BizLink Holding Inc. (3665) Q4 FY2025 Earnings Call Transcript & Summary

Good afternoon, everyone. Welcome to BizLink's 4Q 2025 Earnings Call. My name is Derrick Yang, I'm the coverage analyst of BizLink here at Morgan Stanley. It's our honor today to have BizLink's top management with us today to discuss the 4Q '25 financial numbers as well as the outlook and industry dynamics. So without further ado, let me pass it to BizLink's IR Manager, Mike Wang, for a brief rundown of this call.

Thank you, Derrick. And also thank you to Morgan Stanley for hosting today's earnings call. Good afternoon, everyone, and welcome to our fourth quarter '25 earnings conference call. My name is Mike Wang, Senior Investor Relations Manager. Joining me today are Roger Liang, our Chairman; Felix Teng, our CEO; Florian Hettich, our COO; and Charles Tsai, our CFO. Our earnings results were released earlier today and are available on our Investor Relations website, where you can download the latest earnings materials and access them through MOPS. Today's call will begin with Charles, who will review our financial highlights, followed by Florian for operational highlights. Felix will then conclude with corporate updates before we move to the Q&A session. You may submit your questions at any time through the public or private chat function, and we will address as many as time permits. Before we begin, please note that today's discussion may contain qualitative forward-looking statements based on our current expectations, and they are subject to, of course, risks and uncertainties. Actual results may differ materially. Please refer to the safe harbor notice in our earnings materials for further details. This call is being recorded and will be available on our Investor Relations website within 24 hours. With that, I would like to turn the call over to Charles.

Thank you, Mike. Before reviewing the quarterly numbers, I would like to briefly step back and frame how we think about BizLink financially as BizLink continues to scale and evolve. Over the past several years, we have been transforming BizLink from a collection of strong operating businesses into an integrated platform with formal capital governance designed to allocate capital prudently, manage risk and convert growth into durable financial outcomes. As our mix shift toward more complex system-level engagement, particularly across high-performance computing, infrastructure and capital equipment; we believe it is increasingly important to anchor the financial discussion in earning quality, cash conversion and capital discipline rather than short-term volume dynamics. Our objective is not simply growth, but growth that is self-funded, resilient across cycle and aligned with long-term shareholder value creation. Our global teams executed strongly, achieving our eighth consecutive quarter-on-quarter increase in sales since bottoming 2 years ago. Quarter-on-quarter sales growth was primarily driven by HPC rising 16%. Capital equipment sales remained flat quarter-on-quarter and stay at the level we saw in second quarter 2025. Electric vehicle and factory automation sales continue to gradually recover after bottoming out near the start of the year -- the last year. Healthcare sales remained at third quarter 2025 highs. The quarter reflected the early ramp phase of several new high complexity programs, where production maturity, procurement optimization and yield learning curve have not yet fully normalized. As this program scale over the coming quarters, we expect meaningful cost absorption and margin improvement consistent with historical ramp dynamics. This resulted in our gross margin falling from the third quarter 2025 levels. OpEx was roughly stable quarter-on-quarter and stayed at year-to-date levels, showing that our cost efficiency efforts are flowing through into our P&L. During the second quarter of last year, we completed a review of certain co-invested production assets and recognized the nonoperating, noncash impairment to align carrying value with updated long-term cash flow assumptions. In the fourth quarter, we continue to apply the same conservative framework to reassess asset recoverability as market conditions evolve. The resulting adjustments were program-specific, noncash and accounting driven and do not reflect a deterioration in the underlying competitiveness or strategic relevance of the business. Our effective tax rate this quarter reflects a year-end adjustment that captured full year impact and therefore, should not be interpreted as indicative of our going tax rate. Operating leverage further boosted earnings with our EPS continuing to grow quarter-on-quarter. On a full year basis, our total sales was stronger than expected. Sales was driven by HPC, which more than doubled year-on-year and by capital equipment, which was up more than 50% year-on-year and was more than double 2023 levels. Industrial sales recorded its second annual increase. Finally, we closed our XFS Communications acquisition on January 5, 2026, and began consolidating its financial on the same day. So from a portfolio perspective, our Industrial segment continued to provide a stable earnings base, supported by high mix project-driven demand across automation and equipment applications. Over the recent quarters, customer procurement behavior has begun transitioning from broad inventory correction toward normalized ordering patterns with visibility increasingly tied to new program activity rather than restocking. Our consumer-linked segments are being managed conservatively. While automotive and electrical devices are not expected to be primary near-term growth drivers, they continue to play an important stabilizing role within the portfolio, where prior adjustments and disciplined planning help support operational resilience and limit volatility. In parallel, demand related to AI infrastructure remains structurally strong, while engagement level continue to increase. We are managing this business with appropriate discipline given its step function deployment profile. This diversified mix allow us to maintain financial stability while retaining exposure to long-term infrastructure investment cycles. So what is changing? Growth across the portfolio is increasingly driven by higher system content, deeper engineering engagement and earlier qualification cycle rather than pricing actions. As customer program increasingly follow platform-based deployment cycle, portion of our capability investment occur earlier in the development phase to establish engineering and manufacturing readiness. Importantly, these investments are reusable across multiple customer programs and generations, allowing revenue and content to scale once platform enter volume deployment without requiring proportional increases in incremental capital deployment. Across multiple segments, customers are also engaging earlier in platform design and qualification cycles, modestly extending planning visibility while increasing execution requirements. As a result, growth increasingly reflect platform scaling rather than incremental capacity expansion, supporting more efficient capital deployment over time. At the same time, earning quality continues to improve with cash conversion and return on equity trending upward as operating leverage improved and working capital intensity declined. Growth is being absorbed efficiently rather than funded through balance sheet expansion. At system scale, progress is rarely [indiscernible]. Platform transition and deployment schedule do not move in straight line, even when long-term demand is clear. Our focus is, therefore, on execution reliability rather than maximizing short-term throughput. This discipline is reflected in our capital allocation approach. We maintain tight working capital management, steady CapEx intensity and selective M&A focus on capability enhancement rather than growth for its own sake. During the first quarter of 2026, we completed the acquisition of XFS Communications, adding optical interconnect capability, consistent with our disciplined capability-enhancing M&A approach. Consistent with our framework, integration is being executed methodically within our existing financial discipline. So what does this mean going forward? NPI activity has become an increasingly important forward indicator with many programs aligned to customer platform refresh cycle extending well into 2026 and even into 2027. During the quarter, BizLink achieved several milestones reflecting increased institution maturity, including index inclusion and joining RE100. This step reinforced governance, transparency and balance sheet discipline as the company scale. Taken together, we believe BizLink is positioned to continue scaling with financial resilience as customer architecture evolve. Florian will now provide update on our latest quarterly operational takeaways.

Thank you, Charles. Yes, from an operational point of view, the customer engagement is shifting from component execution toward more system-level collaboration as architectures become more complex. Power delivery, also data connectivity and mechanical integration increasingly interact across customer platforms. Higher performance requirements and tighter deployment schedule raise the cost of execution error and increase the value of early engineering engagement. This increasingly includes both electrical and optical connectivity domains where integration and manufacturability requirements are converging as system architectures evolve. What we are seeing is not a collection of isolated programs, but rather a repeatable operating pattern across multiple customers and industries. So from an operational perspective, we frame our business not around individual products, but around how system function as a whole. As customer architectures evolve, our delivery, data connectivity, mechanical engineering and manufacturability increasingly interact with one another. Customers are no longer optimizing isolated components. Instead, they are designing platforms that must operate reliably under higher density, tighter tolerances and faster deployment time lines. In this environment, execution reliability and integration discipline become as important as technical performance. This shift changes how customers engage suppliers. Rather than sourcing individual parts late in development cycles, customers increasingly seek partners capable of contributing earlier in design and supporting global execution once platforms scale. Our operating model is built around supporting this transition, enabling us to participate where system-level constraints are emerging rather than where specifications are already fixed. As compute density rises, the challenge is no longer meeting individual specification in isolation. Validation becomes multilayered, failure tolerance tightens and execution consistency across regions become critical. In AI infrastructure, higher rack level integration is driving redesigns in both power delivery and data connectivity. Similar dynamics are emerging in semiconductor equipment, where subsystem complexity and engineering constraints are accelerating outsourcing toward integrated partners. As requirements increase, the universe of suppliers capable of supporting these platforms at scale naturally narrows. Our response has been to deepen system-level engagement through expanded NBI, capability and tighter integration between engineering, box build and system integration. Increasingly, customers engage us not only for components, but at the system level where power, data and mechanical constraints intersect. Once architectures are validated, continuity and execution reliability become critical. At the rack level, power delivery is becoming a primary architectural constraint as densities exceed 100 kilowatts. Increasingly, we are being engaged earlier in rack architecture discussions, particularly around power distribution geometry and serviceability considerations. Across capital equipment platforms, we see a similar pattern. OEMs who are facing engineering and clean room constraints are outsourcing more integrated subsystems, rewarding suppliers capable of multidisciplinary execution at scale. And what is that meaning going forward? As the integration increases, manufacturability and repeatability become as important as design performance. Our focus is, therefore, on building execution capability tied directly to customer programs rather than expanding capacity ahead of demand. We believe this disciplined approach positions BizLink to scale alongside increasing system complexity without increasing operational risk. Felix will now provide updates on our latest quarterly corporate takeaways.

All right. Thank you, Florian. Yes, BizLink was founded on solving customer problems through engineering engagement and execution disciplines rather than focusing on any single product or market. That operating philosophy continues to guide the company today. As systems scale, the constraints our customers face are changing. Power density, data movement and integration complexity have become binding factors across data centers, semiconductor manufacturing and emerging automation environments. We view this not as a short-term technology cycle, but as a long-term industrial transformation. Importantly, these constraints are not emerging only with the AI infrastructure. Similar shifts toward higher system integration and earlier engineering collaborations are appearing across semiconductor equipment, industrial automation, healthcare and advanced equipment platforms. This reinforces our view that the changes underway represent a broader industrial evolution. This has never been organized around a single product category or technology cycle. Instead, we frame our role around solving physical layer constraints that emerge as the systems scale. Across industries, series of technical advancement eventually shift bottlenecks away from compute or core intellectual property and toward integration, power delivery, reliability and manufacturability. These constraints determine how efficiently systems can be developed at scale. Our strategy is, therefore, not to predict a single technological end state, but to position BizLink at a layer where these constraints consistently appear. By focusing on execution capability, engineering engagement and global manufacturability, we aim to remain relevant across successive generations of platform transitions rather than any single technology way. So what is changing? Value creation is increasingly moving away from discrete components towards system-level integration and distribution reliability. Suppliers capable of engaging early, qualifying globally and executing consistently become more critical as architecture evolve. BizLink's role remains consistent, solving fiscal layer constraints as systems scale. How we are executing? NPI has become a strategic capability, enabling earlier participation in architectural decisions and positioning BizLink ahead of bottlenecks across HPC and capital equipment platforms. We view AI infrastructure development similarly to those evolution of a modern city. Early phases prioritize scale, while later phases optimize efficiency and coordination within the existing footprint. Our role is to ensure the physical foundations, power, data and connectivity, support whatever technologies follow. As integration increases, we are preparing for higher levels of optical adoption over time by expanding our interconnect capability across both electrical and optical domains. We view this evolution not as a binary transition, but as the convergence of electrical and optical interconnect technology, addressing different system constraints as architectures scale. We do not view this as a binary transition. Electrical and optical interconnects address different constraints and are expected to coexist for an extended period as architectures evolve. Our focus is, therefore, on building manufacturing and execution capability rather than betting on a single technological outcome. AI infrastructure represents a long-term industrial upgrade cycle rather than a single demand wave. The same interconnect disciplines underpin emerging applications such as robotics, autonomous systems and intelligent manufacturing. We see this area as a long-dated extension built on top of today's infrastructure foundation, reinforcing the importance of disciplined execution across current platform transitions. Now let me turn the call over to Mike.

Thank you, Felix, Florian and Charles. This concludes our prepared statements section. Now let us begin the Q&A section.

Please type in your questions and then we will answer as many of them as possible in the remaining time. I want to remind everyone that there will be no forward-looking quantitative comments. Looking at the questions, for the CMS -- sorry, Morgan Stanley, Derrick. That was Helen, [indiscernible], Billy, [indiscernible], Doris all have asked about our outlook, growth trajectory, quarterly pacing. It seems there's still a lot of questions about how sensitive our HPC business's potential moderation in AI CapEx spending. There's been continued concerns about an AI bubble. So for this, for more sort of mid- to longer-term outlook, especially in regards to this AI CapEx sustainability and industry transition, I'll hand it over to Felix.

All right. Yes, we view the current discussion around AI capital intensity as a natural progression of the infrastructure build cycle rather than a signal of weakening demand. You can think of this as a transition similar to building out a transportation network. Early investments focus on adding more vehicles, while later investment shift towards highways, traffic management and efficiency, so the system can operate reliably at scale. So earlier phases of AI investment focused primarily on accelerating compute development. As clusters scale, the industry's priority increasingly shift towards deployability, efficiency and long-term operation of installed capacity. This changes spending composition more than spending direction. Operationally, we are seeing greater emphasis on rack-level architecture, power delivery, thermal coordination and reliable high-speed connectivity, areas that become more critical as power density rises. Even if the rate of CapEx growth normalizes over time, infrastructure content per rack continues to increase because higher density systems require more sophisticated integration. So as a result, our engagement is less tied to short-term procurement cycles and more aligned with multi-generation platform architecture. From our perspective, the key driver is not simply total CapEx, but how AI infrastructure is being built and that trend remains constructive.

Thank you, Felix, for the colorful outlook as well as to the analysts that post the questions. Now for the next set of questions, talked about rack-level power, how that's becoming central. This is coming from actually kind of the same group, but what I've seen, [indiscernible], Derrick [indiscernible] billing. So why is rack-level power becoming more important? And how does this expand business opportunity? Now for this one, it's more operational. I'd like to hand it over to Florian.

All right. Thanks. Yes, I mean power is indeed one of the major topics for future development. And as AI workloads scale beyond the, let's say, traditional server environments, the rack power density is really significantly increasing, in many cases, now exceeding several hundred kilowatts per rack. And at those levels, power delivery is really becoming a system-level reliability constraint rather than, as I said earlier on, only a component decision. So at lower densities, the power design is similar if you compare it to wiring individual homes within a city. But if you go at AI scale rack densities, it becomes closer to, let's say, managing a power distribution for an entire city block or part of the city where stability and also coordination matter more than any single device. So inside individual servers, power paths are relatively localized and also maybe standardized. But at a rack level, however, operators must manage significantly higher current, also higher thermal load. Redundancy design is important and also serviceability across the entire cluster segment is an important topic because a fault at this level can affect a large portion of the deployed compute simultaneously, and this is elevating validation and reliability requirements tremendously. So the impact of an effect on this deployed compute is really serious. So hyperscalers, therefore, increasingly define rack power architecture mainly independently from individual PSU vendors in order to maintain flexibility and also avoid single vendor dependence. And this shifts engineering responsibility toward integrated rack solutions where, as I said, the electrical and thermal and operational considerations must be validated together. And as architectures get mature, qualified, designs typically remain fixed for platform life cycle. So this dynamic increases the importance of early engineering engagement. That is the key, I think, the early engineering engagement and also the execution afterwards, the consistency, execution and structurally raising the content in infrastructure when we scale AI systems. And also as power density increases, the infrastructure reliability increasingly determines how much of your compute power is usable and is bringing into performance. And this shifts actually the value creation towards integrated rack level solutions where engineering execution and validation become, as I said, more important than the individual component specifications on the component level.

Thank you, Florian, for providing the feedback to those questions from those analysts. For the next set of questions, thank you to Morgan Stanley, Derrick, [indiscernible] Doris and Daiwa's Helen for this topic on our HPC strategy positioning. How should investors, of course, analysts understand our HPC strategy within AI infrastructure. So this is a very big picture sort of view. So for this one, I'd like to hand it over to our Chairman, Roger.

Well, our HPC strategy is built around enabling the continuous scaling of AI infrastructure. As AI deployment evolve from individual servers to increasingly large [indiscernible] cluster, the industrial primary challenge are shifting away from compute performance alone and toward the physical infrastructure required to support system level expansion. In practical terms, the limiting factors increasingly become how power is delivered into dense rack and how data move reliably and efficiently across the growing cluster. So accordingly, our focus is on 2 infrastructure layers that directly determine how fast AI system can scale, rack-level power delivery and high-speed data connectivity. In many ways, AI infrastructure is evolving from stand-alone computers towards something closer to an electrical grid, where power delivery and data connectivity determine how much compute can actually be utilized. On the power side, AI workloads are driving a significant increase in rack power density as racks move towards much higher power level. Traditional data center power approach required redesign to improve efficiency, thermal performance and operational reliability. This transition [indiscernible] the importance of high current power connectivity, including power [indiscernible] power system and integrated rack distribution solution. Our long-standing experience in high-power interconnect and system integration position us naturally within this architectural shift where power delivery is becoming a defining element of next-generation AI infrastructure rather than a standardized supporting component. On the data connectivity side, the majority of current industry investment is focused on scale-out architecture where thousands of escalators must communicate across racks within large cluster. This deployment requires solution that balance bandwidth, power consumption, signal integrity and the serviceability and the high speed at hyperscale. Our portfolio of direct attach and active electrical cable solutions support this requirement by expanding reach and maintaining performance efficiency across real-world database environment. As clusters continue to expand, reliable short to mid-rach connectivity remains essential to enabling critical deployment at scale. We also participate in scale-up connectivity environment, including PCI-related solutions, which broaden our exposure across different system architecture and provide diversification as compute platform continue to evolve. While scale-out deployment currently represent the largest area of infrastructure expansion, participation across multiple connectivity layers allows us to support customers throughout the system architecture. From a financial perspective, this positioning also support greater durability in our operating model. Engagement increasingly occurs earlier in customer platform design cycle and extend across multi-generation architecture rather than single product developed deployment. As system complexity increased, qualification requirement, validation cycle and execution continuity become more important, which tend to lengthen the program life cycle and improve visibility. As a result, growth is increasingly driven by expanding system content and deeper engineering participation rather than short-term volume fluctuation, supporting more stable cash conversion and disciplined capital deployment over time. As interconnect requirement advance, the industry is exploring multiple technology approaches to address performance and efficiency needs across different business and use case. Our strategy is not centered on any single transmission meeting, but rather on supporting customers' evolving connectivity architectures. Today, our disclosed activity are primarily focused on electrical interconnect solutions, where we have deep engineering and manufacturing capability. Recent development has expanded our capability into adjacent areas of next-generation interconnect technology, including optical connectivity solution that complement our established electrical expertise. This capability allows us to participate more broadly in evolving system architecture as customer balance performance, efficiency and manufacturability across both electrical and optical domain. Importantly, we view electrical and optical technology as complementary rather than mutually exclusive. AI infrastructure increasingly apply different solutions to different portion of the system depending on distance, power efficiency and operational consideration. By focusing on the underlying infrastructure challenge, delivering higher power density and enabling reliable high-speed data connectivity; we believe our position remain relevant across technology transition and the future architecture change. In summary, our HPC strategy is centered on supporting the physical infrastructure layers that enable AI cluster to scale. By addressing both rack-level power and high-speed data connectivity, we participate in areas that become more critical as AI deployment grow larger, denser and more complex while maintaining flexibility to evolve alongside future technology development.

Thank you, Roger. That's our company's bigger picture view of HPC. And of course, thanks to the analysts that post those questions. Now this is going to be a little bit more financial in nature. Again, thanks to [indiscernible] and Derrick, Billy and Doris. This is about our earnings growth mechanics and generally what drives earnings expansion as the infrastructure scales. So for this one, I'd like to hand it over to our CFO, Charles.

Thank you. And yes, that's a good question. And we really look closely. Our earnings expansion is increasingly driven by structural content growth and system complexity rather than cyclical shipment recovery. As AI infrastructure scales, high power density, tighter performance requirement and deeper system integration are increasingly increasing the amount of data interconnect, power delivery and engineering content required per deployed. Earlier, computing cycle were largely unit driven, where growth dependent on a number of servers deployed. In contrast, AI architecture requires significantly higher power density, tighter signal integrity requirement and more integrated subsystem design. This increases the amount of data interconnect, power delivery and engineering content required per rack. You can think of this transition as moving from simply adding more identical machine toward upgrading the electrical and structural infrastructure inside a building. Even if the number of buildings grow steadily, each one requires substantially more internal infrastructure to operate at higher performance levels. So as a result, our growth is influenced by 3 structural factors. First, content per deployment increases as power and data connectivity requirements scale. Second, product mix shift toward higher-value system-level solutions, where reliability and engineering capability become more important than commoditized components. Third, earlier engagement in platform design allow us to participate across multiple generation of a program, improving operating leverage over time. Therefore, while deployment volume remains important, we believe our earnings progression increasingly reflects the growing technical intensity of AI infrastructure rather than pure cyclical shipment growth. And I hope that answers the question.

Thank you, Charles. So that was our first set of financial-related questions. Of course, thank you to the analysts that shared those questions. Well, the next one is also for you. This is from KGI, [indiscernible], Morgan Stanley, Derrick, of course, Masterlink's analysts. This question is about operating leverage from AI infrastructure scaling. I think the summary is going to be something like how much further can operating leverage improve as AI revenue grows. So this is going to be another financial-related question. So Charles, please, if you may.

Yes. Thank you. No. As AI infrastructure evolve, growth dynamics are shifting away from traditional volume-driven expansion toward architecture-driven scale. Increasing compute density requires substantially higher power delivery, thermal management coordination and system integration complexity at the rack level. This evolution increased the amount of data interconnect and power content required per system deployment. Importantly, much of the incremental content leverage existing engineering platform, manufacturing processes and qualification framework that already developed for hyperscale customers. In practical terms, once the platform architecture is engineered and validated, incremental deployment resemble adding additional floor to an already designed structure rather than redesigning the foundation each time. As a result, revenue growth increasingly benefit from rising content per deployment rather than proportional increases in operating expenses. Engineering investment made in earlier platform generation that support multiple successive architecture, allowing incremental revenue to scale more efficiently. This creates structural operating leverage, where operating expenses grow at a slower rate than the revenue as AI infrastructure adoption expands. From a financial perspective, this leverage reflects platform maturation and architectural progression rather than short-term pricing or cyclical utilization changes. over time, we expect improvement in operating efficiency to be driven primarily by deeper system participation and increasing platform reuse across customer deployment. So yes, I hope that also answered the question.

Thanks again, Charles, for taking the second one right after the first one and also to the analysts for those questions. Now let me see the next set of questions from both [indiscernible] Doris [indiscernible]. And again, Morgan Stanley's Derrick. This is going to be on HPC value capture within the AI. In fact, we have received lots of questions around AI. So this is going to be, as AI system scale, I guess, what the analyst is trying to ask about is, where is the value increase being created within the infrastructure stack. This is going to be another bigger picture question. So I'd like to hand over the floor to Roger.

All right. As AI systems scale, value creation within the infrastructure stack, I mean, increasingly shift toward component that enable reliable operation of high-density compute environment. While advancing semiconductor drive computational capability, critical deployment of these systems depend on power delivery, signal integrity and system integration layer that ensures stable performance at scale. Higher compute density increased electrical thermal sensitivity and data connectivity complexity, which elaborate the importance of infrastructure solution that manage power distribution and data interconnection reliability across entire rack architecture. In many ways, this is similar to large transportation network where long-term value ultimately concentrates in the infrastructure that keep the system operating reliably such as the highway traffic coordination and energy distribution rather than any individual vehicle moving through the network. BizLink participates in this portion of the value chain through solutions that support both power and data connectivity at the system level with architecture developed for higher density and greater integration and related importance of this enabling layers increase alongside the compute performance. So rather than depending solely on unit shipment growth, value expansion increasingly come from high content intensity per system and deeper integration into platform architecture. This positions the company to benefit from continued scaling of AI infrastructure, even a system design across multiple technology generation. In this context, [indiscernible] increasingly align with enabling the infrastructure scalability rather than just us find a discrete component.

Thank you, Roger. Another big picture question for the analysts and for the audience today. For the next one, lots of questions around copper and optics lately. And then, of course, thank you to [indiscernible]; Nomura Kenny; JPMorgan, Bill; [indiscernible]. Putting a lot of these questions together talking about copper and optics architecture expansion. How should everyone think about copper versus optical interconnect over the longer term? So this -- we did touch on this in the remarks during Florian's section. So I'd like to hand it over to Florian.

Yes. Thanks. I mean, this is an ongoing discussion since several years, and this is also in the community very controversially and discussed also in the last couple of weeks and months. But we do not actually view that as a binary, let's call it, binary transition. It is less, in our view, like replacing one technology with another. It's, like, you can imagine like a transportation system. I mean, you have streets, you have railways, you have airplanes and you need all of them because you have different distances which you need to bridge. And so each of those transportation systems is optimized for different distances, efficiency requirements and so on. And the same is as AI infrastructure scales. Optical adoption is increasingly driven by system architecture requirements rather than isolated component substitution, which expands overall interconnect complexity rather than reducing the electrical connectivity. And electrical and optical interconnects address different engineering constraints within this AI system architecture, particularly when it comes to distance and power efficiency, but also latency and serviceability. Copper-based active electrical solutions remain highly efficient, especially for shorter reach, high-density environments where reliability and integration and also power efficiency are the more critical factors. On the other side, optical technologies become increasingly important as bandwidth requirements and also transmission distances expand and go up. So what we are observing is not so much a substitution, but architectural expansion. This architectural expansion also introduces additional connectivity and distribution layers within the rack environment to support what I said, the serviceability, modular replacement or operational reliability as system scale increases. And as systems scale, overall, the complexity of interconnect systems increases because the designers optimize different parts of the system for different performance requirements. And having said that, in addition, as integration moves closer to high-power ASICs, manufacturability and also thermal coordination and yield management become even more critical challenges. So our focus is, therefore, on building manufacturing and engineering readiness that allows us to support evolving architectures rather than relying on single technology outcome. That's also what we heard earlier on. So I hope that answers the questions. And back to you, Mike.

All right. Thank you for that, and the whole debate on copper and optics is something that we believe there is going to be a binary transition. For the next set of questions, again, we did touch on this high-voltage DC, quite a few questions again, thanks to Derrick, Doris, Kenny and Billy. And I think I see some questions also from the audience there. So this will capture some of that as well. Why is the industry moving toward HVDC? And what does that mean for copper to connect? So this is going to be another operational topic. So Florian, if you don't mind?

Yes. Thanks. One is the optics versus copper discussion and the other one is the HVDC discussion. So yes, the transition towards higher voltage power architecture is primarily driven by physics, not so much by the technology preference. It's just -- well, it's physics. We are reaching now power densities that traditional low-voltage data centers were not designed for and were also never intended to support. And as now the GPU clusters scale, rack power requirements are moving from former tens of kilowatts historically now towards hundreds of kilowatts per rack, and this is really a tremendous jump. And at these levels, electrical efficiency and also, as I already said, the thermal management become the primary constraint on scaling rather than the compute performance or the networking bandwidth. So delivering the power at low voltage is something similar as trying to move large volumes of water through many small pipes. Flow becomes inefficient and heat losses increases and so on. So higher voltage allows the same energy to move more efficiently through smaller currents, improving scalability. So physics says that power losses increase with the square of current, so meaning that delivering large amounts of power at low voltage requires extremely high current. High current introduces excessive heat generation, larger conductors, reduce efficiency and also increased reliability risks. And when we now look at the scale of the AI systems, the industry is, therefore, moving toward or needs to move towards higher voltage power distribution architectures. And in practice, some cloud service providers are adopting a plus/minus 400-volt HVDC, high-voltage direct current designs, while others are moving towards 800-volt HVDC. So that's reflecting just different platform architectures. But both approaches significantly reduce the current levels. So they are then getting to the goal that they improve efficiency. They lower the thermal stress and also enable a safer and more scalable power delivery across the whole dense AI deployment. In this sense, a higher voltage HVDC distribution is not simply an upgrade. It is becoming a physical requirement more for continued scaling of AI infrastructure. A common market assumption, by the way, is that higher voltage automatically accelerates optical adoption and reduces copper usage. In practice, these are solving different engineering problems and HVDC addresses power delivery efficiency, while interconnect technology choices are determined primarily by distance, latency, as I said, and power consumption and serviceability requirements. In fact, improved power architecture can enhance even operating conditions for electrical interconnects. Higher voltage distribution improves, for example, grounding stability, reduces electrical noise and also lowers thermal stress within the rack environment. So these are all positive effects to improve the signal integrity, allowing short-reach electrical connections to remain efficient and reliable even when data rate increases. So AI system design is increasingly concentrating on compute into tightly integrated rack environments where connection distances are extremely short. Within these ultra short reaches, electrical interconnects will still remain highly attractive because they provide, as I said already, lower latency, lower power consumption, simpler serviceability compared with optical solutions. So as a result, HVDC tends to push optical solutions outward towards longer distance connectivity while preserving and in some cases, even reinforcing the copper's role inside racks and near rack domains. So structurally, this reflects a broader industry transition from in the past, more compute limited environment towards now an energy limited environment. Constraint is energy. And as power delivery improves, system density increases, which drives more localized connectivity and greater overall interconnect complexity. The likely outcome is not replacement, but a hybrid architecture. That's how we see it in which both electrical and optical technologies coexist, each optimized, as I said, for different distances and efficiency requirements. And in that environment, electrical interconnects continue to play a critical role in enabling scalable and also energy-efficient AI deployment. I hope this answers the questions. Back to you, Mike.

Thank you, Florian. We still have a lot of questions, but I think we'll try to address something on AEC. [indiscernible] from the audience, our largest AEC customer guided to 50% year-on-year growth for its fiscal 2027. Do we have enough capacity to meet the majority of this increase in demand? So I'd like to hand this over to Felix. And again, this is the last question that we will have time for today.

Yes. So actually, we are actually seeing quite a lot of [indiscernible] and also high demand, increasing demand from our customers. So we are actually -- for the past year or 2, we are actually preparing to meet our customers' requirements. So this is not a surprise to us. And actually, there are a couple of locations, we are increasing our capacity, already in progress. And so yes, we do have enough capacity to meet the increasing demand. And it's not just in AEC, but in multiple sectors.

Thank you, Felix, for that last question also to the investor that posted that question. So again, thank you, Roger, Felix, Florian and Charles. This concludes our Q&A section. A replay of the conference call today will be available on our IR website within 24 hours from now. If you have any further questions, please feel free to reach out to BizLink's Investor Relations team that includes me as well as Jimmy. We thank you very much for joining today's call. You may now disconnect.