-performance memory and for compute. Teradyne is positioned to benefit from that, especially from the business in the VIPs. Now, the edge story, cloud is now, edge is tomorrow. It's just getting started. Right now, in 2023, there were about 50 million smartphones that were AI-enabled. In 2024, that grew to over 200 million, but right now, it appears that AI is not delivering a compelling user experience in a smartphone. We believe that in order for there to be a genuine large language model operating locally, you need certain technologies. You need two nanometers, you need gate all around, you need higher performance memories like LPDDR6, and that is what is really going to ignite compelling features from AI in smartphones.
That is going to begin happening in 2026, and our view is that is going to increase the refresh rate of phones and also increase the complexity of the semiconductors that are going in there to drive larger TAM in the mobile space and also to drive recovery in the mobile part of the memory market. Okay. How does this turn into Teradyne? Right now, AI is already a positive impact on our business. We're seeing auto ADAS processor business that started back in 2023. We have AI compute and networking business that started in 2023 and strengthened in 2024. In the latter part of 2024, we entered the HBM performance test market, and we captured significant share at a leading supplier in that year. In 2025, at the beginning of 2025, we achieved first customer acceptance for system-level test of AI co-processors.
You'll see that system today, and that's going to be deploying over this year and into the future. There's going to be a heavy emphasis on solid-state drives in cloud computing. Going forward, we'll have a product line to deal with that in 2026. We have achieved acceptance for production board testing of complex server assemblies, and that's going to be ramping in 2025. With the addition of Quantifi Photonics, we expect Quantifi Photonics to help drive our business as AI accelerators start moving towards co-packaged optics in the 2026 timeframe. Moving to 2026, that's when mobile comes into the picture. We expect that that's going to drive larger TAM for mobile. We expect it's going to drive LPDDR6, and we also expect that it's going to drive protocol transitions in flash memories. We expect a growth in 2025 in the breadth of this again in 2026.
In 2027, we think that the data demands for AI-driven smartphones are going to be pushing new data standards, wireless data standards, and that's going to be an accelerant for LitePoint. The picture that emerges is AI is going to be this building impact to our business over this midterm. Now, robotics is a little bit of a different situation. It's kind of hard to wrap your head around how does AI make robots better. Let me just give you sort of a brief overview of two AI-based applications that are actually in the world right now. The first one is our new MiR Pallet Jack. The Pallet Jack uses AI to recognize pallets. What we did is we took an AI and we trained it with millions of real and synthesized images of pallets.
If you look at this pallet, you see the shrink wrap that's going around it. You see how it's dripping below the level of the pallet. That would actually stymie some products in the market because they would not recognize it as a real pallet. Because we've trained this on a variety of different pallet images, it is able to recognize pallets even if they're dirty, broken, suboptimal, or obscured in some way by gaining clues from the whole surroundings. What does that mean? It means that that Pallet Jack will be able to deliver higher mission success rates than Pallet Jacks that don't use AI. That's sort of the way that AI will make our MiR products better. If you look at the next case over on the other side, this is an AI-based solution that comes from one of our OEM partners at UR.
What Ocado did is they took the UR platform, and on top of that, they built an AI system that allows it to do random bin picking from a cube storage system. Without knowing what's in those bins, it's able to look in the bin, figure out what it is, figure out how to pick it up, and then pick it up and put it in the place that it needs to go, all in real time. These are things that are really hard to get a robot to do unless you use AI. We believe that they're going to expand the market for robotics in a pretty serious way. If you look last year, in 2023, we had about $1 million of AI-driven business in our robotics unit. In 2024, that ramped to $11 million.
By the time we get out to 2028, we expect that that's going to grow to at least $150 million as AI pervades the entire MiR product line and more and more of our UR partners begin to use AI to expand their markets. All right. Just to wrap it all up, here's the takeaway message. Teradyne is positioned to grow robustly through the period of 2028 based on verticalization, electrification, and AI. We have leading product portfolios in all three of our divisions that are going to drive balanced growth in all parts of the company. At the midpoint of our market, I'd like to illustrate what I mean by that balance. Robotics, from 2024 to 2028, we expect robotics to grow from $365 million to about $790 million. At that point in time, it would represent 16% of total Teradyne sales.
Thirteen grows to 16% of a larger total. Product test is going to grow from $330 million to about $520 million, not including any upward impact from the Quantifi acquisition. If it's at $520 million, it would represent about 10% of total Teradyne sales. Semi-test is going to grow from about $2.1 billion to about $3.7 billion. We expect growth across all of the segments of semi-test, whether it's the integrated system test group, whether it's mobile, compute, auto industrial, or memory. We expect all of those to grow in a balanced way. In 2028, if you look at semi at that segment level and then robotics and product test, no part of Teradyne is going to be contributing less than 10% of total revenue, and no part of Teradyne is expected to contribute more than 20% of total revenue.
What that means is we think we have valuable resilience in this model. We have many paths to success, and we have upside potential in all of these areas. The picture is 15% revenue growth over this period, and we expect that to take our EPS from $3.22 to $8.25 at the midpoint in 2025. We believe that Teradyne is positioned to benefit from the trends that are going to reshape the world. We're aligned to the largest and fastest growing companies. We have the right products. We have the right culture. We have the right business model for a complex and changing market. That's it for me. I would like to pass the baton to Rick for his very last presentation. Here you go, Rick.
All right. Thanks, Greg. All right. Thanks, everybody. Welcome to the home of semi-test.
This is where it happens, right? This has been the core business since the company was founded by these two guys with a very simple idea. Computer automation combined with test delivers throughput, cost, and quality to all of our customers. Despite the bit of a bumpy ride, Teradyne's been a trendsetter ever since. Now, I'm going to build on that short history lesson that Greg talked about for those who are new about how Teradyne led this test market from a break-even market to a healthy duopoly and set up the market dynamics that we see today. Now, prior to 2000, most of our customers were IDMs, and many of those IDMs built their own ATE. That dynamic, combined with lots of competitors, set up that unhealthy environment Greg talked about, making money in the upcycle, giving it all back in the downcycle.
By the early 2000s, we were down to six major competitors, four of which had about 20% market share: Teradyne and Mixed Signal, Advantest and Memory, Verigy and Digital, and others, mostly LTX and Analog. Teradyne focused on the mobile phone market, and we grew our share to about 40% by 2010. This touched off an attempt by LTX to buy Verigy. Now, those of you who study this know that Advantest eventually prevailed, and that left us with the duopoly that we see today. That was not it for Teradyne in the 2010s. We focused on establishing financial discipline to make money throughout the cycle. The way we did that was by outsourcing manufacturing, focusing on profitable segments, controlling our costs, and building value through product differentiation.
That was mostly through developing high-density instrumentation based on our own custom ASICs, exactly the model the vertical integrators are employing themselves. We combined that with multi-site software that gave our customers the productivity to be able to utilize all that instrumentation. By the end of the decade, our revenues grew to $2 billion and our profits to 25%-30%. You will notice at the end of the 2020 timeframe, there is some share oscillation going on there. That is mostly customer buying patterns. Big customers in mobile could swing the share as much as 10 percentage points. Under the hood, Teradyne was pulling share away in the memory market and in the VIPs with a simple recipe: compete head-to-head and win. You will notice now shares moved away from Teradyne a little bit in the last few years.
This is really due to two things. About a billion dollars' worth of business in China shifted away from us by U.S. regulations. Also, the growth of complexity in the computing market is driving the whole market in that direction, as Greg described. What I'm going to do is spend the rest of this presentation explaining how we're going to win this back with that age-old recipe: compete head-to-head and win. Greg mentioned the trends behind this complexity growth, and we're talking hundreds of billions of transistors, huge complexity like we've never seen before. Let me tell you what these mean to semi-test. First, in verticalization, for semi, this really means new customers looking to create unique value through their own ASICs, right? They have a higher focus on quality and brand impact, and maybe most importantly, they're willing to try new things.
In the electrification and energy transformation space, this is mostly about wide-band gap technologies. Why wide-band gap? More energy efficiency than regular silicon. That is because each switching device can hold off higher voltages than its silicon counterpart. That enables faster switching and smaller devices in the case of GAN, and it allows higher voltages and higher temperatures in the case of silicon carbide. Finally, GenAI, it is exploding across all the segments, bringing smarter, more adaptive features to enable devices and systems to take advantage of centralized data center training, but also localized learning. This is what we see in the trend towards the edge. We think this is going to move to the edge in mobile and automotive devices to cut the inference latency. If you are about to hit the curb, you do not have time to go to the cloud and back, right?
All of that is driving an enormous amount of data growth, and high-speed and optical interfaces are what are going to enable that needed data movement. All of this sets us up for our midterm outlook. Now, that set of growth drivers sets up the TAM forecast you see on the left. With continued growth at our historic 9% CAGR, that rolls up to a market size of more than $8 billion by 2028, driven by AI-infused ADAS and electrification in the automotive space, AI features in phones and new smart devices boosting mobile, AI accelerators and optical networking, growing computing and memory. Do not forget service, that little line at the bottom there. All this time, installed base continues to grow, and new services like automation fuel growth in this segment.
Now, before I get into the segment details, let's take a look at that last item. My semi-customers are really challenged finding talent. They're building lots of new fabs. You see this in the news. Those fabs are all highly automated. These customers have a large collection of existing facilities, and these have incomplete automation. Automation drives productivity, OEE, quality, and traceability, all factors they care about. These existing facilities have human-centric layouts. They're really tight. No elbow room, no room for overhead transport or conveyors. Enter the mobile cobot and the AMR. Together with Teradyne service personnel on nearly every factory floor, we think this sets up an enormous TAM, over $1 billion. My customers, they're asking me for Teradyne robots. Ujjwal is going to fill you in a little bit more on how we're going to deliver that through Teradyne automation services.
Now, let's get into the segments. Automotive and industrial, long a strength and the foundation of our diversified revenue and customer base. Now, Greg described the massive buildout happening in data centers, where the efficiency of wide-band gap is a key opportunity to manage those power demands. I'm going to focus a little more on vehicles, somewhat overlooked due to those tepid unit growth numbers that Greg showed. The combination of ADAS and infotainment, that drives what looks like a data center and a smartphone on wheels. You add on top of that powertrain electrification, whether it's battery or hybrid, and the content of the average car doubles from $800 a car last year to over $1,500 by the midterm. Now, put on top of that, life safety, long lifetime, and wide temperature ranges, all of which drive up the test time, got a very rich market for us.
Now, why does Teradyne win in that space? Because our ETS platform is purpose-built for the high voltages of GAN and silicon carbide. We have floating resources that allow full accuracy of the instrumentation at those high voltages, which is unlike the digital ground-based architectures of our competitors, where accuracy degrades as the voltages increase. We announced back in January the acquisition of Infineon's internal tester group in Regensburg. Let me tell you a little bit more about that. Infineon is a key leader in the automotive power semiconductor space. They've been instrumental in bringing wide-band gap to market, and they're forecasting strong chip growth in that space, more than two and a half X over the decade. Why? Because those technologies enable reduced charging time to levels that are similar to filling your tank with gas.
They also allow increased range by squeezing out the most performance of battery systems, including the monitoring of accuracy of those systems. Now, what did we get? We got an internal team of about 80 people making unique power testers. Infineon saw that aligning with Teradyne gave them all of those benefits plus the benefits of the ETS throughput and accuracy I described. For us, it allows us to increase our leadership, accelerate our roadmap, and consolidate share in this space. Why do we care? Because we anticipate that the wide-bandgap ATE market is going to double over the midterm, and our share is going to double as well. You can't talk about ATE without mobile. Once the dominant segment in test with almost half of the market, AI features are driving processor and memory complexity, as well as wireless bandwidth. Why?
Because latency and security-sensitive AI features need local processing and storage. You do not want all your personal data flying around the cloud. Apple Intelligence and similar applications are expected to drive shorter refresh cycles, increasing the unit volume. We also expect recovery in all the other technology areas in the phone through continued advancement in wireless standards, display technology, and integrated sensors. We win because the UltraFLEX family delivers the best performance and time to market and has for years, together with LitePoint for wireless cameras and processors. Our key VIP relationships in this space allow us to bring new solutions to market in emerging technology areas. For example, we are introducing ATE and SLT compatible mission mode testing to give our customers the opportunity to both improve quality and cost simultaneously. Now, why the focus on verticalization?
History has shown that owning the stack drives value. We all know Apple has less than 20% of smartphone units and more than half of the profits in that space. It's a similar story for Netflix and streaming and Amazon and retail and web services. More important to these companies than cost is differentiation from store-bought semiconductors. What those VIPs like Apple, Samsung, and Huawei did in the mobile space has been noticed by the computing and automotive VIPs like Google, Amazon, Tesla, Microsoft, and BYD. These companies do it all. They design the chips, the boards, the end products, and the services running on top of them. They're sophisticated software companies, not satisfied with traditional solutions. These VIPs want the whole package, not just semiconductor tests, but product tests, automation, and a software environment that ties it all together.
They want intelligent software workflows that enable small, nimble teams inside their companies to do it all, whether it's test solutions, data analytics, or yield learning. We have invested millions of dollars in AI-enhanced development tools and libraries for our .NET framework to speed these companies to market. Our expectation is that AI accelerators are going to evolve just like mobile, with about half the market vertically integrated by 2030. We use the measure shown on the charts here of transistor count or complexity rather than just units because these VIPs tend to build premium products, which really drive the TAM. You might be a doubter, but just in December, The Wall Street Journal quoted Amazon's data center lead that 50% of the CPUs they've deployed in the cloud in the last two years were internal silicon. That is faster than our forecast for adoption.
Today, we currently have AI accelerator production running on over 250 UltraFLEX and UltraFLEXplus machines. It's happening, and it's happening now. Speaking of which, computing, the booming growth area. Now, before 2022, this was kind of a sleepy segment, right? Less than 10% of the TAM, mostly x86, a few GPUs and networking, Intel dominated by using one of the last homegrown ATEs left in the marketplace. GenAI is now surging, mostly NVIDIA GPUs right now, but pulling through a lot of networking as well. The VIP accelerators are just now starting to ramp. High test time due to full reticle transistor counts of these devices and multiple chiplet packaging creates a huge potential.
Their focus on power efficiency complements that wide-band gap discussion we just had, and they drive a data bandwidth explosion that's going to push up networking and shift us towards silicon photonics and co-packaged optics. Now, why? Because silicon photonics offers about 10 times the bandwidth with about 30% less data center power. Now, we're going to win in this space because all of that massive amount of complexity is outstripping the incumbent tester infrastructure, driving our competitor back to single-site testing. We haven't done that in years. The UltraFLEXplus architecture, based on advanced networking, deep memories, has the capability to deliver thousands of amps to power these kilowatt devices and cool them as well. It can handle multiple sites. It also supports advanced interfaces like PCIe Gen 6, HBM, and soon co-packaged optics to support real-time data moving in.
It also has an architecture built with distributed computing to be able to keep up with the data manipulation, data logging, and calculations associated with all of that test data. Together with Technoprobe, we're going to bring to market later this year several of our joint development projects, which are deploying new capabilities impacting us financially in 2026. Along with this, we have Titan HP. You'll see it on the tour later today. This brings our proven asynchronous architecture for SLT to the computing space to find software and timing-induced bugs that escape ATE testing. Now, clearly, Adventest has incumbency in this space, but there are two disruptions that are opening the door. First is the VIPs that I talked about, where we believe today we have more than 50% of the installed base loading. The second is silicon photonics and co-packaged optics.
Regan's going to talk about those in more detail. Now, finally, memory. Booming along with AI computing, HBM's finally ramping. The quality of HBM stacks is still a leading yield issue for complex AI packaging. It drives more performance tests, which is ideally suited for the near-DUT test architecture of the Magnum systems. Magnum 7 delivers both the highest performance and the highest site count for both DRAM and flash tests today. Now, memory is really an interesting share gain story, and it's an example of why I'm confident that we're going to gain share in computing as well. Back in 2007, Teradyne was barely in the memory test business. Now, let's have Young Kim, President of Next Test Memory Business Unit, tell us a story. Young.
Thank you, Rick. Good afternoon, everyone. I want to show you three graphs.
I want to show you three graphs today in this slide: Teradyne memory revenue and memory AT10 and Teradyne memory market share. Since Teradyne re-entered the memory business in 2008, we have achieved remarkable eight-fold revenue growth in 16 years. We had a record revenue with a record profit last year. We believe that this upward trajectory will continue driven by memory AT10 expansion and our increasing market share. AI is boosting memory AT market significantly. If you look at the graph, during the decade of smartphone era, memory TAM stayed at about $500 million level. Memory chip makers built only seven new memory fabs worldwide during that time. However, in 2017, the rise of big data fundamentally changed the memory AT landscape, doubled memory TAM to $1 billion level. Last year, AI surged memory TAM to more than $1.4 billion.
We believe that this memory ATE will continue growing because of three key factors. Number one, more than 30 memory new fabs are being built to meet explosive memory chip demand from big data and AI, compared to only seven new memory fabs during smartphone era time. Not only four times more memory fabs, but also two to three times bigger fabs. Number two, the rapid pace of technology transition, like new HBM device every year: HBM3 last year, HBM3E this year, and HBM4 next year, and so on. Also, LPDDR5 is moving to LPDDR6 for mobile phones, as Greg mentioned. On the NAND flash side, UFS 4 is moving to UFS 5 for Android phones, and there will be new PCIe generation devices for non-Android phones. Also, high-speed NAND flash speed is increasing rapidly for enterprise SSD.
All this new, every memory device generation will require new system purchase or retooling existing systems. Lastly, increasing memory device speed is decreasing memory test parallelism by half. That requires a lot more testers. A lot more new fabs and then technology transition and then memory test parallelism, combination of these three key factors will generate a powerful engine for the memory ATE growth. Last one, Teradyne memory market share. We have dramatically increased our memory market share since 2008. The formula to our success with our memory market share is the invention of game-changer products based on our superior AC technology and our system integration technology that's far better than our competitions. Every time we introduce the new product, we increase the market share significantly. Let's look at it.
We start with a flash final test segment, and then we increase our market share only from 4% to 16%, about 12%, by testing twice more devices in parallel against competition at similar cost. We expanded our business into wafer-sort business, increased another 12% market share to average 28% by delivering a lot much better test time against the competition. We penetrated the DRAM final test market with a Magnum Epic product and increased our market share over 40% by delivering better device yield for our memory customers. All these innovative products have enabled us to increase our market share from 4% to over 40%, more than tenfold. Last year, we lost some market share because our competitor got the HBM business first with their existing product.
However, we quickly responded with another game-changer product, Magnum 7H, launched mid last year. We already secured a major HBM supplier in the second half of last year by reducing their HBM wafer test time by 40%. Last week, we received a first PO from another major HBM supplier for HBM4 production testing. These two major HBM design wins will help us to achieve another significant market share increase. With this innovative product and with our continuous innovation and customer focus, we are very confident that we will be number one memory ATE supplier. Thank you, everyone. Back to Rick.
Thanks, Young. I hope you picked up on my excitement for this space. Everything I shared today is why I expect robust market and share growth across the midterm.
Now, as we announced, I'll be retiring soon, but you can bet, as a big shareholder, I'm going to be watching this very closely. The market I've seen over the last 45 years has been full of intense shifts in competition, that's what makes it so interesting and also makes it full of potential returns. We will see the recovery of the major mobile customers and the growth of automotive electrification. I really have confidence that Shannon and this very strong semi team will continue to win share in compute and memory. I don't think it's going to be very long before Teradyne regains market share leadership. I look forward to answering your questions a bit later on, but for now, let me pass it along to Ujjwal.
Thank you, Rick. I've been with Teradyne for 18 months. I'm Ujjwal Kumar.
My last 26 years have been in the industrial and industrial automation space, from GM to GE to I was running the industrial automation business of Honeywell. I'm so excited to be leading the world's largest advanced robotics portfolio, which Teradyne has built. As you know, we have some work to do to make this a profitable and predictable business, like what you expect out of any Teradyne division. That is what I'll walk you through over the next 20 minutes and what all actions we are taking to take that to the next level. The robotics market, it's not a futuristic dream. It's a present-day necessity. You know the demographic changes on both the directions.
We have working-age population going down and the 65-plus age population going up who need a lot more support in the hospitals and assisted care, which has put this double whammy on the need for working-age population, which is increasing the labor shortages. In addition, to attract this new generation of workers, we need to completely redefine the way we work. That is, this labor shortage combined with the urgent need to change the old way of working is what is driving the growth for this new generation of advanced robotics. You know, I'll touch upon that later today on how is this advanced robotics different than the robotics which you're used to. This AI-integrated advanced robotics, which we plain, it is enabling the growth need for the countries and companies. You see this big number, $220 billion in front of you.
It's pretty impressive, but let me break it down. Around 60-65% of this space is service robots, these last generation of AGVs, the autonomous guided vehicles. You have humanoids trying to get into some of the service operations, but that still leaves around $50 billion-$70 billion of space just for the kind of solutions which we have, which is the cobot and the AMR and everything in between, which is a really good space for us to drive the kind of growth which we have committed over midterm. I was telling you about our unique portfolio. We are the global market leader. When you simplify this advanced robotics space, it is really to help a manufacturer or any kind of application do things which are done by human beings in a more efficient way.
There are in a factory floor, you'll see three different kinds of people: people who are in a cell doing something, people who are moving things from point A to point B, and then you have hybrid workers who do something at one place, move things around, and do something else again. We, as Teradyne Robotics, we have the unique portfolio to have all three. What makes us this market-leading platform is the ability to bring in this leading-edge AI technology, service offering the largest ecosystem in the entire portfolio of advanced robotics. When you look at what makes us unique in the market, number one is the customer experience. We are seen as the easiest to buy, deploy, and operate. The total cost of ownership for our solutions is pretty attractive. We are seen as the most reliable and scalable platform, including the largest ecosystem.
The third one is when a customer comes with us, they know that they will get the latest technology. It's future-proofing themselves. You know the AI and robotics space is changing and evolving at a very fast pace. Customers, when they are transforming their operations around an automation solution, want to make sure that this productivity they are getting or efficiency they are getting is what will keep them competitive in the long run. This slide tells a critical story about our performance in a very dynamic market. The blue line which you see there shows Teradyne Robotics. The gray is the average growth rate of the peers which we regularly track and compare against. First, we have to acknowledge the external factors that have influenced the market.
Initially, strong growth of 20% plus, which you see here, as Greg was mentioning, we outperformed our peers in eight of the nine years in the past. We had these market forces which affected how everybody did in the market, starting with COVID, which gave high inflation, and then we had these wars, and it looks like the market uncertainty has not gone. Like you read the news every day, it's something which has affected the CapEx in the industrial automation space despite such a glaring need to address our labor shortages. Of those last nine data points which you see here, in 2023, we had a disappointing year. In 2022, we had started our channel transformation, which took its own time to show up in P&L, and we were able to completely turn it around in 2024.
We do see those OEMs which we had signed up started showing up in our pipeline and in orders book and in P&L, which is one of the key ways in which we see us expanding into newer segments. This turnaround is a proof that our investment in innovation and market expansion, both through launching of new products, which I'll touch upon, and this new channel is paying off. Our ability to outperform even in a down market, like what we have been in the last few years, demonstrates the resilience of our strategy and the strength of our combined team's execution. As I highlighted, we didn't sit still during the recent downturn. 2024 was a very busy year for us, as you can see on the chart here. In early 2024, we strategically reorganized both our R&D and the operations team.
We focused R&D with specific competencies to rapidly bring new products to market, and we realigned operations to launch those products with high quality. This reorganization, combined with our strategic investments, paid off handsomely. We delivered a record number of new product introductions, which you can see here. We had to really squeeze it on a slide to show you the whole thing together. I feel like our team, they have become an NPI machine, which we can use to continue our growth as we go into 2025. I look at these launches in two ways. One is it allowed us to expand our served available market quickly. Those new products got us into new segments. It also helped us monetize our install base. We have more than 100,000 robots out there.
These new offerings allowed us to add a new service model, which Greg was talking about, to our portfolio. The second one is around differentiation. We command the highest gross margin in the market, and that market-leading gross margin is driven by us seen as the leading in technology and the platform advantages which they get. When they buy our robots, they do not get just the robots. They get the whole ecosystem. It is like you buy your iPhone, you get the whole iPhone store with it. When they come with Universal Robots or MiR, they get this platform on which you have the highest number of AI-enabled innovators developing all kinds of solutions on it. The market-leading platform combined with services and ecosystem is what allows us to command that market-leading gross margin.
Let me now show you the diversification of our portfolio, which allowed us to outperform our peers, as you saw in 2024. On the top, you see the highlighted area. Those are the growth-focused areas. The four segments on the bottom are the core segments. In the core segments, we have seen some stagnation. Over the next four years, we expect low double-digit growth. That is how the advanced robotics market in general is expected to grow. We will be able to outperform the market exactly the way we have done in 2024, which is the focus on those new, faster-growing growth areas, which gave us growth from 2021 to 2024. This is life sciences, logistics, new emerging segments like construction and agriculture.
These days, with labor shortages becoming even more stark in the U.S., Europe, and different parts of the world, we learn about new areas where people want to use robotics to address their labor need even more and more. Over the next four years, we will add two more segments to our market expansion game plan. One is food and beverage and semiconductors. And semiconductor is an area where we, as Teradyne, have this unique advantage in terms of customer relationship, as well as what Rick was talking about. We have joined our expertise between the semiconductor and the robotics side to create a unique solution for the semiconductor customer through this team we have created called Teradyne Automation Solutions. So, as you think about how will we expand into these new segments, launching of those new products, which you saw on the earlier slide, is a piece.
The second one is what you see on the bottom right corner here. The new solution providers, new OEMs, global system integrators. We also have started working direct with some focused customers. While we will leverage our industry-leading distributor network, our distributors work really hard to get us that differentiation in the market. While on the geographical split, I'm not expecting things to change too much from now to 2028. 80% of our revenue comes from Americas and Europe. That is where we see huge labor shortages, and we expect the need to be even more as the year progresses. 80% of our revenue will continue to come from our core regions of Americas and Europe. On the channel mix, I'll highlight that this 23%, we expect that 23% coming from OEMs and direct large customers.
We expect that to grow to 40% plus over the next four years. Taking that to the 2025 strategy, building on the strategic diversification, which I talked about on the previous slide, our 2025 strategy is centered on continuing this market outperformance through three key pillars. Number one is we'll maintain the momentum on new product launches, which is giving us SAM expansion as well as differentiation, and the channel transformation, which is enabling expansion into these new segments. Number two is we will continue to make strategic investments in AI and these new segments. AI is one of our key positioning in the market. We are the market-leading AI platform in the advanced robotics space. Number three is consolidation of sales, marketing, and services.
Several of you were asking me this at the start of this session today on how did this reorg we just did affect the strategy for Teradyne Robotics. Our strategy doesn't change. It just gave us a new way to further improve our commercial effectiveness. What you see here, 50% of the revenue for both UR and MiR came from these common channel partners. We had one-third of our top 20 customers; they are the same between UR and MiR. Right there on the slide, you see Toyota. If you have a UR salesperson going to Toyota and a MiR person going to Toyota, it makes practical sense to have one account manager developing the overall UR plus MiR go-to-market strategy. This was a way in which we could drive cost efficiency while improving our commercial effectiveness.
It has been very well received by the market as well as our employee base. It simplifies our go-to-market model, while we can refocus those resources in those growth areas. As you saw, we have several of those we are focusing on. Now let's talk about what makes Teradyne Robotics truly special and gives us a unique competitive edge in the market. It starts with our industry-leading hardware, which is standard, intelligent, and reliable. We have shipped more than 100,000 cobots. For those of you who follow us on LinkedIn, you must have noticed it's a very proud moment for the advanced robotics industry for reaching that milestone. Our nearest peer in the advanced robotics space is less than one-third the size of ours. I think it's just the start.
You have seen all kinds of bets going on on LinkedIn on when we'll reach the million-unit mark. This is an industry which is ready for that scale-up. For rapid scale-up, you cannot do custom work application by application. You need to have a standardized, scalable platform on which you allow an open-source and configurable software on top and let the innovation engine of your ecosystem just work on it and come up with newer ideas. That has been the core architecture for UR and MiR from the beginning, which is what allowed UR and MiR to build this largest ecosystem of robotics innovators on the planet. We have this army of UR plus partners, OEMs. These are original equipment manufacturers and global system integrators on the MiR side.
They are able to develop their solution so quickly because they have these building blocks, which not just UR and MiR develops, but other ecosystem partners develop, which you can mix and match to solve customers' problems in newer ways very quickly. That gives us those partners, which, again, if you're following us online, you would see earlier this year, we announced the partnership with Analog Devices, so ADI in semiconductor space, Matthews in the logistics space, NVIDIA and Siemens, they have been our technology partner to expand us into AI-based solutions in several new end markets. We have a long pipeline of other collaborations we have been working on. I'll show you how NVIDIA thinks about us in two minutes.
Before that, let me just end with my final slide here on when your core strategy is to expand through these new verticals, primarily because we have several of those. The market is rife for coming up with new applications. We do not need to create new products for those new applications, as you saw in the earlier chart. We have our framework through which we expand into these new segments. We have four key criteria for selecting a growth segment. Number one, it must be growing faster than our legacy segments. As you saw, our legacy core segments, they will be growing low double-digit. We want to pick areas which are growing much faster than that so that we can get 3-7X versus our core segments in each of those markets over the next four years.
Number two, it must be a segment which is facing long-term labor shortages because that is our core hypothesis. If you pick areas where there are real labor shortages, there'll be long-term resilient growth. Number three is it must leverage our core strength in AI, in our expansive ecosystem, our strategic partnerships, and the services capabilities. Number four, we want to be in a segment where we should be able to create at least more than $50 million over the next four years. When we expand into these new segments, we have a platform approach to the product, the hardware, the software, as well as our go-to-market model so that we don't invest custom work or very specific thing in a big way just for that specific segment.
On the product side, anything which we come for, say, logistics or for food and beverage, it's a derivative of the product which we have developed for our core market. That is the same model through which we expand on the services side and developing the go-to-market side. Overall, we are expecting, as you can see here, 4X growth on the two segments which I'm highlighting here on the semiconductor and the logistics side. Before I introduce you to the NVIDIA video, which I just talked about, and in the break, you will also see a video by Siemens. You would think, why do these big companies like NVIDIA or Siemens, they can work with any player in the market? Why do they dedicate their engineers to work with Teradyne Robotics? It's not just because we are the largest ecosystem and the largest advanced robotics player.
It gives them immediate access to this big ecosystem. It's also our core architecture on which they feel like this open platform allows them to innovate at a much faster rate than any other player in the market. They'll be able to solve customer problems and bring in new solutions at a very fast pace. You will hear in the video, the NVIDIA video, you'll just see our market-leading AI integrated portfolio and the world's largest ecosystem of innovators, how that allows us to outperform our peers, as you saw in the numbers, and how it'll continue to keep Teradyne Robotics as the world's largest advanced robotics player. With that, we can go to the video.
The need for robotics and AI into the physical world has always been there. The question is, what's different now and why now?
My name is Deepu Talla, and I lead robotics and edge AI at NVIDIA. Robotics has always been the ultimate incarnation of AI. If you think about it today, robotics is mostly deployed in a highly automated manner, which are pre-programmed. Those happen to be high-volume, low-mix applications. When you talk about the future, you're talking about low-volume, high-mix type of applications. The moment we go into high-mix type of scenarios, it's unstructured environments, and the automation does not really work because the cost of setting them up is rather high. When you take that high-mix capability, whether it's low volume or high volume, we can use AI to increase the deployments. I believe the market's 10 times bigger. You're still in the first 10% of this journey to solve these robotics problems. The partnership with Teradyne happened a little over a year ago.
Teradyne Robotics was a perfect ecosystem partner for NVIDIA. In fact, the DNAs of NVIDIA and Teradyne Robotics are very similar. Universal Robots is one of the most broadly adopted cobot arms in the world. Almost every researcher in academia uses a UR cobot arm. Almost every company doing prototyping uses a UR arm. We have a deep technology collaboration with Teradyne Robotics. We have regular meetings across our engineers and leadership teams to make sure that the technologies we are developing with AI and the technologies that Teradyne Robotics have, we're carefully integrating and optimizing it together. The whole robotics and physical AI revolution is going to be happening in the next 10-20 years. The partnership with Teradyne Robotics I expect it to be deep both technically and also in the go-to-market in the coming years.
Perfect. With that, I'll hand it over to Regan.
Thanks, Ujjwal. I love that video. It actually exemplifies everything that is Teradyne and partnerships. It runs through our entire organization. Welcome, everyone. I'm Regan Mills, and after 27 years being here at Teradyne, I'm really excited to be here to represent the product test division. While it's true that our founders, Alex and Nick, started this company with the idea of automated test for semiconductors, it didn't take long before people realized that you need to ensure the quality of the entire electronics assembly through the entire stack, all the way out to the product. Over time, what Teradyne did is, through organic development as well as through acquisition, we built up test businesses that do just that. They serve the assembly and product test. Now, what we're doing is we're pulling all of these businesses together into the product test division.
That allows us to take advantage of the synergies that exist both on the customer side as well as on the technology side. Now, you've heard a fair amount about AI today, but I am obligated to talk some more about AI and what it means in the context of product test. AI is in front of us. Everybody knows it. Cloud computing is everywhere. If you look a little bit further out, as Greg was talking about, in the consumer realm in particular, the impact of edge AI, we're just really starting to see the opportunities here. Whether you're talking about an assistant on your phone or if you're driving your vehicle, braking decisions, lane changing, ultimately even full routes, these opportunities are out in front of us, and they're going to grow, particularly on the edge.
Now, if you look at how AI is actually implemented, yes, there's tons of software, but ultimately it ends up in hardware. AI is really complex. It takes billions of nodes and decisions, micro decisions that are being made to make a macro decision to whether you brake your car, whether you add something into your phone. All of these micro decisions rely on the hardware, and we cannot allow AI to be making bad decisions because of faulty hardware. This is why the need for quality in an AI world is only increasing. That's where Teradyne's product test solutions provide a complete suite of solutions that ensure that the semiconductors, that the ATE from Teradyne's ATE, known good semiconductors, are going into assemblies. Those assemblies need to be made sure that those are high quality.
Our solutions are aligned to making sure that that middle layer of hardware is always good all the way out to the product. One thing I will say is Rick was talking about the impact of process node technology and things shrinking and how that leads to increasing numbers of dynamic types of defect models. The same is true in these assemblies. Everything is getting smaller. As things get smaller and tighter, the types of defects and the opportunities for defects increases. The necessity of product test is only increasing in this AI world. That is not the whole story. We talked about verticalization here. If you look at each one of these segments, the largest players and the players gaining share are the ones that have figured out that it is not just about the software or the end product design.
It's about managing the entire stack all the way down to custom silicon. Now, why do we care about that? We care about that as Teradyne because we are the only company. We're the only company that offers a complete set of test solutions from semiconductor wafers through final test, assembly, and end product test. That gives us this unique ability to have incredible synergies between the semiconductor test group and the product test group. That gives us access to customers that we can penetrate from multiple locations and expand our footprint within those major accounts. What I'd like to do is take a moment and give you an example of how this plays out in this disruptive technology called silicon photonics.
Silicon photonics, if you're not familiar with silicon photonics, think of silicon photonics as optical components, devices, modulators, waveguides being fabricated in standard silicon wafers using standard semiconductor processing equipment. What that does is it allows us to get the cost of photonics down, allows the industry to scale the volumes up. All of that is an enabler. On top of that, the fabrication technology is one technology. The other one is packaging. We're going to talk about co-packaged optics. This is putting optical elements into a single package with large ASICs. Why do these technologies matter? These technologies matter because the data centers are consuming a ton of energy. The world cares about that energy consumption. Silicon photonics, when you look at an AI data center, there's a ton of nodes that need to communicate with each other. Lots of processors.
For a macro decision, you need to move data around the data center in a much greater way than general purpose processing. The number of switch elements and the number of data transmissions is going up. An AI data center that's re-architected to take advantage of silicon photonics can reduce the power consumption by 30% and achieve greater performance because the latency and the bandwidth can increase tenfold. This is an amazing combination of higher performance at lower power. That's why the industry is interested in pursuing this aggressively. What does this look like in terms of a transition? This graph on the right is something that really shows a transition scenario to co-packaged optics or CPO. The first thing I want to point out, though, is what I referred to on the prior slide.
This verticalization trend, optimization of the silicon, these are AI accelerators. These are devices that are specifically designed to accelerate AI in the data center. The trend has already started. It's only going to continue. What we're talking about here is that transition to take advantage of an AI accelerator using co-packaged optics. Now, over this midterm, the volumes are going to start. They're going to start during this midterm. The more important thing and the pertinent thing this year is decisions are right in front of us. Companies are deciding now which partner they want to have along this journey and which equipment they're going to use. Silicon photonics is an enabler for this growth in this transition, but it's a necessary but not sufficient condition because test is also essential. There are some real test challenges here.
First, you start with a wafer, a photonic IC wafer, or you may hear it called a PIC. This is a wafer that has photonic elements on it. In order to test this, you need to be able to align a fiber array to within 50-100 nanometers of precision to that element in order to capture the light and stimulate the device properly. That is 1,000 times greater than what we do in the electrical world in terms of accuracy. Challenging active alignment problem. You take that wafer, these photonic elements, you combine them with high-speed electrical devices into an optical transceiver, also called an optical engine. This is the device that basically takes information from the electrical world back to the light world and vice versa.
High-speed, high-performance device, challenging combination of very high-performance electrical instrumentation and high-performance optical instrumentation in the same tester. You take these optical engines. You have a massive data switch or a processor, and you surround it with these optical engines in order for it to communicate with the outside world. Now, you have co-packaged optics. Each one of these optical engines might have 8, 16 channels of optical information. You surround it with 20, 30, 40 of these devices. Now, you're into hundreds of optical channels that you need to deal with. All of that is great, except this device, this network switch, this processor, it's still got all of the challenges that you had before. These are huge devices that take thousands of watts. You need to deliver that power. You need to cool the device all while you're doing these high-speed tests.
The biggest challenge for all of this is it needs to be economical. It needs to be economical in order for this volume to scale. This is where Teradyne's productivity strategy comes through. We've developed an open architecture, which means by that, what I mean is on the top of the UltraFLEXplus, we're able to work with any prober company, any handler company to enable the endpoint the industry is inevitably going to get to, which is interoperability. We're going to provide the best electrical instrumentation as well as the optical instrumentation from Quantifi Photonics. That delivers the highest throughput tester in the industry. This entire solution is optimized for throughput and manufacturing productivity. That is innate. That's part of our DNA. That's why we win in the market consistently.
In this current time, when all of these different technologies are coming together, having all of the applications expertise from thermal, electrical, and optical in one company to support customers is critical to success. Let me just give you a quick visualization. This graph on the right is really a visualization of this problem that I'm talking about. As the photonics, which exist today in a data center, if you go out to a data center and you look at the edge of the data center, you'll see photonics. It's all used for long-haul communication. Data center a kilometer, 10 kilometers, 100 kilometers away. It's all optical communication. What we're talking about is all of that optical communication coming back inside the server itself. Inside the server, it's all optical communication.
That massively increases the channel count that you need to hundreds of channels inside a co-packaged optical device. Now, that does not make sense unless you can do it economically. That is where the throughput focus of the Teradyne UltraFLEXplus, the throughput focus of Quantifi Photonics, manufacturing productivity is part of our DNA. It is part of why we are going to win in this market. With that, I would like to wrap up by talking about what does all of this mean in terms of the numbers. The last couple of years since COVID, the mobility market, the automotive markets have both been a little bit slow. If you look at the underlying growth drivers for the future, artificial intelligence, what it is going to do in terms of edge AI adoption, volumes increasing, quality demands going up, those are fundamental drivers for this business.
On top of that, as you move to edge AI, the importance of connectivity between devices goes up. That connectivity is something that is implemented through wireless technologies. Our leadership position that we've maintained with the LitePoint business unit is going to persevere. We're going to be able to capture the largest share in wireless connectivity. On top of all of this, the reorganization with the product test division, it was thoughtful. There are synergies between the different business units that we can capitalize on. We can also take advantage of the vertical integration that's happening. The customer synergy with these large accounts between semi-test and product test is something that's going to be able to drive us to double-digit growth in a very profitable manner over the midterm. Now, this 12% plan in terms of growth rate doesn't include anything from Quantifi Photonics.
When we get past the close, we'll integrate. Right now, we only see upside from Quantifi Photonics. There are a number of tailwinds in the other business units. We're super excited to welcome Yannick and his team to Teradyne. With that, I'm going to wrap up. Traci, if you want to come. I'll see most of you later in the breakout session, I hope. Looking forward to meeting you.
Thanks, Regan. We're going to go ahead and take a quick 10-minute break. We have refreshments outside. If we can all reconvene here at 3:05 P.M., that would be great. We can hear Sanjay wrap up with the financials, and we'll do Q&A. Thank you.
All right. Welcome back, everybody. It's nice to see some familiar faces in the room. For those that don't know me, my name is Sanjay Mehta.
I'm Teradyne CFO. Incremental to my CFO duties, I also manage IT facilities as well as global operations. Thanks for taking the time to get to know us a little bit better. I joined the company in 2019. I joined the company with very strong core values that are positioned well for long-term sustainable growth. There we go. At Teradyne, we believe we have a resilient business model that's going to deliver attractive returns for shareholders. How are we going to do that? First, we're going to profitably grow our semi-test business, our product test business, and robotics businesses. Second, we're going to take that diversified portfolio of businesses and operate through a flexible business model that's going to be able to generate tremendous free cash flow through the industry cycles and deliver those returns.
Third, we're going to take that cash, and we're going to apply it to a balanced capital allocation approach. In January, we issued our January earnings model. We're committed to that earnings model that goes out to 2028. That model, if you recall, has a revenue range of $4.5 billion-$5.5 billion. Now, while we can control our execution and the performance of our company, we don't pretend to have the ability to forecast with exact precision over the midterm. We definitely aren't in a position to call the up-and-down industry cycles that occur. However, we believe we have many paths for success. We're confident in our earnings model from the top line to the bottom line. To make it simple, I'm going to do my presentation using the midpoint of our earnings model for the financial metrics. For example, I'll talk about $5 billion.
I'll talk about $8.25, which is the midpoint of our earnings per share, which is $7-$9.50. Okay. Let's get to it. In 2024, we had a pretty balanced revenue mix. From that jump-off point, our earnings model forecasts that we grow at 15% cumulative average growth rate from 2024 to 2028. Looking over history, we saw that growth occur from 2016 to 2020. That growth over history up to 2021 was really mobile-driven. Now, the ATE market is compute-dominated. Before I get into SOC and memory, I'd like to spend a few words on the IST business, product test business, and robotics business. In IST, we've had some ups and downs of revenue over the years. In 2021, our revenue was just under $300 million. By 2024, that revenue had contracted to $85 million.
Looking forward, by 2028, we expect revenue to grow to north of $400 million. It is really four growth drivers. First, high-performance compute devices are going to require SLT to maintain a high level of quality. Second, the SSD market is going to be driven by client PC and AI online storage for data center growth. Third, we are predicting the recovery in the smartphone market, which is going to drive more SLT revenue. Lastly, data centers, big topic of today, are going to drive exabyte growth and enable HDDN market to grow. In product test, we heard from Regan that that revenue is going to grow to over $500 million. That is a cumulative average growth rate of 12% from 2024 to 2028. We believe there are some key drivers of that growth.
First, at LitePoint, Wi-Fi 7 and 8 adoption coupled with smartphone and PCN market recovery. In production board test, as the automotive market recovers, so will the growth of the production board test revenue. Lastly, looking at our defense and aerospace business, we believe we're going to grow really tied to specific government programs over time. Turning to robotics, our revenue growth has varied. From 2016 to 2024, our revenue growth was roughly 17% on a cumulative average growth rate basis. In 2023 and 2024, our revenue growth actually contracted over that time period in a very challenging and weak industrial automation market, which we believe at this point is at a trough. Looking forward, our earnings model forecasts that we're going to grow roughly between 18-24%, with a midpoint of 21% on a cumulative average growth rate basis.
This growth is going to be driven by a couple of things. First, recovery in the industrial automation market. Second, we're going to introduce new products that are going to enable new markets to be penetrated. Third, AI is going to drive the adoption of these robotics technologies in both incumbent markets as well as growth markets. In 2025, we believe growth is going to be positive and grow to upwards of 10%. Through the midterm, we expect that growth to accelerate to 21%. Now, the predictability of this business in a very weak industrial automation market has been very challenging. You have seen that in our predictability over the last couple of years, for that matter. That is one of the key drivers.
When we stepped back and thought about it, that's one of the key drivers of why we resized our operating expense envelope to make sure that, given this unpredictability, that if the growth doesn't show up in the near term, we believe confidently that it's going to grow over the long term. In the short term, we wanted to resize the business to ensure that if growth didn't occur, this business would not be a drag on our financials. I'd like to put that in a little bit of context. Our earnings model has revenue growth at 21%. As Greg noted, the EPS impact at the midpoint of our model in 2028 is about 6% of our total earnings. That's $0.53 out of that $8.25. A couple of examples there. Let's say we don't grow on a cumulative average growth rate of 21%.
Let's say we grow at 10%. By 2028, our revenue in 2028 would be down by about $250 million. That tied to regulating our OpEx, as Greg talked about the operating principles, we would slow our OpEx down. That would translate to about 25 cents lower of earnings per share. Now, let's look at if we exceed the midpoint of our revenue guide. If we grow at 24% on a cumulative average growth rate basis, our revenue would increase about $90 million. That would translate to about 15 cents of earnings growth. The ATE market is currently dominated by compute. From 2016 to 2021, mobile was much more than 40% of the ATE market. The TAM in memory in 2021, roughly 50%, was tied to NAND flash for the mobile market. Mobile contracted from 2022 to 2024.
By 2024, the mobile ATE segment was roughly $800 million by our estimate, which was less than half of its peak in 2021. Looking forward, we are forecasting the recovery of the mobile market. How? Really tied to an upgrade cycle driven by AI at the edge that's going to require two nanometers and gate all around. Turning to compute, that market significantly grew from 2022 to 2024. We predict that it's going to continue to grow, albeit at a slower pace, over the midterm to 2028. We believe that market will grow to just under $3 billion. We do think that under the covers, there are some segment shifts occurring with VIPs. In 2024, we believe the VIP market was about $300 million. By 2028, we expect that market to grow to roughly $800 million.
That segment of the compute market, our estimate is estimating to be just under 30%. In auto and industrial, we expect that market to recover and grow. In auto, we believe there is more silicon going in cars, as Rick mentioned, as well as growth in hybrid and electric vehicles. In industrial, we expect that market to grow really tied to the power requirements required in the growing data center investments. In memory, 2024's market size was about $1.4 billion. We expect that market to grow at a 1.5%-ish, a little bit, maybe plus, and plateau. Why do we expect it to be at that level, a sustained level over the years? One, the recovery of the mobile market will drive NAND flash. Then secondly, HBM. Our revenue in SOC and memory will grow at about 13% cumulative average growth rate.
It'll be about two-thirds of the company's or the enterprise's revenue. We started in 2020, or we start the earnings model in 2024 with a very balanced revenue mix. In mobile, if I can address each of the markets, in mobile, we expect to grow with the market as that market recovers. In compute, we expect to grow with the market, really driven by our strong position in networking, as well as gain share as the segment shifts to more VIPs. In auto and industrial, we expect to grow with the market as well as gain share tied to our leadership in the power space.
In memory, we expect from 2024 to grow with the market, but also to gain significant share tied to the recovery of NAND flash, where we have a strong position, as well as growth in HBM and DDR, which we're well positioned for. Historically, Teradyne's had a very good record or a strong track record of delivering very strong gross margins. We expect to continue that progress and obtain our gross margin model of 59-60%. Under the covers, I'd like to share a few highlights. At Teradyne, we have a very disciplined approach to making investment priorities. We invest in complex problems that have large profit pools. Let me give you an example. In 2004, or sorry, in the mid-2000s, we had about 4% market share in memory. By 2023, we were north of 40% market share. You heard Young and Rick talk about it.
We did that by making very conscious investments of what to invest in. Young went through the product roadmap. What we chose not to invest in was DRAM wafer sort. We did not invest in DRAM wafer sort because it was a commoditized market segment with low profit pools. Fast forward to 2024, HBM, a DRAM wafer sort technology, is much more complex. Of course, we invested in there and are in position to win significant share. Going now more to our supply chain strategies. In 2021, we had an outsourced contract manufacturing factory, which had over 35% of our revenue. In 2022, that factory was located in China. In 2021 and 2022, we moved that factory to a different location. Another component of our resilience and our business continuity strategy is that hundreds of thousands of components go into a tester.
If you're ready to ship a tester and you can't ship because of one component, you're not shipping product to your customers. As I just mentioned, we had a factory in China with a deep supply chain. What we went and did in 2022 and 2023 is we qualified second sources for all of these components. We did that because we wanted to ensure we had business continuity in the event of a natural disaster, if we had tariffs, if there was any political or other issues that came up, we had the ability, depending on the circumstance, to point our supply chain to another set of suppliers in different countries to enable the sale and the business continuity of our testers. The other thing I'd like to talk about from a resiliency standpoint is 80% of our manufacturing is done through outsourced manufacturing.
That means when we go through the industry cycles, I talked about the free cash flow through the cycles. When we go through the industry cycles in a downturn, it's somebody else's factory and machinery that's bleeding to their P&L, not ours. Also, they're dealing with the employee impact. Moving on to our strategy for OpEx, I just want to share a couple of key strategic points about how we think about our disciplined approach to operating expenses. First, 100% of our employees are on variable compensation. That means when revenue and profits go up, variable compensation goes up. Everybody's walking around smiling. When revenue and profits go down, employee compensation goes down. When you look at this chart, there's an interesting observation. You're probably thinking, "That's interesting, Sanjay. In 2021, you had revenues of $3.7 billion. In 2022, that contracted significantly.
Why didn't your operating expenses come down?" That's a good question. From 2021 to 2023, we had flat operating expenses. What happened was our variable model actually did reduce our operating expenses. However, we specifically chose to invest in a pivot to compute as well as our channel transformation robotics. While looking at it externally, it was flat. Under the covers, we took that capacity and we reinvested it into our strategic priorities. Another point I want to make is that 35% of our employees are in low-cost regions. Sorry, they call this the Sanjay table because I did this in the rehearsal too. Quick drink. Yeah. 35% of our employees are in low-cost regions as of 2024.
Lastly, the last point I'd make about our operating expense strategy is that in our model, we have a cumulative average growth rate of increasing operating expenses of 8%. That's roughly half of our 15% revenue growth expectation. As we go through the years, we plan on moderating our expenses really tied to the expected revenue growth. Moving on to our operating profit. We expect in our earnings model to have operating profit expansion. Looking back to 2020, 2016, and 2021, we had tremendous operating profit growth, really driven by gross margin as well as revenues outpacing OpEx spending. However, in 2022 and 2023, our operating profit declined.
Again, I'll go back to we chose to invest in strategic priorities of supply chain resilience as well as our strategic investment priorities of, sorry, the pivot to compute as well as the robotics in that channel transformation. Looking forward to 2028, we expect to have operating leverage. That is really driven by revenues outpacing our OpEx investments. That profit is going to translate to free cash flow. Just some quick context. From 2016 to 2024, our net income, about 100% of our net income, translated to free cash flow. Internally, we have very disciplined working capital management goals that are passed on to all the business leaders and the management team, as well as we are asset light. 80% of our manufacturing is outsourced. Moving on to our capital allocation. Our strategy hasn't changed.
What we've done in the past is going to be consistent with what you're going to see going forward. We're going to continue to fund our dividends. We're going to buy our shares back to offset dilution. We're going to opportunistically buy shares back with excess liquidity. We're going to focus on our core strategies to drive M&A, but we're going to be disciplined about them. Our M&A approach, when I say disciplined, is really—sorry. It's okay. We're going to be disciplined about our M&A acquisitions. First, we're going to look to buy companies where we're going to acquire them, and we're going to have—we're going to buy those companies, and they're going to be accretive relative to share buybacks. So we spend a dollar on the company.
We expect to get more return than we would if we bought shares back on an EPS basis over the long term. Secondly, we want to buy companies that have an internal rate of return that are higher than our weighted average cost of capital. Just some interesting facts I have on the slide. Since 2015, we spent about $4 billion in our share buybacks. We bought back roughly 80 million shares at about $50 each. That's about a third of our outstanding shares had we not purchased any. We dispersed about $600 million in dividends, and we spent $1.1 billion in M&A over that same time horizon. I'll wrap up like this. I'll summarize what you have to believe in our earnings model for us to hit our earnings model. First, you have to believe the mobile, auto industrial market are going to recover.
Second, you have to believe that we're going to gain share in compute through the change of the—hold our share in networking and grow our share with a segment shift to VIPs. Third, we're going to introduce AI-enabled products that are going to grow markets, grow incremental markets, and drive that business forward. Our experience management team is going to operate our diversified portfolio of businesses in a flexible model. It's going to generate tremendous free cash flow through the cycles. We're going to apply that cash to a balanced capital allocation approach that's going to deliver tremendous value, attractive returns to our shareholders. Thank you. With that, I'll—thanks. With that, I'll ask the speakers to come on up for Q&A.
All right. We'll ask anyone who has a question to raise their hand and state your name and your company, please.
Hi. It's Tim Arcuri at UBS. I actually had two. First, I'm wondering, Sanjay, is there a profitability line in the sand for IAA? Because we heard last call that you're cutting OpEx in IAA, and now there's a lot of growth expected. I'm kind of wondering how you're going to balance that. Is there a—you're not going to let profitability be less than 5% op margin? Is there a line in the sand that if the growth isn't there, that you won't invest the money?
Yeah. I think, as I noted in my prepared remarks, the predictability of this business has been quite challenging in a tough industrial market. As the team outlined, we've outpaced our peers in the industry. I'd say that we've reset our OpEx envelope because of this lack of predictability, such that if the growth doesn't occur, we won't be a drag on the bottom line. Historically, our expectation is, given the fact that we expect positive growth this year, that we are going to incrementally add to the bottom line this year as opposed to the prior years. Now, historically, we've managed the business at a 5-15% level when we were growing at a much higher rate.
As we go through the cycles, and sorry, as we project to grow, I would say that you should expect to see operating leverage where, as revenue grows, for every dollar of revenue growth, you should expect, say, $0.50 of OpEx growth, just like the operating principles Greg laid out.
Yeah. To that point, I think in terms of the mechanics, in the robotics business, we're kind of from Missouri right now, that if we demonstrate growth in a particular year, that is going to help us set the operating expense budget for the following year. If the growth doesn't come, we're going to constrain the investments in that space to try to make sure that we don't get ahead of the cost curve again.
Okay. Great. Can you somehow frame—I know you talked about a new HBM customer, which I thought was pretty interesting. Can you frame how important that could be, how meaningful it could be? Was that in your forecast for this year? I know that right now there's some macro, sounds like some export controls are hurting that business, the memory piece of the semi-test business. Was that customer in your forecast, and how big can it be? Thanks.
You want to talk about that, Greg?
Yeah. Sure. The customer was in our forecast. We've been engaged in qualification for a number of months. I don't think we've broken out the quantification, but as Young mentioned, it's HBM4. HBM4 really takes off towards the end of this year. The big impact is going to show up in 2026.
Yeah. Hi. It's Krish Sankar from TD Cowen. I had a question for Greg and one for Rick. Greg, on the pre-announcement for the full year, taking it down, I'm just kind of curious. Can you tell us where exactly it came from? Seems like mobile was already weak. Mobile test, you're kind of under-indexed to compute. Is it fair to assume it's pretty much auto analog industrial and maybe some memory? Any color on that would be helpful. I had a follow-up for.
Yeah. As far as the change, the information that we have right now is really focused around the dynamics for Q2. That is no cancellations, some push-outs, and some customers that we expected to be placing orders that have deferred placing orders to later. We have pretty good visibility about the impact that that will have in Q2. We have less visibility about the full year. We are trying to make sure that we give you the best idea that we have, but there is a lack of visibility into the second half of the year. We tried to be careful about how we set the first half versus second half revenue. In terms of the balance of that, you are right that the impact is smallest in mobile because mobile is already relatively small.
It is really balanced across the rest of the businesses, that we're seeing similar stories from customers in different spaces around tariff concerns and trade restrictions. They are taking a bit of a wait-and-see perspective to see if this is something that goes away quick or turns into a problem.
Got it. A quick follow-up for Rick. On the mobile test, we saw over the last several years the emergence of SSN, the structured scan network, which improved efficiency, but unfortunately for you, lowered the test times. Do you envision something like that happening in compute, that by 2028, the market is not as big as you think because it's gotten more efficient? Just curious on SSN for compute. Thank you.
We really think that that's a technology that isn't specific to mobile. It's a feature of most modern digital design. So it already exists in the compute space. We're not anticipating a step function like we saw when mobile was big and that technology got introduced.
Good afternoon. Joe Dooley from Stifel Securities. My question has to do with why you win in the VIP space. Someone made an interesting comment about, I think, how Advantest was now testing only one up. I'm assuming that's GPU devices. I'm curious, is that part of the strategy to win business where you're able to test more than one up? Does that tie into your investment in Technoprobe? I have a follow-up.
You want to take that?
Yeah. Sure. Yeah. I think that's a big part of what we see as our advantage. Our competitor system is architecturally limited. The result of that has been to pull back to single-site testing. The UltraFLEXplus was architected for much higher pin count expansion, much higher power levels. It's capable of multi-site testing. I like to say, eventually, the accountants take over, and people start to care about cost and efficiency. We're starting to get to that point in the maturity curve, and that's going to tilt the winds our way.
Do you think you've talked about winning 50% share, I think, in the VIP space with these hyperscalers and guys that are just designing their own custom silicon. I'm curious, could you actually win business back at NVIDIA? Or could you win business at NVIDIA with a 50% throughput advantage? I would think, as you mentioned, that the accounts would start to take over. That seems like a big enough cost reduction and a roadmap advantage that you might actually be able to win a slot or two there.
I think we have to make investment decisions around where we put our money to try and gain business. Over the past couple of years, we saw the opportunity in VIPs where the concrete had not yet set, that there was not this strong preference. We definitely prioritized that for the majority of our development resources. We're always trying to gain share in these accounts. In our view, the primary way that that would happen is through the disruption that is going to happen in the transition to co-packaged optics. As things go forward and you go to new connection technologies, that really changes the battlefield and makes it much easier for us to try and demonstrate an advantage. That is sort of the key that we have towards our overall share gain strategy in this space for the traditional suppliers.
A couple of follow-ups. I want to go back to the revised expectation for 2025. If I just take your revised guide, there's about $200 million-$220 million that comes out of the calendar year, $80 million-$82 million from Q2 to Q3, and then more. I am a little bit confused. I was wondering if you could help me. Obviously, this kind of adjustment is going to—semi-tests would essentially have the biggest impact. I am thinking you're at the component level. I do not see any change to the wafer stars or wafer shipment. You do the test afterwards. I imagine, even if there is a tariff, more of EMS companies would see the adverse impact or would cut this early in the change of a tariff regime.
Why would you see it in the semi-test, given the fact that you're still up in the food chain and far from a finished good coming to the U.S.? I have a follow-up.
Let me tell you how we revised our expectations. Since the January earnings call, we got new information from our customers around our plans for Q2. It was clear that we had that reduction coming for Q2 to move us to that flat to minus 10%. That much we knew. After we looked at that, we then looked at the resulting balance of first-half revenue versus second-half revenue. We wanted to try to maintain—we already had a relatively bad skew towards the end of the year. We wanted to be careful around whether we would make that worse or not. We tried to keep that balance about the same. I want to emphasize that we have very little visibility about what's going to happen beyond Q2. There are really three cases.
One case is that the Q2 impact was sort of on its own, and the second half stays where we thought it was going to stay. That is case one. Case two is that the Q2 came down, or the Q2 slides out to the third quarter, and then some stuff in the fourth quarter slides to the following year. That is the sort of—that is the case that we modeled in the guidance. The other thing that happens is people took a breath, sit back, and they go, "Oh, wait, we need that capacity." Q3 comes roaring back. It is entirely possible that we could end up with the Q2 revenue moving into the second half. That would strengthen the second half even more than it was now. We felt uncomfortable setting a target around that. Did that help?
Yes. Thank you. On Quantifi Photonics, have you disclosed what has been the revenue, the trailing 12 months? If you look at the product test mix, today we have some feel for it. What would the mix be in 2028, including the photonics opportunities?
We're going to wait to answer questions about the QP acquisition until we actually get to close. We're not going to disclose any of that stuff right now, but we'll give you a better picture once we have the transaction finished.
Let me just circle back. There is a new HDD medium that is planned to come to the market soon. Have you seen any activity related to HDD tests? It's been at a depressed level for quite some time.
Are you talking about Hammer?
Yes.
You want to talk a little bit about that?
Yeah. Let's see. I don't want to make any specific comments about the deployment of technologies at our customers. What I would say is we are seeing pickup in activity in the HDD space, exabyte growth rates starting to revive, and actually a little stronger than what we were thinking. It's kind of early days. There's a little bit of capacity to absorb first before we really get into buying.
Yeah. One of the things about this is it kind of ties into this midterm resilience that if you look at the exabyte growth rates, those customers are going to need more test capacity. Depending on whether the exabyte growth rate goes at the lower range or the higher range, that's either going to be 2026 or 2027 or 2028. We are trying to be a little bit careful about the in-between. We are certain that the upward pressure from data centers is going to drive growth.
Hi. Samik from JP Morgan. Maybe just going back to the mobile DRAM. When I look at your 2028 forecast, obviously, that's a big pickup. Just help me think through how much of that is dependent on units coming back to a higher level versus how you're envisioning sort of the overall increase in terms of testing intensity, including two nanometer in that roadmap. How dependent are you on the units rebounding on that front? I have a follow-up on the IS side.
You want to take that?
Sure. I think we're forecasting both initiatives, right? A continued growth of complexity similar to what we've seen, primarily limited by technology improvement over time. What's going to drive the consumption in that direction is these AI features that Greg was talking about. We also expect the refresh rate to come back. Unit volumes are down quite a bit from the peak. We might see a peak like we saw in 2021 if there's really dramatic improvement in the AI functionality. Our model is more tepid than that, right? Today, unit volume growths are in the kind of 2% neighborhood. Our model's built around AI features driving us to something in the 2-4% unit volume growth rate.
Yeah. If you're trying to balance it, I'd say that it's the complexity. Because in addition to just sort of the raw transistors that you need to be able to support AI, there's significant changes in architecture. You need a ton more memory bandwidth. That means that there's more connections to the device and things like that that make the test more capital-intensive than it is now. I think we're weighted more towards complexity than refresh. Please, in 2026, everybody go buy an AI phone would be a really good thing.
On the IS side, when we think about the go-to-market approaches that you've tried, like in terms of now working with integrators, etc., on that front, I know sort of there are a lot of questions on the profitability side. When you think about even the runway in terms of trying these different partnerships in terms of the go-to-market, where are you in terms of currently trying out the latest plan? How much runway do you think in terms of seeing whether success is sort of driven by the current sort of go-to-market approach? Do you need to pivot to another sort of backup plan in terms of what would be the go-to-market if this current sort of push doesn't materialize on that front?
Yeah. You want to take that, Ujjwal?
Our pivot towards OEMs, which we started three years back, really, has been a bright spot for us. As you saw in 2024, that was one of the key reasons for outperformance versus peers. I feel that OEM program is matured. That is a key vehicle through which we will be expanding into these new end markets. As you get into, say, food and beverage and semiconductors, and in semiconductor, we are focusing on our large-account strategy. Say, life sciences, logistics, that OEM structure, which gave us good success, is something which we'll double down on in expanding into new segments. We have good initial traction and learnings on the large account. That is one focus area which we'll be investing more on in 2025. We started announcing several of those strategic partnerships.
Those strategic partnerships are helping us get good foothold in some target large accounts.
One thing that I'd like to add is just sort of the different flavors of partnerships. If you look at the Matthews partnership that we just announced, they're a fantastic global integrator for logistics. They're like a channel. That is an augmentation to our integrator channel that's going to be fantastic. Analog Devices is an end customer. The partnership that we came to with them is really around applying AI to these back-end semiconductor challenges that we can leverage in other places. With NVIDIA and Siemens, both of those are really part of the platform and ecosystem play. Honestly, a part of it is actually around marketing.
If you look at how much money NVIDIA is putting into marketing physical AI and robotics, we really want to benefit from the visibility that they're adding to this space and make sure that the robots that are being used to demonstrate physical AI have blue end caps and have MiR logos on them. Because we think that that's going to help open up the opportunities and drive growth. The different partnerships really play in differently. I don't see it as like right now we have a really good idea of the roadmap for our channel strategy, and we're building it out. I don't think it's going to change.
Yes. Gus Richard with Northland. Looking at your expectations for 2028, compute grows as a percentage of your SOC mix, while that comes down as a percentage of the market. I'm just wondering if you could kind of give us a little color on what's driving that. Is that purely gaining new VIP share, or is it co-packaged optics, or do you think you can turn an AMD, NVIDIA, or Intel for that matter?
You want to take that one?
Yeah, sure. I think what I can say is we're primarily focused on the VIPs. Our expectation is that silicon photonics, as Regan showed in his chart, is going to become a meaningful part of the space and the revenue stream. It's kind of an open playing field right now. We have an opportunity there to make inroads across the industry.
In thinking about your 2028 projections, what do you expect the buy rate to be? Is test going to outgrow semis going forward, or do you think it's going to hold in the same?
The buy rate has been pretty consistent around 1% since 2010. I am not anticipating a huge change in the buy rate through this midterm.
Yes. Rob Mason with Baird. Greg or Ujjwal with this question on robotics. How are you defining your AI-driven business? How does it go from, or how are you measuring it going from, I guess, $11 million to $150 million? You started offering the AI accelerator last year. Is it defined by that, or are you going to be relying on the ecosystem to attach AI to your?
Yeah. Let me answer that one because I spent a fair amount of time torturing Ujjwal and his team to try to make sure that we got a measure that we can trust around this. The way we count AI-driven applications is if the product that we sell has AI features. The MiR1200 Pallet Jack, we would consider that to be AI-driven revenue. Over time, we're going to be adding AI-driven features to more and more of the MiR product line. For UR, since we are providing a platform that our partners are building AI solutions on top of, we're enabling that ecosystem. We've set a rule that we need to be able to identify the exact partner and exact technology in order to count that as AI-driven revenue.
If some rando buys a robot and builds an AI solution with it, that would not be in our AI-driven revenue envelope. If there is a partner that we know their solution and it is built on an AI technology, then we would count that revenue as AI-driven.
I'll just stick on robotics for the follow-up. How are you seeing that market fragmentation start to consolidate either on the arm side or the AMR side?
You want to take that one?
Ten years back, it was just Universal Robots. There was another company which folded. We did see a bunch of players who jumped in. Now we see a lot more consolidations or exits from that market. I expect the next ten years to be very different, which will be the scale-up phase of any market that goes through that. Both the cobot and the AMR markets have reached that stage where customers have their core operations running on these technologies. From here on, only the players who are well-integrated with customers' operations in a reliable way are the ones who will survive. We feel very well-positioned with the key large customers in the market.
The other thing that we've seen is the focus on humanoids is drawing a lot of sort of entrepreneurial zeal. We were seeing a lot of copycat stuff coming into the AMR and the cobot space. Now there's sort of this new hotness. I think we're seeing fewer entrants, and we're seeing more consolidations. We're also seeing some of the low-price players in this market struggling to either go public or to get the pricing that they want for an exit. There are early signs of consolidation, but it's not a trend yet.
Hi. Yeah, this is Dennis Piachin and on for Brian Chun from Stifel. My question is about silicon photonics and the relative complexity of testing for that versus just regular silicon. Could you tell us a little bit more about maybe the unit economics or how much more incremental revenue you can get from testing all these CPO and silicon photonics chips? Maybe also the timing of that revenue, like when's that going to hit the top line?
Regan, you want to take that one?
Sure. In terms of complexity, there's the three different stages that I mentioned. There's some really big technical challenges with respect to active alignment at the wafer level. The combination at optical engines where the data rates are going from 112 to 224 for 48 Gb, combined with optical test, the capital intensity is growing. When you get to copackaged optics, this is very intense from an electrical, thermal, power, and optical combination coming together. These are very challenging insertions. The key is for us to bring high-throughput solutions to each one of these insertions in order to capture the share. Honestly, in this growth period, our focus is more on winning the business. Ultimately, the size of the market will be determined by how quickly the industry can scale.
We've heard that some large foundries may be deploying some of these products towards the end of the year into 2026. Can we expect for your revenue to be coming in at roughly the same timeframe?
As I had indicated, we definitely expect that there's going to be a ramp in this midterm. The early deployments are going to be small, but they are indicative of what I described in terms of people are making decisions now about who they're going to partner with and what equipment they're going to use. Our focus is on being a great partner for those companies. Yeah. I think our best view is sort of manufacturing enablement and sample production in the 2026 timeframe and more substantial ramps in the 2027 timeframe. There is a lot of work to do in terms of the manufacturing enablement phase. We'd expect to see something, but it's definitely one of these things that's gathering strength as you get towards the end of the midterm.
Hi.
Hi. Doug Sanchang from Bank of America on behalf of Vivek. On memory time, I think you said you expect time to be flattish into 2028. I'm curious why you would think so, just given the underlying market is growing so fast. Also, the HBM4 and three cycles are ramping with probably higher test intensity. Thank you.
Yeah. I think if you look carefully at Young's drawings, we have a wide range by 2028, ranging on the low side from flattish to substantial growth, almost another $500 million, $300 million-$500 million in that 2028 timeframe. The reason there's a lot of variation there is you're right. There's an underlying organic growth that's pushing the market up. There's the kind of normal pattern of technology transitions happening in DRAM and flash. In HBM in particular, where a lot of the growth is coming from recently, there's some substantial improvements in productivity rolling out. What you saw last year was a very low productivity capability, and it artificially ballooned the TAM. We think that's going to moderate just because of the economics of the memory market, right? This is a very economically focused commodity-like market.
We think that the efficiency is going to improve. There is a lot of variation in that in the potential outcome there. You can see that in our forecast.
Part of it is also just having a multi-year backward perspective that the memory TAM in 2024 was the largest TAM that we've seen since 2007. Capital intensity staying at that level is pretty remarkable. A lot of the growth happened very suddenly because of this low productivity beginning. There is going to be a lot more capacity added for the same amount of capital spend in those years coming up because of these changes. We have that uncertainty.
Got it. As a follow-up, staying on the topic of TAM, do you plan to update the 2025 TAM estimate for SOC and memory, just given all these tariff impacts?
Yeah. We're going to we gave you the latest breaking news in terms of how we saw Q2 shaping up. We expect that we'll have a much fuller picture in terms of how much impact that would have on the TAM, how much impact that will have on the rest of the year. We'll share that news in our April earnings call.
That's it for time for Q&A. With that, we will say goodbye to those who are on the webcast. For those of you who are in the room and would like to join us for the technology tours, if you look on your badges, you'll see a color code dot, either green, blue, or yellow. Those with a green dot, please meet outside of the doors to the conference center. Greg and Shannon will take you around.