All right. Good morning, everyone, and welcome to Cohu's 2026 Investor Day. I am Matt Hutton, vice president of strategy and investor relations. On behalf of the entire team, thank you for joining us here in New York City, and a warm welcome to everyone participating through the webcast. We appreciate your time, your interest in Cohu, and the opportunity to have a more depth conversation about our business. Whether you have followed Cohu for many years or are getting to know us today, our goal is the same, to give you a clear understanding of where we are going, why we see opportunity ahead, and how we intend to translate that opportunity into long-term shareholder value. Before we begin, please take a moment to review the forward-looking statement disclosure in today's presentation. Our remarks will include expectations about our markets, strategy, and future financial performance.
These statements are subject to risks and uncertainties, and actual results may differ materially. Please also review the risk factors described in our SEC filings. Please note that the midterm financial model refers to a time horizon of three to five years. We will discuss non-GAAP financial measures as well. Please refer to the accompanying disclosures and reconciliations in the presentation materials, including the notes relating to forward-looking non-GAAP measures. Today is an opportunity to step back from the quarterly reporting cycle and look at the broader direction of Cohu. The theme is expanding our growth aperture, building on our foundation in semiconductor test and inspection to address a wider set of customer needs in yield, productivity, and manufacturing intelligence. We have designed this program to connect three things: the market opportunity, the technology that allows us to address it, and the financial framework behind our strategy.
As you listen, I would encourage you to keep three questions in mind. Where can Cohu grow? What differentiates our solutions? And how does that growth translate into value for shareholders? Luis will begin our strategic direction and the opportunities we see ahead. We are also pleased to welcome AMD for a customer perspective, the market forces shaping high performance and AI computing. You will hear from our business leaders about AI compute, package inspection, including high bandwidth memory and advanced packaging, data center power management test, and lifecycle recurring revenue through interfaces and software analytics. Jeff Jones, our Chief Financial Officer, will bring those discussions together with our midterm financial model and capital allocation priorities. We will then open up the floor for Q&A. For those joining us in person, the product technology showcase will follow the presentations and Q&A.
I encourage you to use that time to see the technologies up close and connect what you have heard with the manufacturing challenges our customers are working to solve. To help us keep this program moving, please hold your questions for the Q&A following the prepared presentations. We want today to be a useful dialogue and welcome the questions that will help you better understand our strategy and priorities. Let me close by thanking our speakers, our customer participant, and everyone who helped bring today together. Most importantly, thank you to the investors and analysts for being here and engaging with us. With that, it is my pleasure to introduce Cohu's President and Chief Executive Officer, Luis Müller. Luis will take you through our strategic direction and how we are positioning Cohu for its next phase of growth. Luis, over to you.
All right. Thank you, Matt. Good morning, everyone, and thanks for coming up here with us. My goal today is actually just to frame the strategic plan of the company around four growth factors. Then I am going to let the leadership team talk about those in more detail. They are really the stars of the show today. I have been the CEO of this company now for close to 12 years. Frankly, I have never seen an opportunity in front of us of the magnitude that we have today. This is really exciting times for us. You probably know the company well. We built a strong position in semiconductor test handling equipment over the years. Today, the idea is to frame a discussion around how we are going to double the size of this business over the next coming years. Let us get started here.
Before discussing a lot about that framing, let us talk about and highlight a little bit the strength of the business that we have already built in this company. We have over 25,000 systems installed and actively working worldwide at our customers' locations. We have longstanding customer relationships. We have global service and support organizations around the globe. I think an important factor that a lot of people miss, we have about half of our revenue from recurring. Our global manufacturing, supply chain, and support capability help move programs from development to production. That is an important point for our customers that need to ramp and generally our supply chain constraints in these days. Our solutions across our product lines span thermal handling, test and interface products, inspection metrology, software, and aftermarket services. Again, you need to maintain that equipment and keep it productive all the time.
That breadth is a competitive advantage for the company. As the semiconductor complexity increases, customers actually value partners that can solve broader yield and productivity challenges more so than just provide an individual piece of equipment. We view that that shift plays well to the strengths at Cohu. You will hear from these leaders here in the company responsible for carrying out this strategy. They bring pretty deep semi-cap equipment expertise, good, strong customer relationships, and a proven record of innovation and execution at the company. Their perspectives both reflect our strategy, but also they have direct contact in understanding what is happening across the semiconductor ecosystem. With that, let us talk about the four platforms and the next stage of growth for the company. Like I said earlier, Cohu has historically been viewed as a test handler company, largely serving automotive and industrial markets.
That core remains fundamental to our business. In fact, if you look at what we have been reporting, test utilization across our fleet has been increasing, and it now ended third quarter at about 82%. That view kind of understates our capabilities and opportunities ahead. We are expanding now, as I said, to be a broader provider of solutions that improve yield, productivity, also providing process intelligence solutions to our customers. We are doing that by combining our equipment, our services, the software that we are laying on top of that, providing analytics, so that we can capture a larger share of this value stack in semiconductor back-end manufacturing. Our target markets remain pretty large, and they are driven by some powerful secular trends in the industry, whether you are talking about artificial intelligence or some advanced memory architectures, electrification, and eventually software-driven productivity and yield opportunities.
These trends increase the importance of technologies that we already have, like thermal, vision, interface products, test, instrumentation, software, and they are the building blocks that together with our customer engagement drive the next stage of growth for the company. This is not about reinventing Cohu. It is actually about extending what we do today into faster-growing large markets where we view the opportunity to create value. If I bring that together, the opportunity that we see here is to go from an approximately $3 billion addressable market today to a $5.4 billion market opportunity, okay? That expansion, as I said, comes out of four growth themes. First, AI compute. Obviously, rising power, test intensity, they are creating demand for active thermal control technology at test, and we will talk quite a bit about that today. Second, package inspection.
Whether we are talking about HBM or advanced packaging, they require more sophisticated vision, measurement techniques, and process control solutions. Third, lifecycle business. That relates to that recurring opportunity that I was talking about, and we see a path here to grow that recurring with co-packaged optics interface and software analytics. Fourth, data center power management. The adoption of silicon carbide and gallium nitride technologies is driving the need for newer, higher current, higher voltage test instrumentation, and we will discuss that today as well.
Each growth platform here is supported by some durable technology trends and growing manufacturing complexities. They are also highly complementary, if you think about it, because they all come around test and inspection, right? They help us expand this addressable market, strengthen our recurring business, and deepen our customer engagement and relationships, and that is how we are going to double the size of the company.
The leadership team will provide supporting details on that and dig deeper into it, and my goal today is simply to create the framework around this strategy so that we can talk more about it and close with the financial layout of what our target model is. Let me start with the largest of our growth platforms, which is AI compute. Artificial intelligence is driving a very rapid increase in semiconductor performance. We all know that. With that is coming power consumption, it is coming thermal density, and a lot of package complexity. As these devices become more powerful, testing is becoming significantly more difficult. Large AI accelerators and processors must be validated at pretty demanding temperature conditions and very high power levels, oftentimes actually more than what you see on the spec in the actual use case scenario.
That is where Cohu's active thermal control capabilities create value to our customers. Our Eclipse handler is able to maintain a precise temperature control, a uniform temperature across the die, while our customers continue to move and advance the technology of these processors. This eventually enables efficient characterization, improving yield, and improving production quality of the devices, ultimately translating in value to our customer base. As processor architectures continue to evolve, test intensity and power requirements continue to be a secular strong driver for the industry and for Cohu. Our second platform is package inspection. High-performance computing is driving more complex package architectures, whether you take HBM stacks, different types of stack dies, micro bumps, heterogeneous integration. At the end, these are creating real challenges in package capability, in process inspection, and process control.
Detecting these defects early in the process has real economic value to the customers. Our Neon platform, which we will talk about today, addresses these challenges by combining a myriad of vision technologies. We will talk some about infrared, dark field imaging, three-dimensional metrology. Again, increasing this package complexity and detecting defects early in the process is the long-term value creation opportunity in this space. In lifecycle business, we seek to expand that base of recurring revenue. What I said, it is about half of our revenue today. We are going to do that through co-packaged optics interfaces and software analytics. As data centers evolve, co-packaged optics and high-speed interfaces are becoming increasingly more important. Concurrently with that, you are handling some really high ASP devices, so ensuring the quality during that manufacturing process and greater care of the device becomes critical. That is where the software analytics comes in.
It gets installed in our equipment, and it processes data for our customers. We have what is called PAICe Digital Twin software that combines test data, inspection data, and converts that into actionable intelligence. It is quite a meaningful opportunity, actually. When you come to think about it, just some modest improvements in yield and uptime and equipment utilization can translate into significant value to our customers. Fourth is testing power semiconductors. As AI is increasing both computing demand, it is also driving the need for more power-efficient solutions across delivery and endpoint of use in the data center infrastructure. As these scale consumption increases, you see more of the penetration of silicon carbide and gallium nitride, which are more efficient semiconductor solutions that help with the power systems in a data center. These devices require higher current, higher voltage instruments, basically new class of instruments in the industry.
Our Diamondx tester, along with our Neon handler, are providing a differentiated solution for testing these power management devices. We will, like I said, discuss that as well. Now, stepping back, these four growth platforms really have a common pattern. If you think about it, each one of them addresses a large and growing market. Each one of them builds on capabilities that Cohu already has. Each basically strengthens our differentiation and customer value. Combined, they create close to a $2.5 billion opportunity for the company to grow into. They expand our SAM to about $5.4 billion, and they support our strategy, which is to grow Cohu revenue to $1.3 billion, and most importantly, deliver $5 of EPS. They also position Cohu at the intersection of important secular trends, which is AI compute, advanced packaging, co-packaged optics, intelligent software, and power-efficient solutions.
These are not short-term cycles. These are durable technology transitions that we believe can drive value creation for many years. Our leadership team, like I said, will now go deeper into each one of these platforms and describe why we are confident we can actually deliver the growth that we are talking about. But before we do that, I would like to introduce our guest speaker. I am honored to welcome Keivan Keshvari, Senior Vice President of Global Operations and Quality at AMD. As we all know, AMD is a leading innovator in high-performance computing, AI accelerators, and advanced semiconductor technologies. Keivan will share a customer's perspective on the opportunities emerging in the high-performance computing ecosystem and how they are changing test requirements, which is what affects us. Keivan, thank you very much for joining us here today, and the stage is yours. As soon as I collect my papers.
Thank you, Luis. Good morning, everyone, and thank you for Cohu team for inviting me. I am Keivan Keshvari. I lead global operations at AMD. Please take a moment and note the disclaimer at the back of this presentation. I am here today as a customer. I will not speak for Cohu, and I will not talk about anyone's financials. What I can give you is the view from the perspective of a customer's factory floor.
What is happening to demand for data center compute, why the products that serve it are so much harder to build and test, and why that matters for the suppliers we choose to depend on. Here is my message. AI demand is growing exponentially. Data center products are growing more complex, and our quality bar is rising with both. Together, those forces are creating production test requirements that very few suppliers can meet. Cohu is one that can.
We have partnered with them for more than 25 years, and we count on them to support the growth ahead. Over the next few minutes, I will walk through the demand, then complexity, and what that means for tests, and finally, our experience with Cohu. Let us start with demand. This chart shows worldwide monthly inference token consumption from early 2024 through this past September. Two years ago, AI was only beginning to reach the mainstream. Today, millions of people use it every day. The world now consumes more than 35 quadrillion tokens every month. That is up roughly 95 times in just two years, and the curve is still growing steeper. Why does this matter to an operations leader? Every one of those tokens runs on a silicon, and that silicon has to be built, tested, and shipped at quality before it can generate a single token.
The more useful AI becomes, the more compute the world needs. All of that demand lands on a supply chain that includes production testing. This demand has two engines, the training and inference. Training remains foundational. Since 2020, the compute used to train the most advanced models has increased five times every year. That is what has given us better reasoning, new capabilities, and a growing number of specialized models built for specific industries and tasks. The bigger shift over the last year is inference. For the first time, the world is using more AI compute to run models than to train them. In just two years, inference has grown roughly 60% of global AI compute capacity. As billions of people are using AI every day, the workload moves from building models to putting them to work. Agents will accelerate that shift. A chat model answers one question.
An agent takes a goal and keeps working it until it has solved the problem. From a compute standpoint, that is a step change, because one request becomes dozens of steps, reasoning, tool calls, data access, over and over. You need GPUs to do reasoning, and you need lots of CPUs to orchestrate every step around it. More agents, more tools, more inference, and more compute. For us, that means more units, more complex units, and a quality expectation that does not move. To meet that demand, no single chip or system architecture is enough. You need a broad portfolio. This slide spans the full range of AI, from real-time physical AI at the edge to inference and development on workstations, to fine-tuning, to distributed training, and finally, to mega-scale inference and training models of a trillion parameter or more.
At the edge, we have Ryzen AI, embedded and Kria solutions, plus FPGAs adaptive SoCs for real-time processing. In the middle sits EPYC, Radeon, and Instinct. At the far right is Helios, a 72 GPU rack scale built on MI450. The value is placing each workload on the compute that fits it with consistent open software approach. From the operations perspective again, every product on this slide has a different package, a different power profile, and a different test requirement. A portfolio this broad means a test and handling partner has to cover a very wide range, not just one device type. Here's the same portfolio organized by product family. You see the EPYC processors and servers, Instinct accelerators for AI training and inference, Pensando programmable DPUs, AI NICs for networking, Versal adaptive SoCs for AI at the edge, and Ryzen AI and Radeon AI for client. NICs for networking.
I'm not going to walk through each one. The point is breadth. We build processors, accelerators, networking silicon, adaptive devices, and customers increasingly deploy them together as one system. This is great for our customers. It's very demanding for our operations. Each of these families has to be tested to a high standard at production volume, so the partners we select have to be able to cover the whole portfolio, and they have to scale with us. This slide is about what changes as AI moves to rack scale. The industry is moving from components to systems. Customers don't want strong silicon alone. They want complete solutions that are ready to deploy. That means we have to think across the whole path, from chip to package to server to rack. Interdependencies are multiplying. Performance now depends on how well compute and memory and interconnect power, cooling work all together.
A change in one area will affect the entire system. That raises the bar for coordination, collaboration across partners. Speed matters as much as capability. Lastly, quality. When one failure can take down a rack, not just a chip, the cost of an escape goes up sharply. That's why we keep innovating our test solutions. We need to deliver a high-quality product and still meet demand. We have a history of pushing limits at the silicon and package level. Now we have to apply that same discipline to the rack level. With partners like Cohu, we can turn that complexity into a real advantage. Let's start with the physical side of testing, the handling element. A few years ago, a handler had to deal with one or two formats. Today, test happens in nearly every level of integration.
Reading across the slide, wafer-level testing, singulated die testing, substrate-level testing, final package testing, module testing, and board-level testing. Each of those changes the physical object that the handler must hold, align, contact, and move. A wafer, a bare die, a large package, and a complete board all behave differently. The two points that matter is, the first is, we can only push the power and performance of the part once the product has been packaged. Wafer and die-level test can only screen basic functions, but the real operating envelope is only reachable in final package test. Second, comprehensive testing is critical before the ASIC is mounted onto the board, and well before the rack-level testing. Once the device is soldered down, a failure is expensive to find and difficult to fix. We need to catch it at the earliest point possible.
Few suppliers can enable testing at these conditions, and that's the reason the supplier list is short. Now let's look at what happens electrically when device goes into test. The chart shows a bounding box of test requirement. The smaller box is where the test is today. The larger box is where AI needs it to be. The size of the exact box varies by test methods and tester and instrument type, but the direction is clear. Both power and IO are growing. Test power is growing roughly two times. Test IO count and package size are growing two to four times. Here's what's behind that. On power, we're dealing with higher power density and vertical power delivery. On IO, the connection types are broadening, electrical SerDes, optical SerDes, and memory IO. We're testing larger multi-die package with very high power loads and a lot more connections.
All of that has to stay accurate and reliable at production volumes. The challenge is no longer simply removing heat. Handling, test interfaces, and thermal control all have to work as one system. That's why working with Cohu matters. They help us combine advanced handling, test interfaces, and active thermal control. That makes accurate, repeatable testing possible, and it does so at production volume. Power creates heat, so let's talk about the thermal issues. The thermal is where the evolution is most visible. On the left is the traditional approach. A single external temperature sensor provides feedback. The device is tested at steady rate with slower reaction time. Controls are passive, meaning cooling or heating only. That worked well for low to mid-power devices, up to roughly 500 watts. On the right-hand side is what AI devices need.
These devices draw far more power, and they produce fast localized heat spikes. So we now use multiple analog and digital sensors inside the die to find our hotspots. We detect transient spikes and react quickly. And we use active thermal control, which is closed-loop, multi-zone, and predictive. This is for ultra-high power, beyond 2 kW. Why does that matter for quality? If you can't hold the device at the right temperature during test, you either fail a good part or pass a bad part. Neither one of those are acceptable when we talk about data center requirements and their reliability requirements. This is where Cohu's active thermal control technology comes in. It provides precise closed-loop control for high-power devices with support up to 6 kW. This is why it's critical to have a partner with Cohu's capabilities. Let me close with our experience.
We've worked with Cohu for more than 25 years. We started with Delta Summit handlers and then integrated Cohu's thermal subsystems into AMD's custom systems, and we have continued with the latest Eclipse handlers. Today, Cohu's handlers are used across majority of AMD portfolio, including data center products. They are deployed across our manufacturing partners for production tests, so the relationship reaches across our manufacturing network. It's not just in one site, in one factory. Why does this matter right now? Because AI demand is outpacing supply across the industry. When capacity is tight and requirements keep rising, a proven partner with deployed qualified solution is not a convenience, it's a necessity. To the Cohu team, thank you for your continued support and work with AMD. AI's compute demand is growing exponentially. As it grows, products become more complex and quality requirements get tougher.
Production test is where those pressures meet, and few suppliers can meet them. This is why the right partnerships are essential. The opportunity ahead depends on an ecosystem working together to handle greater complexity while maintaining quality and execution. Cohu is a partner we have worked with for more than 25 years, and we rely on their support for the growth ahead. We value the relationship and look forward to continued innovation together. Thank you.
I want to start by saying thanks to Keivan. It's always nice to get a customer's perspective of what we provide and where the market's growing. There are a couple things that Keivan said during his presentation which resonate with me. AI demand is growing exponentially. These products are becoming more complex, and there's a rising bar in the quality requirements. Those are exciting to me in the business that I run, as well as the products that we bring to market, because it ties back to what we deliver. Today's story that I'll walk you through comes down to three things, both in HPC tests as well as inspection. Why we can win with differentiated technology, system architecture, and know-how, where we can grow in both test and inspection, and how we can create long-lasting value through upgrades, software, and services beyond the initial equipment sale.
My name is Troy Taylor. I've been with Cohu for over 25 years. I'm the vice president of our emerging products group. Over my tenure at Cohu, I've developed equipment solutions for successive products. Today, let me walk you through that story I mentioned. You've likely seen our update from this morning where we've announced yet another win in HPC test. That's another funnel opportunity checked off. HPC gives us the opportunity to improve our growth profile in both systems and recurring revenue. The investment is not simply that AI requires more chips, it's that increasingly demanding devices need more capable solutions. You heard that from Keivan as well. This creates opportunity for us beyond the initial equipment sale. Our HPC pipeline represents a $1 billion opportunity with $770 million in systems and $230 million in recurring, and that mix matters.
Recurring provides an opportunity for us to support our customers with services, software, and aftermarket offerings to help them with their production requirements. With the additional win that we announced this morning, our qualified category now represents $210 million in systems and $60 million in recurring. We have another $440 million in the qualification category with further opportunities beyond that. Again, Keivan mentioned increasing complexity and raising the quality bar. That's leading to longer test times on these devices. An example would be devices today ranging from 30-minute test times to, say, 90-minute test times. To put that in perspective, that means, let's take the example of a 30-minute test time solution. That means one system is testing one device every 30 minutes. So every system is outputting two devices per hour.
Combine that with complexity, quality requirements, and demand, and that's how we come to our opportunity funnel. What makes this attractive, again, is that AI not only needs more chips, but the increasing demand on those chips, greater test intensity, and more complex thermal requirements. That puts greater value on the capabilities that Cohu brings to the market, turning technical relevance into qualification, production deployments, and continuing revenue. Let me connect that pipeline to customer requirements that are driving it and to the thermal and execution capabilities that we need to turn that opportunity into growth. We believe AI is increasing the value of semiconductor tests. Each new generation of devices requires greater thermal control, more data, and greater correlation between test conditions and real-world operation. Active thermal control is about managing the temperature of the device throughout the test as workload changes. It's not simply providing cooling.
There are three trends behind these needs. Processor power is rising. Today, we already see devices at 4 kW, with several customers talking about roadmaps at 12 kW by the end of next year. Advanced packaging is driving higher power density, and that's making heat transfer more difficult. Processors are becoming larger and more complex, and as Keivan Keshvari's slide showed, that creates hot spots around the package, and that's driving the need for multi-zone active thermal control. Customers need both sufficient thermal capacity and control as power levels change during tests, and that puts active thermal control on the critical path for testing advanced compute devices. That's why we offer dynamic multi-zone thermal control. It's what we believe makes Eclipse particularly relevant as AI processors become more demanding. We estimate a $1.5 billion serviceable market, growing at a compound annual rate of 36% based on our internal modeling.
These applications, as the slide shows, include CPUs, GPUs, custom ASICs, and network processors, serving hyperscalers and fabless customers around the world. Our competitive assessment positions Eclipse as the only solution on the market with a 6 kW capability in dynamic multi-zone thermal control and a roadmap to 12 kW by the end of next year. Sorry. Our first priority is winning more AI and advanced package test opportunities, bringing them from qualification and into production. But customers also need manufacturing capacity, supply chain execution, and application support, and Cohu brings those capabilities together globally. It's important. Customers don't only need thermal performance and thermal control. They need a partner, a partner to support accelerated ramps and worldwide deployment. As customers' requirements evolve, we plan to expand the value that we provide through upgrades, services, alongside with recurring software and aftermarket revenue.
Our objective is a longer-term production relationship with our customers, rather than ending our engagement after the initial system sale. Our growth plan is to translate differentiated thermal capabilities into qualifications, production ramps, and extend those relationships ongoing. The expanding market requirements create the opening, while execution determines how much of that opportunity turns into revenue. Thermal performance under these demanding conditions is the foundation of the Eclipse architecture, and it gives our customers paths to upgrade as their requirements evolve. The next question is how that thermal technology performs under these demanding conditions. This is a slide of actual data taken on a production handler, and what you're looking at here is power levels on the device changing throughout the test program. What's important for our customers is that the thermal control system tightly controls the guard band of that package throughout the test.
Thermal capacity matters because it enables power-hungry devices to dissipate that energy. But control matters because it's what allows us to keep the temperature of the device within a tight guard band. Together, this connects thermal performance to test quality, test efficiency, and time to production. It's not simply a production spec on a product sheet. What's important here is the dynamic power spikes. Those dynamic power spikes create hotspots on the package, and our ability to remove that power quickly enables us to control the guard band. Our ability to do that with multi-zone thermal control allows us to control each area of the package with different hotspots at different times and control the package to an overall temperature guard band. The graph illustrates that fast thermal control is important to keep those devices within the limits.
Without adequate thermal control, customers might have to reduce the amount of testing that they can do in parallel. What does that mean? It means longer test times, it means higher cost of test, and ultimately, it means potential quality escapes because you're not testing the part in real-world operation. On a CPU device example, we expect between 5% to 10% improvement in test times with faster thermal control, and that matters. Equally important is dynamic multi-zone thermal control, because again, it allows our customers to test the devices in representative operating conditions and support better test coverage and better quality. The takeaway here is Eclipse provides the combination of high-power capability, dynamic thermal control, and a production-proven platform, giving our AI compute growth opportunity a concrete technical foundation. Now, let me connect that to broader customer adoption and longer-term value of Eclipse.
Eclipse gives us the production-proven foundation, but that's just the starting point. From there, configuration upgrades, broader AI and HPC adoption, and software attachment create opportunities to deepen those relationships with our customer beyond the initial deployment. We see upgrade and aftermarket revenue representing about 20%-30% of the value as customer product lifecycle changes versus the original equipment sale of a Tri-Temp handler. We see upgrades and aftermarket expanding our connection to the customer. The expanding number of qualifications indicates that customers recognize the value of that configurability and adaptability and upgradability. It's evidence of our engagement and adoption potential. PAICe prescriptive analytics adds yet another dimension, recurring software value and deeper customer engagements. Alongside equipment and upgrades, PAICe creates additional ways to support our customers during production. These are complementary opportunities to extend the relationships beyond that initial purchase.
When a customer selects a piece of equipment, they need to answer three questions. One, can it meet the demanding test requirements? Two, does it fit the production needs, and can it be supported as volumes ramp? And three, how can it adapt as devices evolve? Cohu's value proposition addresses all three of these. Thermal capability makes Eclipse relevant to demanding applications, and its production foundation and global support help customers deploy it. The upgrade path extends its usefulness beyond the needs of just one generation. There's more to its selection than a simple one-line power specification on a product sheet. The takeaway here is simple. Eclipse creates opportunities to expand customer value well beyond that initial sale. Now, how do customers build on their original investment on Eclipse? I'll walk you through that next. This slide shows the upgradability and configurability and versatility of the Eclipse architecture.
Customers can start with the capabilities that they need today and upgrade as their requirements increase. That is the central value of Eclipse architecture. It is providing a path for customers to evolve their test capability rather than treating each new application as a new platform decision. Again, configurability and upgradability. The slide shows our versatility. Passive cooling for low-power processors and mobile to air-cooled active thermal control for 30 to 50-watt applications, moving to liquid-cooled active thermal control for 50 to 700-watt compute devices, all the way to AI and HPC data center applications requiring Tri-Temp active thermal control for devices anywhere from 100 watts to more than 6,000 watts. Different applications need different configuration, and that is the value that Eclipse brings with interchangeable cooling solutions on a common platform. For customers, that means they can extend the usefulness of their platform and reduce replacement friction.
For Cohu, field upgrades create additional revenue opportunity throughout the platform's life cycle. The customer benefit and business opportunity are linked. Continued value from the equipment supports continued partnership. The investment thesis is straightforward. AI and HPC create more demanding test requirements. Eclipse brings differentiated thermal capability on a production-proven platform, and the upgradability extends the value beyond that initial sale for both customers and Cohu. Our task is to take that relevance and convert it into qualifications, deployments, and continued revenue.
That is the foundation of our growth strategy. Next, I am going to continue to our second growth platform, package inspection. AI compute doesn't stop at the processor. It is driving greater demand with high-bandwidth memory devices and increasing inspection challenges associated with complex high-value devices. HBM and advanced package inspection increase both market opportunity and inspection challenge. Our internal model estimates a 34% compound annual growth for inspection metrology market.
Capturing it means meeting a demanding customer requirement, providing confidence in defect identification on a complex high-value device without sacrificing throughput, and that is critical, without sacrificing throughput. The combination matters, and it is our value proposition, high-end vision technology on a platform that adapts to how customers manufacture their devices. Neon is designed to deliver both strong vision and automation efficiency on a single platform. We see strong customer interest, and a key differentiator of Neon's ability is to perform singulated inspection. Neon inspects six sides of the package, top, bottom, and four sides of either individual die or stack. We have visible optical, infrared, and luminescence imaging technology providing complementary inspection methods. We have AI-based inspection to support yield optimization through better classification, reducing over-rejection of good devices and under-rejection of defective ones. The objective is to improve inspection decisions, not simply collect more images.
The result is advanced image inspection capability designed for a production environment where every device matters and needs to be evaluated efficiently. I mentioned earlier Neon's second differentiation, configurability. Is this the right slide? Yeah, configurability. Limited platform configurability can become a constraint when customer processes change, but Neon's configurability enables customers to adapt Neon without requiring new products for each application. Multiple inputs and outputs are just one example of configurability features, and that includes as many as six outputs for sorting bin classes. Neon's flexibility also allows customers to change and adapt their processes while protecting their capital investment. Flexibility extends beyond a single use case. Configurations are being developed for customers in HBM and advanced packaging and die sort applications. With Neon, we already support elements of automated material handling, and we plan to release fully lights-out support for production by the end of 2027.
The technology story is broader than crack detection. Neon integrates that visible infrared luminescence inspection all in one platform, and that includes 3D ball metrology. Complex HBM devices and advanced packages demand one-micron-level crack detection and comprehensive metrology. Neon brings this all together in one platform, one configurable platform, to address yield-critical defects and support customer quality requirements. Very small defects can be identified using wafer-level sampling techniques, but the challenge is detecting those devices on individual devices while maintaining production-level output. That's why our objective is to combine greater sensitivity with 100% device inspection. Our capabilities draw on internally developed intellectual property, deep inspection expertise within our organization, and collaboration with specialized partners in advanced lighting and optics. Neon gives us several avenues for growth. As HBM capacity and inspection requirements increase, we can expand the install base.
Its configurable architecture allows us to serve additional customers without having to develop a completely new platform for every application. The install base creates further growth opportunities as customer requirements evolve, platform upgrades, expanding inspection capabilities, and software attachment. PAICe Inspection adds another recurring software value and deeper customer engagement, again, extending that opportunity beyond the initial product sale. For customers, Neon combines advanced inspection, process flexibility, and automation on a configurable platform. For Cohu, that creates opportunities to serve more applications and build value throughout the product life cycle. It's not just that initial equipment purchase that matters. In summary, Neon provides three important advantages. First, every device inspection with the visible infrared, true 3D, and luminescence capabilities, all on one platform, supported by AI defect classification. Second, inspection performance and process configurability are provided together on one platform.
Lastly, we have a technology roadmap and install base opportunity that positions Cohu to participate in the continued growth of HBM and advanced package inspection. I walked you through the story, and by now, it should be clear. We have three things we talked about: winning through differentiation, configurability, technology, and system architecture, growing system recurring and recurring revenue in both HPC test and inspection for systems, and creating long-lasting value through upgrades, software, and services beyond that initial sale. I will now turn the presentation over to Ian Mazsa, who will discuss data center high-power management platform.
Thanks, Troy. Good morning, and thanks for joining us. My name is Ian Mazsa. I am the Vice President of Semiconductor Test at Cohu. I joined in 2021, and today, I'm pleased to share with you how we are positioning our semiconductor test business for its next phase of growth. I'd like to discuss compelling growth opportunities within semiconductor test, the convergence of AI infrastructure and advanced power management that drives it. From data centers to intelligent edge devices, AI increases semiconductor content, complexity, and power control requirements. Customers need higher currents, higher voltages, and greater parallelism. These requirements push beyond the capabilities of some legacy air-cooled systems. Our Diamondx platform is the platform customers need today for the AI future. Over the next few minutes, I'll discuss the market opportunity, our competitive positioning, and how we're transitioning these trends into a growing revenue pipeline.
Let's look more closely at the market dynamics driving this opportunity. AI data centers require dramatically higher power density than conventional computing environments. That is accelerating adoption of gallium nitride, or GaN, in a $300 million segment for test of power devices across the power distribution network. Customers require higher test parallelism to maintain throughput while managing that increased complexity. As the cost of manufacturing GaN falls, it will expand the addressable opportunity beyond traditional power testing and across microcontrollers and advanced power modules, for example. Our execution strategy is straightforward and meant to take advantage of these trends. First, expand Diamondx within mixed-signal accounts where legacy testers struggle to meet customers' requirements for high-volume GaN testing, the immediate pain. Second, secure both wafer test and final test insertions, particularly at OSATs that value flexibility and scalability. Give choice.
Third, deploy AI-augmented workflows to improve productivity and utilization across a complete test cell. Let me say that again, a complete test cell, including tester, handler, and docking equipment. You give choice. Finally, focus our applications and technical support resources on the customers that will win, the win in power management for high-performance computing today and physical AI in the future. The takeaway is that we have products to solve these winning customers' challenges today. Diamondx's unique differentiation is the combination of test coverage for complex devices alongside the capabilities required for high-power testing at high throughput. That gives customers flexibility across high-power semiconductor and module power management applications. Diamondx supports this breadth with up to 1,000 amps of pulse current. Applications include high-power RDS(on) measurements, GPU point of load power transient testing, and testing emergent 800-volt power conversion modules for next-generation AI infrastructure.
Together, these illustrate the grid-to-core opportunity, power conversion from the utility grid to the high-performance computing socket. Customers need more than a peak current specification. They need to optimize test cell throughput across different device power levels. Diamondx combines precise power control, accurate measurements, and the flexibility to test multiple devices at once, or combine testing resources for more demanding applications, from individual power devices to complex power modules. We have invested in Diamondx's power capabilities to compete for this grid-to-core application space today and support customers as their test requirements continue to evolve. In summary, Diamondx gives us a test solution to participate in AI-driven semiconductor growth. Diamondx addresses two opportunities: offloading digital tests from high-performance computing automated test equipment, and also accelerating GaN testing across increasingly complex high-power grid-to-core conversion paths.
Agentic tools help generate test cell solutions and reduce what used to be a very high bar, a very sticky situation, into more utility to pick the right tester, the right platform for the economics of your device. Our analytics support factory automation and continuous yield improvement. Together, these capabilities help integrate device manufacturers and outsourced assembly test providers as they prepare for more devices with neural processing units, bringing AI inference to the edge. For Cohu, these capabilities create three paths to growth: new system sales, upgrades, and recurring service and software revenue. That is why we expect semiconductor tests to become an increasingly important contributor to Cohu's long-term growth and profitability. I will now turn it over to Chris Bohrson for our recurring business.
All right. Thank you, Ian. Hello, everyone. As you can see on the slide, my name is Chris Bohrson. I lead Cohu's global lifecycle recurring business, and I really want to cover three topics here today. First, I will give you a little bit of background on the recurring business, just to put the rest of the presentation into context. Then I really want to spend my time focusing in on two, what we see as exciting or interesting growth areas in that recurring business around our contactors and around our software. Goal with this slide, I wanted to give you a representation of the breadth of Cohu's recurring business. I have tried to do that by, you see these eight boxes on the outside of the test cell on the slide.
Hopefully this does a good job of representing the complete portfolio of the products and services that we provide. These products, by the way, I think enable customers to We support the equipment throughout the lifecycle of the equipment. The goal is really to enable the customers to drive higher productivity of that installed base over its lifetime. As you can see on the slide, the number of systems we have installed globally now is more than 25,000. I have also tried to show two growth areas in the orange boxes. You can see that they represent about a $250 million SAM, or served available market. First is, if you look in the upper left-hand side of the slide, you will see the Hydra series of contactors.
That opportunity builds on the growing complexity for testing these high-performance computing semiconductors, including co-packaged optics, which I will sometimes refer to just as CPO. Second area is our software platform on the upper right-hand side. That provides, as you heard Luis talk about, the AI-driven data analytics. It pulls data off Cohu's systems as well as other sources of data on the test floor, and then it translates that into actionable insights for customers. Again, it helps customers improve the productivity of the installed base of our equipment as well as the entire testing process. Let us start first with the Hydra contactors. Again, this is a product line we call Hydra. The Hydra series is targeted at the growing market for, again, for testing high-performance computing semiconductors.
As you can see on the slide, the estimated SAM that we have for Cohu for that is about $200 million annually. Again, these devices are driving much higher testing complexity, so what customers are telling us is the contactors are becoming an increasingly important part of the test cell, I should say critical part of the test cell, and therefore creating an attractive recurring revenue opportunity for Cohu. Now, I also want to bring your attention to the lower right-hand side of the slide, where I have highlighted the co-packaged optics, or CPO. These present a particularly challenging set of test requirements, which I will get into more detail on the next slide. But overall, I think Cohu is very well positioned to participate in testing these advanced devices, and that is both for engineering characterization as well as high-volume production.
The strength that the company brings is the combination of the interface products that we have, the automated handling systems, as well as we wrap all that around with our multi-domain expertise. When I describe multi-domain expertise, I am really referring to our depth of knowledge in electrical testing, depth of knowledge in tightly controlling the temperature of the device under test, as you heard Troy talk about, and then now the new requirement for these devices for optical interfacing. I promised to talk about co-packaged optic devices, and again, these present a particularly challenging set of test requirements. To explain that, I thought it would be useful to describe that in two questions that I often get about this. The first is, why are these devices so challenging to test?
Well, the simple story is they introduce more complex handling, more complex interfacing, and more complex testing requirements. Specifically, they require very precise mechanical and electrical interfacing. They require very precise optical alignment and coupling on up to four sides of the package. Then they obviously provide, as Troy talked about, the very tight, precise temperature control of the device under test. Given all that, what does Cohu bring to the table? What unique capabilities do we provide? Again, I will continue to describe that as multi-domain expertise, as well as providing an integrated solution. Specifically, we provide the Cohu handler, the Eclipse handler with active thermal control. We provide a device-specific kit for precise mechanical alignment. We provide a high-density, high-speed Hydra CPO contactor for electrical connectivity.
Then precise optical alignment and coupling to the device under test for CPO devices, and then obviously integration to the customer's test system of choice. If you look at the lower left-hand side of the slide, you will see a little footnote, and per that footnote, we have actually applied for a patent in this area, and that includes the unique capability we have for aligning and coupling the fiber array to and from the optical engine on the module for, or on the device, including for high-volume manufacturing. The opportunity from my point of view for Cohu is to provide a complete handling and interface solution for this. Then the opportunity for my team, of course, is to grow revenue with the Hydra series of contactors, and that includes, by the way, of aligning that contactor growth with the growth of the Eclipse platform that Troy talked about.
All right, the second area of growth I want to talk about is our PAICe software platform. PAICe incorporates machine learning, generative AI, and agentic AI, again, with the goal of pulling data off the test floor, including off of our equipment, and translating that into actionable insights for the customer so they can take action to fix whatever the issue is on their test floor. To date, we are looking at providing three buckets of value to the customer, targeting that $50 million served available market you see on the slide. First bucket of value is around improved yield. We have a product, as you can see on the slide, called PAICe Inspection. PAICe Inspection provides AI inspection actually in line with the high-volume inspection process.
The goal there is to reduce false defects, as well as to make sure they have no escapes in the no quality escapes during the production or the inspection process. Customers have told us what we've seen from their use of the product in high-volume manufacturing is up to a 1% increase in yield on the test floor. Second bucket of value is really around increased uptime for the equipment. Again, you can see on the slide, we have a product called PAICe Prescriptive. PAICe monitors the equipment signals coming off of our equipment. It looks for patterns of normal behavior of the equipment, and then of course looks for patterns of abnormal behavior, what the issue is. It then combines that analysis with other documentation sources, both from ourselves and from the customer.
The goal is to provide corrective actions to the customer to tell them specifically what they have to do to correct the problem on their test floor. Again, the goal is reducing unplanned downtime. The third bucket of value is really around increased test cell productivity. Again, you can see on the slide, we have a product called Test Operations that's actually in development. Think of Test Operations as an AI troubleshooting tool for the test cell. I forgot to mention one thing about Prescriptive. I want to go back to this because it's very important. We actually chose the name Prescriptive very consciously because it demonstrates the differentiation of the product. It moves the customer from purely reactive, obviously, but even beyond predictive to prescriptive, meaning it tells the customer specifically what they have to do to correct the problem.
Customers have told us, that's one of the most valuable aspects of the product, so I wanted to make sure I covered that. Again, customers have told us they've seen up to a 45% reduction in unplanned downtime from use of the product on their test floor. Back to Test Operations. Again, an AI troubleshooting tool for the test floor. It gathers data from multiple sources on the test floor. It diagnoses issues, including production bottlenecks. It then automates the analysis to correct the problem, which engineers used to have to do, and therefore reduces the amount of time it takes to resolve the problem, as well as it frees up engineers to work on other perhaps more important aspects of more challenging problems. I also want to bring your attention to the Eclipse reference on the lower right-hand side of the slide.
You heard Troy talk about the importance of Eclipse for thermal management on the test floor. Turns out what customers tell us is dealing with thermal problems on the test floor is one of the most challenging aspects of what they do. Sometimes it can take them days or even weeks to resolve the problem. We've got a specific implementation of Eclipse to help them automate that, resolving the problem, which we've seen at least indications right now, up to now, they're able to reduce that time to resolve issues from maybe weeks or days down to less than a day or even hours. That obviously reduces the cycle time of resolving the problem as well as frees up. They don't need a thermal engineer to actually do the work. They don't need an expert.
Summary, PAICe delivers three buckets of value, improved yield, improved uptime of the equipment, and then improving test cell productivity. It obviously provides also an attractive recurring revenue opportunity for our entire installed base of systems, but specifically also for Eclipse. It is a subscription-based product, so it obviously adds to our recurring revenue, growing revenue stream. If I have done a good job here today, hopefully you will get four takeaways from this. First is we have 25,000 total systems installed globally. That provides a very solid foundation for our recurring revenue through the life cycle of the product. Second, the increasing device complexity for high-performance computing devices, including co-packaged optics devices, creates a $200 million opportunity for our Hydra series of contactors. Third, PAICe, our software platform, targets a $50 million opportunity for improving yield, improving uptime, and improving test cell productivity.
And fourth, we are aligning both Hydra and PAICe to increase Eclipse differentiation for high-performance computing, and obviously to drive further growth in our recurring business tied to the Eclipse platform. I think these opportunities together demonstrate how we are building on Cohu's installed base. We are leveraging our technical expertise. We are leveraging our customer relationships to deliver more value, but also to generate more recurring revenue. Okay, so with that, I will turn it over to There we go. I will turn it over to Jeff Jones. He will discuss how we will reflect these growth opportunities in our financial model and how we are, our approach to capital allocation.
Thank you, Chris. Hello, everyone. Thank you again for joining us today. As Chris said, I am Jeff Jones, Cohu's CFO. You have already heard from my colleagues today about markets and products and strategies. My objective today is to connect all of that into our new financial target model and to show you how we plan to allocate capital, support growth, and shareholder value. In doing so, I will cover five things, obviously. First off, I will walk through the new financial model. This is over the midterm, which to us means three to five years. The headline here is that we plan to more than double revenue and significantly grow earnings per share. Number two, I will walk through the growth drivers behind the increase in revenue, and then I will talk a little bit more about our steady, resilient recurring revenue.
Number four, I will review our track record of margin expansion, and how we plan to hit the gross margin target. And five, I will finish with operating leverage and our plans for capital allocation. Our objective is disciplined, profitable growth, not growth at any cost. We intend to build a larger, more profitable, and more resilient Cohu by combining systems and revenue growth, recurring revenue growth, I should say, improve earnings per share of the business, and maintain the financial flexibility to invest through the cycles. So, let us turn to the new model. Be the wrong way. Surprise. Okay. Over the midterm, we plan to more than double Cohu's revenue from an estimated $625 million in fiscal 2026 to $1.3 billion. And we expect the model to deliver significant cost leverage, taking non-GAAP EPS from about $1 to $5.
For fiscal 2026, we estimate revenue of approximately $625 million, gross margin of about 45%, operating expenses of about 35%, and non-GAAP EPS of about $1. Looking ahead to fiscal 2027, we're projecting revenue growth of 30%-40% year over year to a range of $810 million-$875 million, and that step-up comes from the growth programs we've already talked about today. Our midterm model is $1.3 billion in revenue, a 50% gross margin, operating expenses at 22%, and $5 of EPS. The revenue growth is supported by multiple broad drivers. Expansion in HPC, growth in recurring revenue, increasing demand in our core markets, and new opportunities in emerging strategic markets, as you heard from my colleagues. Operating expense as a percentage of revenue goes from 35% this year to 22% over the midterm. That reflects operating leverage. We're not planning to cut investment.
We're planning to grow revenue much faster than our cost base. The result is EPS of about $1 this year and $5 in the midterm model. As you can see, the revenue mix changes. This year, recurring revenue is about 52% of total, and systems about 48%. At $1.3 billion, it shifts to roughly 60% systems and 40% recurring. This doesn't mean recurring revenue declines. It means systems revenue grows faster as the new platforms scale, while recurring revenue continues to expand in absolute dollars and provides resilience and attractive margin contribution. Profitability comes from both mix and scale. You heard Troy and others talk about differentiated products, as well as our recurring offerings, which support gross margin, while operating expenses grow more slowly than revenue. That's the leverage that takes us from $1 EPS in fiscal 2026 toward our $5 midterm target.
Now, let's head over to the revenue growth drivers. This chart clearly shows the major drivers behind roughly a $675 million incremental revenue opportunity. HPC systems are the largest contributor at about 60% of the total growth. As you've heard from Troy, our Eclipse opportunity funnel is strong and more than capable of supporting this level of growth. Our core systems contribute about 15% of the growth. This reflects systems growth as industrial, automotive, mobile, and other established end markets recover and expand. The assumption is a more normal demand environment for our core business. Recurring revenue accounts for about 25% of the growth. More Eclipse, Neon, Diamondx systems in our customers' test facilities increase the foundation for more consumables, spare parts, and service revenue.
Additionally, co-packaged optics, interface products, and AI analytics, as you heard from Chris, are emerging and strategic recurring opportunities built around the Hydra and PAICe products. Together, these three main growth vectors bridge the model to our $1.3 billion target. Next, talk about Cohu's steady and resilient recurring business. As you can see, at the end of Q3, our trailing 12-month revenue was $309 million, about 54% of our consolidated revenue. Most of it is consumables, which wear out as customers run production, so our revenue tracks with equipment usage. Recurring revenue has grown sequentially for the last seven quarters, while customers' equipment utilization rate rose 10 points to 82% at the end of Q3. We don't forecast utilization, but in prior cycles, it has run into the mid to a high 80% range.
Almost half of the recurring revenue comes from annual service contracts, and an 88% renewal rate indicates that customers value them. We also run cloud-based automated ordering that covers roughly 14,000 spare parts, keeps us deeply embedded in our customers' workflow. What drives all of this, as you've heard from Luis and from Chris, about 25,000 systems at 108 customers across 279 volume facilities in 30 countries. Systems typically run 10 plus years, so as customers adopt Eclipse, Neon, and Diamondx, revenue from service, upgrades, contactors, components, and analytics grow with the base. We support the installed base with 440 field engineers, which you heard extremely important, as Troy described, and their average tenure is 13 years, and that's the experience and the knowledge that our customers require. What does this mean for the financial model? Well, three things.
One, it gives us stable base of revenue and cash flow that cushions soft systems demand through the industry cycles. Number two, it supports gross margin and our margin contribution target. And three, it grows with our strategy because every system we install adds to the base that we serve. Eclipse is a good example. At its heart, a complex active thermal control module, which creates a real opportunity for field service, including replacements and upgrades. Now let me talk a little bit further on gross margin. As you can see, Cohu has a solid track record of growing gross margin as our portfolio, mix, and operating model improve. From fiscal 2015 through prior target periods, we gained about 10 percentage points of non-GAAP gross margin. Now, as you can see, progress isn't a straight line.
Gross margin moves with revenue level, product mix, factory utilization, supply chain conditions, and market cycles. We expect about 45% gross margin this year, and our midterm target is 50%. The chart also shows targets we set in prior years. So you can see, we set a number and then we work to deliver it. How do we go from 45% to 50%? It's four items, four drivers. First, you heard from my colleagues, differentiated products. We compete on technology and performance, which supports pricing and value. Second is recurring revenue. Consumables, service, and software typically carry higher margins than systems. And third is operating leverage. As volume goes up on a largely fixed manufacturing footprint, we absorb manufacturing overhead more efficiently. And we're adding Eclipse manufacturing capacity in Malaysia and Philippines ahead of the demand, and these expansions will finish in the first half of 2027.
With the related equipment, we're raising our 2027 CapEx target from 2% to 3% of revenue, which is still easily absorbed on a much larger revenue base. And then fourth is our supply chain optimization. Better sourcing, logistics, and manufacturing efficiency, plus scale, help us cut costs as new platforms move into higher volume production. These levers are largely in our control and consistent with what has worked in the past. Going from 45% to 50% is a five-point step, and we've already delivered about 10 points before. That's why we're confident in this model. Gross margin's only the first part of the profitability story, so let's look what happens below the gross margin line. The midterm model creates significant operating leverage. Our objective is approximately double the revenue while targeting roughly a sixfold increase in operating income and corresponding growth in earnings per share.
The first driver is gross profit. Higher revenue and expanding gross margin give us a much larger pool of gross profit to fund innovation and increase profitability. The second driver is operating expense discipline. We will keep investing in research and development, particularly for high-performance computing, inspection, power, test interfaces, and analytics. Internally, we are realigning resources historically supporting our core business to now focus on HPC products and customers. This cost realignment reduces the overall increase in operating expenses and better leverages profitability as revenue grows. The model includes approximately 30% increase in HPC-related research and development investment. At the same time, total operating expenses declines as a percent of revenue as the business scales. Selling general and administrative expense also benefits from scale because we do not need to replicate the full support infrastructure as our revenue grows. More incremental gross profit reaches operating income.
The model targets approximately 45% operating fall-through. That means that for each additional dollar of revenue, approximately 45%, 45 cents, drops to operating income thanks to higher gross margin and largely fixed cost infrastructure. EPS grows in line with operating income from about $1 in 2026 to $5 in the midterm model. If you connect the last three slides, the story is consistent. Revenue grows from $625 million to $1.3 billion. Gross margin moves from 45% to 50%. Operating expense drops from 35% to 22% of revenue. Combined, operating income goes from about $65 million to $370 million, and EPS from about $1 to $5. That level of profitability creates significant cash flow, so let us turn to capital allocations. We are starting from a position of strength with $508 million in cash at the end of Q3.
That gives us the flexibility to invest, to return capital, and to manage our balance sheet through the cycle. Let me walk through our five priorities. First, capital expenditures. As I mentioned, we are expanding in Malaysia and the Philippines across fiscal 2026 and 2027 to support future high-performance computing growth. It is an investment ahead of demand so we can deliver when our customers need us to. Including this expansion, we estimate our 2027 CapEx to be about 3% of revenue. It is a fairly asset-light model, which helps free cash flow conversion as revenue grows. Second, R&D. Our first priority for cash is funding organic investment in growth markets. We have invested about $800 million from fiscal 2016 through Q3 of this year.
That is what has built the product portfolio behind the growth drivers we reviewed earlier, and we intend to keep funding it, as I noted on a previous slide. Third is M&A. Our focus is on acquisitions that expand our serviceable addressable market and our technology portfolio in HPC and in recurring revenue. We are not pursuing acquisitions for scale alone. The test is whether a deal strengthens our strategic position and accelerates progress toward the financial targets in the midterm model. Fourth, share repurchase. The board previously authorized $140 million, and through Q3 of this year, we have repurchased about 4 million shares for $117 million. The remaining authorization gives us a way to potentially return cash to shareholders while still funding growth. Fifth is debt. Our cash of $508 million more than covers our total debt of $310 million as of the end of Q3.
Of that, $288 million is convertible debt, which matures in January 2031, so we have no near-term maturity pressure. To summarize, we have a clear path to more than double revenue to $1.3 billion, expand gross margin to 50%, reduce operating expense to 22% of sales, and reach $5 of EPS. We have a resilient recurring base, a record of delivering margin expansion, and a balance sheet that lets us invest and return capital.
I want to thank you, as that concludes our presentations, and we will take a quick five-minute break so we can set up chairs on stage here and move to Q&A. Why don't we reconvene here at approximately quarter till? Start the Q&A session. We have got a couple of microphones here. When you are ready for a question, raise your hand, and then somebody will run over and hand you a mic. Sorry, Christian, I saw Craig's hand go up first.
No worries.
Craig Ellis, B. Riley Securities. Let me start by saying, really appreciate the detailed updates that you guys have given us and getting that from across the business. We often get a chance to inquire to Jeff and Luis, so I am going to start with our GMs, and I will start with Troy. Troy, you showed a slide that showed customers wanting to engage at 12 kW. Can you talk about the breadth of customer engagement there for those chips that are that real high power intensity? When do you see that really moving into solution uptake, given how we would think about AI data center product roadmaps?
Sure. The first part of the question is, what is the breadth of customers? I think different customers are at different stages of their products. You see GPU powers going up quicker than CPU power. I will not talk about a specific customer, but it is a general trend as these advanced packages are getting bigger, just more compute power being packaged in them across the board, not a specific market, just across the board. In terms of when do we expect these to go into production, the way we see it is customers coming out with products that need to start testing towards the end of next year and into production the following year.
That is really helpful. Then the follow-up is to Ian. Ian, really liked seeing the points that you made on GaN in the data center, because a lot of the field work we are doing shows that is going to be essential in 800-volt racks. So, the question to you is somewhat similar. As we think about GaN and we think about what Cohu is doing there, when do we see the real system inflection related to that, and how broad would you say that that customer engagement is? Thank you.
The customer engagement across GaN is everyone at this point. I would say everyone in power management either is manufacturing with it, using it, repackaging it, developing, and has developed a strategy for going to market with it. Where we see it very publicly intersecting HPC is in the point of load. I think over time, it will work its way back more to IBCs, the intermediate bus conversion, as well as the grid conversion stage. There is work that has to be done, right, to get to those higher voltages and that capability.
But the attraction of GaN being able to be much more compatible with existing CMOS processes, right, another layer in the lithography stage to that simplicity, is really attractiveness compared to silicon carbide. I think the physics supports that GaN is going to get there. It is going to take time, but I think Diamondx hits the spot quite nicely with the ability to do it today and scale to higher site counts in the future, which is going to help reduce the cost of test and make GaN more ubiquitous going forward. It is going to take time to equal CMOS volumes, but it has a path to it.
Great. Thanks.
Thank you.
Jeff, Christian Schwab, Craig-Hallum. Can you just give us the rough mix of revenue in 2026 between the product families of Eclipse, Neon, and Diamondx?
Yeah, sure. The Eclipse, I think we talked about $120 million in HPC in 2026, with roughly about $100 million coming out of the Eclipse product and about roughly $20 million coming from the HBM inspection product.
The rest would be Diamond?
No, I mean, Diamondx and the tester revenue would be roughly in the $80 million-$90 million this year. You've got a recurring business that's quite large, as we talked about, over $300 million for the year. I think those are. Then you've got the core handler system business, which would make up that difference there to the 625.
Great. As a follow-up to that, as we move to our target model, it appears that the Eclipse is the driving family of that. Can you just remind us of your visibility? We did give 2027 guidance already for the year, so let's. Sometimes you guys give multiple quarters. Now we're going to give five quarters. So visibility must be extremely strong?
Yeah, we have a much higher level of confidence in the opportunities that we're seeing primarily in high-performance compute, AI compute, with Eclipse, right? That whole funnel opportunity that we've been talking about now for a few quarters, which has grown, I believe it came out at $750 million to begin with, and here we are at a billion-dollar opportunity. As we work through the funnel, moving from the different buckets, the engagement in qualification, the qualified, that gives us a lot more confidence and quite a bit more visibility to the opportunity over the next three, four, five quarters.
That was really the driving force behind the guidance for FY 2027. As you mentioned, the HPC aspect is the largest contributor, as I said, 60% from systems alone. Then there's revenue that's recurring revenue that's generated from those systems as well. Luis said, this is a unique opportunity for Cohu. We have not seen this over the last 20 years. I think what you're seeing in terms of guidance and the model is our confidence in that high-performance computing area.
My last question, just on the thermal handling Eclipse product line. Can whoever would like to take the question, discuss your product differentiation and competitive profile there that gives you confidence that nobody else will be able to come in and get in the way of your growth objectives?
Yeah, I think Troy can hit on what differentiates us from primarily the main competition. So you want to talk about that?
Sure. I think what's changing right now, and I mentioned this in the break to somebody, is what's good enough in the past is no longer good enough. What we see is the level of power increasing fast, the requirements of dynamic thermal control increasing quickly, and that whole topic around hotspots. Keivan mentioned it in his presentation. We talk about multi-zone thermal control. What gives me confidence is that this is technology and IP that we've developed over 20-plus years, 25-plus years in our industry, historically supporting the IDMs where they have seen the value. Now we see other customers coming in and requiring those same capabilities that IDMs of the past have really required. So it's not, as I said, it's not just cooling, right? It's about dynamic multi-zone thermal control.
David, you want?
Yeah. Thanks for taking my question. I guess my first question has to do with the recurring revenue stream. I think you mentioned $309 million of trailing 12-month revenue in that area. How much now is the Greek monster Hydra?
I would say Hydra is a relatively new product, so it's probably single-digit millions in terms of revenue this year.
It seems like you are going to have, with the ramp-up of the Eclipse installed base, you are talking, I think, $200 million next year in sales or something like that. That is going to produce a lot of units. There is a recurring revenue stream attached with each unit. Could you just elaborate a little bit on that? It seems to me, after a couple years of selling a few hundred systems a year, that recurring revenue stream can be quite significant.
Yeah, Dave, you are right about that. I will let Troy just sort of describe the elements of the recurring revenue that are related to Eclipse, which is similar to other products, but I think there is one point in particular that should be emphasized, which is the active thermal control module and the heaters and so forth. You want to talk a little bit about that?
Yeah. Again, I think our market is changing a bit in terms of what customers expect for test results. Therefore, that is driving recurring revenue from new thermal solutions that are being provided and shipped to customers with new product families. There is new thermal technology and thermal heads that are being deployed. On top of that, device kits is another piece, system upgrades, features, options, as well as the software attachment that we talked about. All those things combined, I agree, with an increasing install base of systems, customers want to continue using that for next-generation packages. Therefore, they will continue to buy upgraded thermal solutions as their packages evolve.
If I may add to that, I think what Jeff is really alluding to is as the powers go up and devices change, they have a relatively short lifetime, 18 months, 24 months. It is going to be more common now that you have to replace the thermal head entirely or the thermal modules in the head to adapt to the new devices that are coming out, the geometry, the larger sizes, and the different power levels, the different power zones that the new devices will require to control. I think what we will see is an increased rotation or sale of upgrade thermal systems or thermal subsystems into our Eclipse handler, more than what we have seen up to now in our core handler platforms.
Just kind of summarize what you just said, they tend to wear out after 18-24 months.
Devices change.
Right. Each time a new device is introduced, you are going to have to have a new set of thermal heads. With the high-performance computing market on an annual cadence, you are going to have to be upgrading the thermal heads in these systems fairly regularly. It is a bit longer than an annual cadence for device life, right? I mean, you see new devices coming out year to year and h alf, but the old products will continue for a period of time. I would say more year and a half to two-year cadence, we start seeing thermal head upgrades, different geometries, different power zones. That is what we are seeing more often.
Okay, final question from me is you put out a press release this morning on another customer, but maybe you could just elaborate on that. Is it a networking customer or a GPU customer, or how significant do you think this particular win is?
If you noticed on the funnel, I don't know if you've had a chance to analyze the different buckets, but it was essentially a shift. They were already in the funnel. Maybe that's obvious, but they were already in the funnel, and it moved from this particular opportunity was in engagement and jumped down to qualification, qualified essentially. About $30 million in systems from the in-engagement bucket down to qualified, about $10 million in recurring. And I think it just goes to show that with each customer, the timing is different, the qualification process is a little bit different, the motivation to get to market is a little bit different. And in this case, as I said, they went from an engagement to you are now the plan of record. Make sense?
Yeah.
Okay. Denis?
Thank you very much. Denis Pyatchanin with Needham & Company. For my first question, maybe you could tell us more about active thermal control versus passive thermal control in the real world. How would the testing process look like with a Cohu product with ATC versus using a competitor with passive?
Yeah, I think that's for Troy again.
Lucky me. Passive thermal control goes back to that cooling, right? If you were to look at, if you remember that chart that I showed up on the slide, things during tests move very quickly, right? As workloads are changing. With active thermal control, we are actively quickly, and I mean quick, reacting to temperature changes on the device. With a passive cooling, you're not reacting at all. The device is going where it's going, and now as Keivan mentioned, right, tighter guard band controls, requirements for quality is becoming more critical.
Therefore, that high-speed response is what customers are now looking for. That chart that I showed you, when we show that to customers live in front of them, new customers in the funnel, the first thing they say is, "Wow, that's amazing." Right? I'm not over-exaggerating. "Wow, that's amazing. That's what we need." Right? I think it's those customers that are starting to see the value and the challenges that they're dealing with in their production side that are coming to us and saying, "Let's get into a qualification.
Great. For my second one, you'd mentioned there was a CPO test-related alignment patent. Could you tell us more about that and what advantage it brings to the system?
Chris, is that you? Yeah.
Yes, Denis. The advantage really is in the, for the testing the CPO devices, again, in the integrated solution. Specifically, the change that is occurring right now is the ability to actively align and couple the external communication path from the optical to the device under test. That requires both modification of the contactor, but also an innovation in terms of how you provide that optical connectivity mechanism, the mechanical mechanism during the actual testing process. That is both for engineering characterization as well as for high-volume production. That is part of what we have done with the, that I mentioned a reference to the patent we have applied for in terms of how we do that specifically.
For my last one, we will move to the recurring business. Can you tell us more about what are the largest components in that business and where you expect to get the most growth? Can we assume that most of this business is attached to your handlers as opposed to your testers?
Yeah, so the recurring. Sorry.
No, no. Yeah, go ahead.
Yeah, the recurring business is tied to all the systems. Obviously, our handler installed base is the largest, so it has the largest component of that in terms of just the raw recurring. As Troy mentioned, that includes a whole variety of things that I had on the slide between spares and repairs and upgrades and kits and contactors, et cetera. The area where we expect most of the growth of the two I mentioned, which were really the contactors for high-performance computing applications, and one of the areas specifically was for co-packaged optics devices. Secondly is the growth in the software platform that I mentioned for targeting yield, equipment productivity, test cell productivity, et cetera.
Thank you.
Thanks, Denis.
Hi, this is Vedvati from Evercore ISI. Thanks for taking my question. I wanted to double-click on services again. I understand with Eclipse, you generate high recurring content. Maybe could you help us size, like what kind of revenue attach you see per Eclipse handlers and what that used to be for your core handlers? Is there like a percentage attached that we can think about?
Well, I think it's probably easiest just to look at the funnel there, and we have an estimated recurring revenue portion of that, and I think it's 20%-25% of the total system. So that would be my guidance on the best view of the recurring. It's not that far out of line, or it's pretty similar, maybe a little bit higher than our core business, I think primarily because of the shorter life cycles and the more complexity of the thermal heads.
Understand. Then on gross margins, you have 5 percentage points of an improvement. You talked about the three levers that drive it. Could you maybe help us understand the contribution from each of those levers? The second one I have is, we're also in an environment where we are seeing increasing costs across the board, so how do you manage those input cost increases?
Yeah, the main driver there on the overall margin improvement will be the leverage that we get with the increase in scale. While we are currently increasing capacity and investing in our Malaysia and Philippines facilities, over the long term or over this midterm, the manufacturing costs are relatively fixed. So, as we flow more Eclipse through the factories, more interface products through the factories, then we more efficiently leverage those manufacturing costs. So, the main driver of that five-point increase is really leverage of those fixed costs. Now, the other aspects that I talked about, in terms of supply chain and mature platforms and things of that nature, they contribute as well, but I would just guide you to that the volume is the main driver in the margin contribution. Sorry.
Brian Chin from Stifel. Thank you for the analyst meeting and the opportunity to ask a few questions. Maybe firstly, you mentioned that the thermal handler SAM is $1.5 billion. You said you estimate it growing at a mid-30% CAGR. That would take that market in several years to like $4 billion. If I look at your $1.3 billion target model and approximate sort of how much of that is thermal test handler, maybe it is like $600 million, something like that. That would be like a 15% share of that potential SAM. That would seem kind of conservative relative to the momentum you have right now in the product and the potential for differentiation. There is nothing minuscule about doubling your revenue of a company over that horizon, but how conservative could that particular input be?
Well, thanks for the question, Brian. I would compare that to the opportunity funnel we have in front of us, which is really where our focus is. It is the $1 billion. We do not expect to get 100% of that over the midterm, but certainly a high percentage. So far, our track record has been excellent. That is a big market number, as you described. I think for us, it is more about focusing on what is in front of us. Could we be conservative in that number? Well, if it turns out to be a $5 billion market, then potentially could be. I think for now, the real emphasis and focus is on how do we capture as much of that $1 billion opportunity as possible.
Thanks. In terms of what triggers a new platform sale as opposed to an upgrade of an existing chassis or handler platform, would it be the size of the package in particular? Is that kind of what would trigger that? Also, in terms of the complexity for the technology roadmap as you try to intersect customers' future needs, which is more complex? Is it the power? Is it the size of the package and surface area that you are having to actively control? Is it either those dimensions and/or there is something else. I guess there is like a third dimension that seems kind of important there, but-
Yeah
I think the test time, just the test time against that larger surface area and the higher power.
So, Troy, you want to talk about upgrades and then complexity?
I think the answer is yes. Everything that you said drives decisions on platforms and upgrades. I think what's happening right now is just the increasing power and thermal requirements require something different, and that's opening these opportunities up. But yeah, power, package size, all that comes into play. If customers already have that equipment and they need to change from one product generation to another product generation, that might be upgrading a thermal system, but there's also the capacity increase, right? So, I talked about test times increasing. That comes into play as well in terms of platform decisions. So it's not necessarily, "Hey, I'm changing a platform." It's that I'm growing my install base to support my test times and my complexity in products.
Peter, you want to go?
Thank you. Peter Vogel with New Street Research. Question, as chips are getting more complex, can you just discuss what's happening between interface side on whether they're sockets or pins, load boards, and how that's tying into the equipment, and whether you'll sort of see more vertical integration than we've seen in the past?
Mm-hmm. Yeah. Chris, you want to talk about higher pin counts and power and so forth?
Yeah, so, sorry, Peter?
Peter, yes.
Yeah. Yeah, so that's a good point. We are actually seeing that. Again, as I mentioned earlier, the customers are interested in an integrated solution, so they're seeing higher pin counts, they're seeing higher performance devices, and therefore they are interested in a more integrated solution from a single supplier because of the risk and the time to market of being able to do all that. That doesn't say that our customers are still interested in kind of best in class in both areas, but I think we are seeing an increasing trend towards customers saying to us. We don't want to take the risk. We want to cut down the cycle time, so let's look at putting together an integrated solution for those types of applications.
And maybe a follow-
Peter, you're going to see in one of the demonstrations we'll go over, there's an interface booth that's set up, and our guy Dan will kind of talk through that point for you.
The other question, capacity for the industry is quite tight, and physical size comes up in a lot of my meetings, the size of the handler. The size of chip packages is increasing dramatically, and I'm talking to some of your peers who have to change the physical designs. Can you speak to that? Because that kind of goes to what you're talking about, a 10-year life cycle on the equipment. How much is modular that can just be upgraded with new robot arms or something versus wholesale changes to the footprint of the equipment that would let OSAT actually fit more handlers into a tighter space?
Right. Troy, you want to take that?
Yeah, it is a complex question. I think there are definitely trends right now that drive system sizes larger, right? Take mega JEDEC trays that are coming down the line. Those are driving customers to look at how we are going to do things differently. In terms of bricks and reuse, in general, there is a lot of, for us, it is a lot of reuse of technology in building bricks. Now, does that mean we are going to come with a small footprint handler? Part of the handler footprint decision is based off of what the customers require for production, things like number of bins that they need to sort to, lot sizes, package sizes. There is a lot of different things. So, it is not an easy answer to just give you, "Here is what we are doing." We are supporting what our customers are asking us for.
Sorry, Craig. How about we go in the back here?
Hi, I am Jay Dara with Indigo Capital, and I just had a question about this cycle compared to the 1997, 1998, 1999 buildup phase. How does this cycle compare to that one?
Yeah, I wasn't around. 1997, 1998. I can say from a Cohu standpoint, again, I wasn't at Cohu at the time, but I do know that we were developing active thermal control back then, and the main customer at that time was Intel. We were working very closely with them on their processors, and as their processors became higher speed, this thermal control was essential to their test as well. This is unique. Luis and I have been through multiple cycles. We both joined Cohu in 2005, and we've seen a few cycles, and generally, they're driven by our customers overproducing, having too much inventory, taking a pause on production, pulling systems out of production, and that kind of gets back to why this recurring revenue is so important for us, is that those units that are still in production, they consume parts and so forth.
It's generally driven by that overproduction, and then they finally realize it and it's, "Oops, now we got to take a break." As Luis said, we just have not seen this level of opportunity and what seems to be an insatiable demand from the market for use of AI. So, we'll see. As soon as I say that, of course, it's the same as the old cycles, but this has new characteristics, that unique characteristics that we haven't seen before. I don't know, Luis, I don't know if you'd say it any differently.
Yeah, it's true. It's a lot more technology-driven. For us, what we can see is a lot more technology-driven, the changes in the package construction. You guys were asking about larger package to Troy a minute ago. The changes in moving from one sensor in the die to a couple hundred sensors, maintaining temperature uniformity, power levels, power density. There's a lot more technology innovation happening for things that we touch nowadays that it's causing us to continuously look at refreshing what we just shipped a year ago with what comes next. Troy talked about the power levels going up to 12 kW by the end of next year. Just one example. He touched on mega JEDEC tray that we didn't even talk much about today. That's another example. There are others.
But doesn't this test cycle lack the efficiency that other, that two decades ago had?
It does. In fact, it's going anti-efficient. Troy mentioned test times, 30 minutes, and some new generation devices, 90 minutes.
Craig, you want to go again?
Yeah, round two. Thanks for taking it. I wanted to cycle back to Chris. Chris, we started talking about this on the break, but can you go through PAICe and talk about how the three-stack or the three-module stack evolves over time, either to add additional capability or to tune what you have for potentially higher value? Then on the CPO side, to what extent are we in production today with CPO capability? If we're not, when do you see that really starting to move up the curve? Then I've got one for Luis.
Okay, so on the PAICe software to start with, yeah, as you said, we have three products today, and that's been an evolution over time, and that's been based on the engagement we have with customers. When we develop customers, we have what's called customer success managers, which I think other companies call forward deployed engineers. We put them in there, they start understanding what problems the customer are having, and then how we can apply this technology to help that customer in our space, right? Primarily in the back end with the test cell. So, as an example of that I gave, that I actually talked a little bit about in the presentation, and you can actually see it when Eric does the demo in his room, is PAICe Prescriptive.
That started off primarily with the large install base we have with MATRiX handlers, because that's one of the largest install bases we have to focus on reducing unplanned downtime. As we moved over to Eclipse, one of the major use cases of that was really to resolve these thermal issues, which also involves downtime, but it's a much more complex problem, and we discovered that by working with the customer. So, that's an example of how we're trying to enhance the product itself. As the customer deploys it, they'll keep signing up for the annual subscription because we keep enhancing the product. That also enabled us, those engagements then enabled us, for example, we're developing this product now called Test Operations, to understand other problems they were having.
I think there's a rich future of this on the test floor, because they really don't have any tools today to do this kind of thing, except some very kind of junior varsity internal tools that they're using. So, I think there's a real rich opportunity to keep doing that with our forward deployed engineers or the customer success managers to keep developing the additional functionality. But again, you'll get to see a good example of that with Eric's demo on what we're doing.
Got it. Then the follow-up for Luis. Luis, the HPC funnel that the team has been using for the last three to four quarters is a really helpful transparency and accountability tool. The question is, as we look out to the midterm model and our potential to double sales, what's the assumption in the funnel regarding the number of customers that we're engaged with from now to attainment? Are we getting there with the current customer base we have, or do you have expectations that we're adding customers? To the extent we are, where's the team engaged in terms of how they're reaching out, et cetera? Thank you.
Yeah. Craig, we have plenty in the funnel to achieve the midterm model. I think we have more than enough in the funnel. As Jeff said, we're not even counting on successfully converting all that revenue into Cohu revenue, all that opportunity to Cohu revenue over the midterm. So at the moment, I think it's really all hands on deck making, sorry, all hands on deck making what we have a reality. The latest one that we announced this morning, as Jeff pointed out, kind of surprised us a little bit because it went from in engagement to qualified and POs in hand.
I think it shows a little bit of the drama certain customers are faced with having a solution they can run into production next year. So yes, will there be new customer additions in the funnel? Indeed. We talk about a few, but we want to make sure that we don't just flood the funnel. Instead, execute the funnel we have at the moment. That's the primary focus right now. Hope that answers the question for you.
Hey, let's take one more, and then we'll go to the demonstrations.
Yeah. Hey, guys. Kevin Garrigan from Jefferies. Just back on the cycle question, Luis, how do you balance increasing capacity and fulfilling demand versus the potential of holding the bag if the cycle takes a turn? What are you looking at to make sure that you're ready in case that happens?
First of all, we're really a CapEx-light organization. We don't have heavy manufacturing infrastructure. We rely a lot on suppliers. If you look at our equipment, if you ever visit our factory in Malaysia, for example, we build most of our equipment. Not all, but most of our equipment. You will see that it's largely final integration, configuration, and test. I don't think we'll carry a lot of excess capacity during a cycle downturn. If you look at our tester business, we work with well-known, we work with Jabil.
It's an outsourced manufacturer as well. Similar story. I think we do a pretty good job at maintaining a light infrastructure, a light manufacturing infrastructure that doesn't carry the bag, as you said, during down cycles. Now, is it pain-free? No, not entirely. There is, Jeff mentioned, there's value in that utilization of the infrastructure. But we have a pretty well-established playbook on how we run through cycles in the industry. Yeah.
And then just, Jeff, you talked about potential M&A. Basically, the main themes that I have heard today are customers want more of a complete system. How do you rank either, or what is your kind of focus between trying to buy or acquire a company that is more of a complement to what you guys are doing versus maybe going upstream or to another step of the manufacturing flow?
Yeah. That is a good question, Kevin. Matt, if you would like to take that one?
Yeah, sure. I think with the growth rate we are seeing in HPC, I think I like the word you used, complement, right? I do not think we are going farther upstream at this point. We are going to stick to our expertise in the areas. We see a lot of growth there. We see opportunities to possibly accelerate the growth in there through acquisitions. But the growth opportunities that we presented today are ripe, prime for possible M&A. We keep a pretty robust funnel, always are looking at stuff and always looking at ways to grow inorganically with these high-growth areas.
Okay. I think we're finished with questions right now, so we can transition to
Yeah, great. We're going to do the product showcases now. There are three different areas, and you'll be with a group, and the group is determined by the color on the back of your badges. You'll have a guide with that color flag that will take you to each session. Each session is about 15 minutes. The demo show and tell will last about 10 minutes. Then we have SMEs there that you'll have about five minutes to ask questions in those specific areas.
You'll travel with those groups to all three areas. The blue will start in here, then you'll see other guides with flags. Steph has a green. I believe Chris will have a red one. I'll guide the blue. Once that's finished, there will be an opportunity to grab, I think we have box lunches right where the breakfast was set up. But we will be exiting at 12:30 P.M. Not a lot of time after the showcase, but feel free to grab a box lunch. Again, appreciate everybody.
Thanks, everybody.