Good afternoon, everyone. Welcome to BizLink's fourth quarter 2024 earnings English conference call. This is Mike Wang, IR Manager. I am joined by Roger Liang, our Chairman; Felix Teng, our CEO; Florian Hettich, our COO; and Charles Tsai, our CFO. Our results were just released and are available on our IR website, where you can download the latest earnings release materials as well as access them from MOPS. This one-hour call will begin with Charles to highlight our financials before we switch to Florian to highlight our operations, and then end with Felix to highlight corporate-wide items. We will then move to Q&A before concluding this call. You may type in your questions now, and we will answer as many of them as possible. We will not provide any quantitative forward-looking comments.
Before we continue, please kindly be reminded that today's discussions may contain qualitative forward-looking statements based on our current expectations, which are subject to significant risks and uncertainties and may cause actual results to differ materially from those contained in these qualitative forward-looking statements. We are not obligated to update these statements, which are to be used for information purposes only. Please refer to the Safe Harbor Notice in our earnings deck for more details. I would like to remind everyone that today's call is being recorded. This recording and these prepared remarks will be uploaded onto our IR website within 24 hours after the conclusion of this call. We sincerely appreciate Yuanta Securities for hosting today's call. With that, I'll return the call over to our CFO, Charles .
Thank you, Mike. Let me begin by talking a little bit about our product sales. AI DataCom sales rose 35% year-on-year last year, as high-performance computing sales rose 48% year-on-year, while peripheral sales continued to fall, dropping 18% year-on-year. Electrical Appliance sales rose 18% year-on-year, given new product launches, some of which are continuing to be rolled out this year. Industrial sales rose 4% year-on-year, as Capital Equipment sales rose 44% year-on-year. Healthcare sales rose 8% year-on-year. Factory Automation sales fell 31% year-on-year, and Energy sales rose 63% year-on-year. We're delivering on our various project wins with more incoming this year and next year.
Looking at HPC first, our Data Solution sales grew 20% year-on-year last year, as shipment of our high-speed external data cable product began to visibly grow in the second half of the year, while our Power Solution sales nearly doubled, as shipment of our high-power cable and connector product grew throughout the year. Data Centers are very cost-hungry. Power is evolving at a much faster rate than data. However, first in today's HPC construction, Data Solution accounted for 52% of total HPC sales versus a peak of 74% two years ago. Our Power business, which has stretched into other segments, has grown and has greatly outgrown our Data business. Looking at Capital Equipment Business, we continue to benefit from the outsourcing trend of semiconductor production equipment to make our electrical distribution subsystems, including control boxes, power distribution units, and system enclosures.
Our box build and system integration activity surged last year, given multiple IP intensive technology transitions at our SPE customer, as we highlighted last quarter. The Easys and cable connection deal that closed late last year marked new effort in this. And while our activity is currently focused within capital equipment, we're seeing similar projects emerging in healthcare, electric vehicles, and even in general industrial areas. We continue our growth momentum, achieving sequential sales growth for the fourth consecutive period. While gross margins soared slightly quarter on quarter due to typical year-end seasonality, our product mix stayed favorable, and cost control remained firm. OpEx averaged $2.0 billion for the 12th straight quarter, with OpEx to sales dropping below 15%, reflecting improved scale and potential for further operating margin expansion. Non-operating income was positive, driven by foreign exchange gains.
Our effective tax rate declined due to increased tax shield from financing optimizations, a greater profit share in low-tax jurisdiction, and a successful turnaround of previously loss-making entities. This structural enhancement will continue to drive long-term tax efficiency. Earnings per share (EPS) reached $8.28, marking a new record despite the ongoing conversion of outstanding ECBs, which are now nearly fully converted. Our strategic initiative remains on track to support both sustainable profitability and operational efficiency. On balance sheet and cash flow, BizLink's focus has always been on the long term and is less on specific near-term numbers. We seek to support growth alongside our established brand name customers, many of whom are leaders in their respective industries, and to realize the vision of high-tech startups and industry newcomers. We're leveraging key megatrends such as HPC and capital equipment, which accounted for nearly 29% of total sales last year.
Financially, our goal is to ensure sufficient talent and capital resources to enable this growth aspirations. This includes building and acquiring this capability and capacity needed to serve customers effectively while aiming for optimal cash flow efficacy. We continue to generate cash and have existing unused data facilities. Our 2025 cash needs are well accounted for, but we will be on the lookout for strategic opportunities to accelerate our development. As we expand our offering and become an instrumental part of our customers' roadmap, we aim to grow our balance sheet sensibly, maintaining the right capital structure to support sustainable and profitable growth. Future-facing customers seek financially sound partners like BizLink. We will continue to be a successful incubator of newcomers and new applications, utilizing our growing presence in key parts of the world where everything is happening.
Our Silicon Valley Fremont tech quarters continue to be a strategic location in our global footprint. Speed and agility are key, and we're showing to customers what it means to be closer to them to service them faster and better to gain long-term market share. Some opportunities take time, and we are glad that we're starting to be recognized for it. In the face of macroeconomic and geopolitical uncertainty, especially with the challenges and opportunities of Trump 2.0 for the next four years, sustaining financial flexibility and resilience is vital. We're better positioned than we used to be via our substantial diversification efforts, including strengthening our global footprint and customer mix to mitigate the impact of shifting trade dynamics. We have enhanced operational efficiency to streamline working capital and upgrade in cash flow. This enables us to deleverage and reduce interest expenses.
While we may need to fund future growth from external sources, given the promising growth opportunity in various markets, we will do so responsibly. We aim for our capital structure to stay consistent with historical operations. We maintain a cautious stance amid prolonged volatility, with no clear mid- to long-term direction, as various macroeconomic and geopolitical factors influence the landscape. We're also continuing to move toward having smaller open currency positions. We seek to use as much local financing as possible to rationalize and to add resiliency to our financial structure. This balanced strategy allows us to mitigate risks effectively while supporting our global operations. On a bigger picture, while we expect HPC and capital equipment remain strong throughout 2025 and financially in 2026 as well, we're starting to see green shoots appear in factory automation, suggesting the trough could be here.
Although it remains challenging, we're refocusing our resources toward autonomous driving so that we're prepared for this growth wave. Healthcare will likely continue to grow as world population further ages. Green shoots are also emerging from our exposure in Europe and for INBG. INBG's strengths in 2023 allow CTBG to continue to spend to prepare for the next up cycle, which we saw last year and will continue to see this year. INBG was very good about leading them quickly to adjust to its down cycle. They continue to make progress in various product and market development efforts. INBG looks to be bottoming out now, and manufacturing PMI in Europe are promising as they point to a slow ongoing recovery. INBG's TMP business unit, which is where our customers go to solve their difficult design challenges, is quickly turning out to be a hidden gem.
It has been selectively gaining margin share, and we're growing its global presence. Its capability has slowly expanded beyond capital equipment, healthcare, rolling stock, and industrial specialties, and gradually making its way into more product areas. It has since become INBG's largest business unit, and it's instrumental in some of our megatrends. We just announced our latest tuck-in deal. We will be acquiring a 100% stake in Alpha Elektrotechnik AG for an enterprise value of $34.3 million. Alpha offers high-voltage cable solutions for high-speed trains, intercity trains, local trains, and locomotives. Alpha recorded sales of $11.1 million in its latest fiscal year and is a rolling stock unit spun off from the Pfiffner Group in Switzerland. This acquisition is expected to close in the second quarter of 2025 and will be financed through an all-cash transaction.
The disconnector business under Alpha is now part of the deal. We will leverage our collaboration with Alpha to realize synergy in technology and new application development, as well as in market access development. This deal is expected to be attractive from day one. The attractiveness of the rolling stock interconnect market is underscored by its robust growth prospects, technological innovation, and strategic consolidation with only a few known and reliable solution providers, including BizLink. We warmly welcome the Alpha team as well as BizLinkers and look forward to their contributions. Our CapEx plan for this year marks a new high as we prepare for growth. We continue to rationalize our footprint in China, both in Zhengzhou and Xiamen, and selectively expanded elsewhere, including in Southeast Asia.
We look to spend on equipment, especially for HPC and capital equipment, and we'll be updating our IT system with a particular focus on digital transformation efforts. Florian will now provide updates on our latest quarterly operational takeaways.
Thank you, Charles. So I will elaborate first on the industrial segment. So semiconductors are the core of HPC, humanoid robots, and autonomous driving. As we can see from this chart, this is how we view our long-term AI opportunity. It is an ecosystem that is interconnected and will gradually develop and evolve, supported by semiconductors as the overall foundation over the next few years. We are upgrading our existing capabilities, and we are actively searching externally to fill in any gaps. We have multiple business units collaborating on these megatrends, and we are well-positioned to benefit from their growth.
The next one to two years, we will only see increasing complexity in terms of landscape, technology, and solutions and business models, requiring not just speed and agility, a global footprint and resilience, but customer relationships and foresight. Our efforts in capital equipment were greatly boosted by our Speedy deal back in the early 2020, resulting in sales more than tripling from 2020 to 2022 before encountering a brief down cycle in 2023. Our expertise in rapid prototyping, new product introduction, high-mix, low-volume manufacturing, and engineering and supply chain speed and agility reduces risks and accelerates time to market for our SPE customers who are in their race on their own. Our acquisition of Easys marks a step up in our efforts to build our box build and system integration capabilities and capacities.
We see strong opportunities to expand this business in 2025, and while demand trends for the second half are still evolving, we remain optimistic about future growth. Control boxes, PDUs, and system enclosures are critical components in SPE tools, and these house, protect, and regulate electrical, mechanical, and fluid control systems, ensuring reliability, safety, and efficiency in mission-critical environments. Control boxes serve as the command center for mechanical and electrical processes, controlling wafer etching, deposition, and metrology, while PDUs manage electrical power to multiple subsystems in high-power environments, including high-precision wafer processing machines. System enclosures protect and organize electrical, mechanical, and control components from environmental hazards, physical damage, and electromagnetic interference. They are typically found in cleanroom environments. The bulk of our box build and system integration business up until late last year was mainly in electrical distribution systems, or EDS.
We are looking to shift from being a simple assembler to becoming a strategic partner by not only reinforcing our NPI capabilities to raise exposure to higher complexity projects and expanding our global low-grade Class 10K Cleanroom capacities, but we are also expanding our exposure to fluid distribution systems, or FDS, on the back of our Easys M&A as we seek to differentiate from low-value commodity assemblers. We seek to show the SPE toolmakers proof of quality control, contamination prevention, and advanced assembly expertise, and aim to secure multi-year supplier contracts. There are also industries outside of the semis that need such capability and capacities. The key benefits of having a stronger NPI for more complex SPE projects include early involvement in product design and development, reduced time to market, enhanced process and supply chain control, increased engineering and system design credibility, and higher margins and long-term contracts.
The key benefit of having Class 10K Cleanrooms includes that it enables entry into FDS, enhances capabilities for Box Build and system integration, and it strengthens competitive positioning. These lower-grade cleanrooms allow for pre-assembly and initial leak testing to ensure airborne particulates are low enough to prevent initial particle contamination before final purification steps in higher-grade cleanrooms. They can serve as a bridge to enter higher-value markets by enabling precision assembly, contamination control, and specialized testing in Class 1 to Class 100 Cleanrooms without major upfront CapEx by implementing progressive cleanliness controls. Finally, whereas EDS can be around 40%-50% of total equipment build cost, FDS can be around 50%-60%. FDS are much more complex due to precision contamination risks, dynamic control requirements, and material compatibility issues.
Now coming to IT and data comm, the HPC landscape is evolving rapidly, driven by massive CapEx, hardware innovation, and power efficiency constraints. AI compute expansion is limited by energy supply rather than chip availability and bandwidth speeds, pushing alternative compute architectures. Hyperscalers dictate AI infrastructure standards, which will have GPUs and ASICs coexisting. Those edge computing-driven applications with direct line of sight with their own computer will be those that will be cloud-based. If you think of a computer as an engine, then you will not have the best car at the race if the car does not have the best parts. Our data and power solutions are platform agnostic, and hyperscalers prefer financially solid suppliers with a full product portfolio, and those with a long-term roadmap due to high switching cost.
No single winner will dominate entirely, but direct tier-one relationships with hyperscalers will ensure our place within their growth plans. The AI boom is far from over. It is simply transitioning into a multifaceted market structure. Our tier-one supplier relationship with leading hyperscalers is a key part of our strategy, giving us an early advantage as one of their partners. Our efforts in HPC were boosted by the launch of AECs from 2019 and by INBG data center business from 2022, with sales more than tripling from 2020 to 2022 before seeing a brief down cycle in 2023. Data solutions sales grew more than four times, while power solutions more than doubled during that time. Although data solution sales in 2024 have not yet recovered from 2022 highs, the bias exiting 2024 is very strong, and we may bridge the remaining gap this year.
Power solution sales grew from 2019 to 2024, more than doubling from 2022 to 2024. Our first mover advantage in AECs and power solutions positioning, agility in NPI, and strategic industry positioning and market awareness are key competitive advantages. We expect to see high growth here from the 2023 lows and are working on next-generation platforms and designs with customers and supply chain partners, contribution for which will be for 2026 and beyond. Our HPC solutions portfolio does not just stop at current-generation product. In data, we will continue to develop faster high-speed DACs, AOCs, AECs, and server cables, all of which have mid- to high-level of customization given the wide variety of RAC specifications. We will also develop co-packaged copper and co-packaged optical for future high-speed connectivity as copper's favorable cost-performance ratio versus optical will reverse at 448G per lane.
400G is the current mainstream bandwidth, and they are useful for switch-to-switch, switch-to-server, switch-to-switch, and accelerator-to-accelerator interconnects. 800G may start to see some volume by the end of this year, meaning 800G may be the mainstream speed for 2026 and 2027. Power consumption is still much lower than optical, continuing to favor AECs versus optical. The number of lanes will be eight at 112G per lane, showing a relatively slow progression in terms of data evolution. 800G can still be multiplexed to reach the 1.6T, which is more than enough for the majority of today's AI high-end demand. In power, the requirements are moving much faster in order to support the latest and greatest computing platform.
If you just look at the subsequent iterations of NVIDIA GPU platforms, the anticipated AI GPU peak rack density increases from 130 to 250 kilowatts for the Blackwell, 250 to 900 kilowatts for Blackwell Ultra and Rubin, and then 900 to over 1000 kilowatts for the Rubin Ultra. The average rack density used to be only 8.2 kilowatts back in 2020. NVIDIA's global data centers were consuming power equivalent to that of individual nations in mid-2024. Supplying enough power to our customers' racks is a challenge. Standard power distribution architectures are struggling to keep up with the rising demands of AI accelerators. Traditional DAC power distribution introduces conversion losses. As a result, we are developing high-voltage DC to reduce conversion steps and minimize transmission losses, leading to efficiency improvement, including moving to 48-volt DC architectures or even 400-volt DC for AI workloads.
Busbar power distribution provides high current, low-loss power delivery. Liquid-cooled busbars would further reduce resistive heating and improve power efficiency. Energy constraints could slow AI infrastructure expansion if power generation and distribution fail to keep up, but this is not the case with bandwidth speed. We will probably see more AI data centers being co-located with power plants, including LNG or nuclear-powered ones, and a shift towards self-sustaining AI energy ecosystems in the coming years, which is what one of our major customers is doing right now in the U.S. Felix will now provide updates on our latest quarterly corporate takeaways.
Right, thank you, Brian. Yeah, I will have the corporate highlights here and also the business outlook. So SPE, that's Semiconductor Production Equipment. Toolmakers are outsourcing non-critical tools to external vendors due to several reasons.
Leading SPE companies are focusing on in-house R&D and manufacturing on their most advanced, high-value tools. Less complex and lower-margin sub-components are outsourced to optimize capital efficiency, and it enables them to scale manufacturing capacity without increasing internal headcounts or CapEx investments. Geopolitical uncertainties have driven SPE toolmakers to build more resilient, regionally distributed supply chains. They are outsourcing non-critical subsystems to trusted manufacturing partners, especially to global vendors with footprints in Southeast Asia, Europe, and North America, to mitigate the risk of logistics disruption, tariffs, and export restrictions. They are under time pressure to accelerate deliveries for advanced EUV metrology and deposition tools, and the outsourcing of non-differentiated subsystems allows their internal teams to focus on high-precision, IP-intensive innovations.
Finally, SPE toolmakers are seeing higher materials and labor costs due to inflation and supply chain disruptions, and engaging in outsourcing lowers their overhead costs, improves cost predictability, and allows for more flexible cost structures in SPE tool productions. It's a win-win scenario for BizLink and our customers. The content and value upside in moving from more traditional cable assemblies and wire harnesses projects is sizable. Our bigger picture strategy is to gradually increase exposure to SPE toolmakers by positioning ourselves as a reliable assembler, and then later expanding into system integrations and co-design. Our near-term strategy includes establishing an assembly and box build business for SPE tools, strengthening cash conversion and profitability controls, and initially expanding into higher-value added system integration. I would say we are moving further into this space right now.
Our mid-term strategy includes transitioning from assembly to partial design and integration, enhancing cash flow optimization while scaling growth and expanding our partnership for advanced SPE box build solutions. Our long-term strategy includes becoming a full-fledged strategic partner for SPE toolmakers, establishing financial disciplines for sustainable growth, and achieving differentiations through innovation and market positioning. We have identified and are benefiting from what we have identified to be the low-hanging fruits within our general strategy. This involves low-complexity box build, power distributions, and control panel opportunities within standard maintenance environments. While we will continue to benefit from this in the near term, we are progressing into what we deem to be more mid-tier opportunities, which involve medium-complexity fluid and chemical handling opportunities within class 10K cleanroom environments.
We may eventually grab some high-hanging fruits, and this involves high-complexity full system integration and R&D co-design opportunities in class 1 to class 100 cleanrooms. We'll also look at what our other business units can offer to our SPE toolmaker customers, including cobots. This outsourcing trend will likely accelerate over the next few years, given the semiconductor equipment demand boom, starting with the growth of HPC and then leading to growth in various AI computing-driven applications such as humanoid robotics and autonomous driving. Global semiconductor manufacturing expansion driven by government-led initiatives and subsidies to build local production and supply chains, and industry consolidation and efficiency optimization. We do not expect to see a reversal of this trend, given leading-edge tool demand will likely stay robust for the foreseeable future. Export restrictions, which are widely anticipated to escalate, typically do not target non-critical subsystems.
Suppliers from less politically sensitive countries and third-party suppliers are not aimed at hurting Western SPE toolmakers' business and are not meant to be overly complex so as to be easier to enforce. On the IT and DataCom, AI infrastructure spending is currently being driven by economic, technological, and geopolitical incentives. Stopping or reversing this would need large disruptions or priority shifts, albeit periodic slowdowns and regulatory adjustments are still possible. With humanoid robotics and autonomous driving on the horizon, the chances for sustained high level of investment in 2026 are high. 8-lane 20G lanes currently support up to 5 m, while 6-lane 224G is viable up to 2 m, with cost reductions still possible. The outlook for 3.2T in copper versus optical remains uncertain. Advances in compute cost efficiency could drive earlier adoption of CPO from 224G rather than waiting for 448G, especially in specialized high-performance developments.
However, widespread adoption is unlikely due to cost and serviceability challenges, as CPO is better suited for specific use cases with higher infrastructure demands. The bulk of hyperscale environments may see limited need for CPO, and enterprise adoption of higher bandwidth beyond 400G AEC appears minimal. AEC could stay the dominant solution longer than expected, with ongoing advancements in copper-based high-speed data cables pushing 800G capabilities to 7 m. That said, AECs are not the sole viable approach. As AI models scale, their power demands are rising at an accelerating rate, creating challenges for data centers and power infrastructure. High-power interconnect solutions are vital for improving power efficiency and enabling the next generations of HPC platforms. We are working with industry partners on off-rack solutions to address thermal constraints and efficiency concerns, as in-rack approaches become increasingly difficult to sustain with growing power requirements.
Moving power conversion and distribution outside compute racks may reduce the immediate need for certain components, but will drive demand for higher specification solutions. The expansion of data centers is expected to continue, presenting long-term growth opportunities for off-rack power solutions. We look to leverage existing tier-one relationships with hyperscalers, existing strengths in high-power and high-speed interconnects, and diversify capabilities to hedge against industry shifts. This approach ensures gains across multiple industry trajectories. By executing a multi-scenario strategy, we ensure profitabilities and meaningful market presence regardless of how the AI compute landscape evolves. The key to long-term success is not betting on a single AI chip vendor, but instead to supply the essential infrastructure that all AI players need: power and data connectivities. Only those with long-term staying power can offer and achieve this. Now, let me turn the call over to Mike. Thank you, Felix, Charles, and Florian.
This concludes our prepared statement section. Now, let us begin the Q&A section. Please type in your questions if you haven't already. I've received some in the chat and also some in the open chat as well as the direct chat. And we'll answer as many of them as possible in the time we may need to end this one-hour call. And for the first question, let's just let me read it in the interest of time. It's coming from KGI's Terry. BizLink delivered strong HPC revenue growth in 2024, even without meaningful contribution from GB200 and with only limited contribution from AEC. Please kindly add more color on how we achieved such a goal. Was it mainly from market share win, new client penetration, or content win? For this question, I'd like to hand it over to Florian.
All right. Thanks, Mike.
Thanks, Terry, for the question. Our HPC business has indeed done quite well, as well as also the capital equipment business. If we just look at HPC, our platform-agnostic high power and high-speed solutions both grew last year. And I think we shared a lot of color on them earlier in this call. However, if you want a summary, I would say that this growth was a result of all three aspects: market share on one hand, new client penetration also, but also content win, which is probably kind of the same thing as the first one. However, this would not have been possible without a very, very close collaboration with our customers, providing them solutions through our high agility in development as well as in deliveries.
But also, we realized that our HPC exposure may be fairly one-sided right now in both terms, like exposure in terms of computing platform, but also exposure in terms of customers' exposures. But we fully expect that we will be able, and we have our strong track record in being able to diversify over the time to even things out in that respect. So we are already seeing today fruits from our efforts in the current quarter, and hopefully, we can also share some of those details in the next call in two months' time. Back to you, Mike.
Thank you, Florian, for the colorful insight, as well as from KGI, Terry, for that question. Now, for the next question, trying to take care of all the ones that talk about AI.
From Nomura, Kenny, and Daiwa, Helen, I'm asking this question every quarter, I guess, which is fairly fair, given all the market news and rumors as of late. What changes does the management team see in the order of receiving sales momentum for AI server products currently versus a quarter ago? Are they stronger, flat, or weaker? And then the second question is, what's the implication of things like DeepSeek? So for these particular questions, I'd like to hand it over to Felix.
All right. Yeah, I would say, well, in general, we haven't seen the momentum change much. Yeah, some customers raised their near-term outlook, while others suggested lower. So the total spend is still there. And we see many power and data projects still incoming, and actually, they are many, I would say, with OEM, with ODMs.
There are varieties of customization we need to provide, and we need to develop with our customers. Overall, we see a momentum still there. DeepSeek, again, I mean, this is actually we see that we believe this cost-efficient AI model could indicate that lower training and inference cost may, in fact, contribute to increased demand of AI compute in the coming years. No matter it's like a high-required high power or anyone like algorithm or how that develops, still we see the hardware requirements will just continue to grow. Rather than reduce, I think people may think that we only rely on a few IC suppliers, but actually, we look at the whole AI industry, and we think that this actually is a good thing. More diversified, I mean, the solutions and also will increase the total volume.
So yeah, so as AI applications continue to scale, the need for high-speed networking, power, and compute is likely to remain strong. And advancing computing systems through various stages, from pre-training scaling to post-training scaling and test time scaling, will likely require more computing power rather than less. So actually, DeepSeek exemplifies the approach toward test time scaling. So in general, we think that's actually a good thing to have more players into this field.
Thank you, Felix. And thank you to Nomura, Kenny, and Daiwa, Helen for those two questions. I'd like to address one final question from the direct chats before I get on HPC. This one's coming from Alice Su . Many smaller cable connector suppliers have claimed that they're capable of manufacturing similar products for GB200/GB300 with likely lower cost. What's the competitive advantage of BizLink now as a supplier?
For this one, I'd like to hand it back to Florian.
Yes, thanks, Mike. And thank you for the question. I think we actually touched on this topic quite extensively during our last quarterly earnings call, in both sections, in the prepared remarks as well as in the Q&A section. But let me give you a brief summary. Many of those smaller competitors claim this, and they may eventually gain some market share at a much lower price because of their lower cost, but they work with the ODMs in a very limited product area. They do just not have the balance sheet to work on larger projects from the beginning, as many do not launch their products until everything is specified. And also, hyperscalers prefer solution providers rather than only product suppliers, especially those with a comprehensive portfolio and a roadmap.
This is necessary to get into the hyperscalers and to be known as a partner for the hyperscalers. Many of these small competitors are not able to quickly service such high-caliber customers on a global scale as they have just limited production and service capabilities. So they oftentimes will win content in situations where the design has already been done and the specification is easy and just to copy and make, but there's less value add there. Being a development partner, as I said earlier, this ensures and supports a fast time to market for our customers. This is highly valuable. Achieving this requires a very close collaboration, as already said, with the customers, and very important is high agility and high speed in both areas, development as well as in delivery, so in my view, these are the key strengths and bring value to the table.
This is one of the success factors of our business at the moment. I hope that answers the questions. Back to you, Mike.
Thank you, Florian, for the color as well as MasterLink Alice Su . There's two more questions from the direct chat, and then I'll head over to the open chat area. For this next second to the last one, BofA Doris, any sign of recovery for industrial segment and non-semi-equipment segment and auto segment? For this one, I'd like to hand it over to Felix.
All right. Yeah, so we have several categories within industrial. The SPE capital equipment, which has now become the biggest category, will continue to grow. Healthcare will slowly expand, and energy, even though right now it's a much smaller scale, will also see a robust growth.
Factory automations used to be the largest category, and we feel that there are signs that this is bottoming, and a sustainable recovery may take some time. Overall, our industrial sales have increased quarter on quarter for the fourth straight time, with the total sales at all-time high, with more upside to come. The way that peers define their industrial business may be different than ours. As for automotive, we stay the most conservative here, and there is more time needed before there's a break. Back to you, Mike.
Thank you, Felix, for the feedback on that one and BofA Doris for that question. The last group of questions before we go into the open chat area. Actually, they also talk about, in general, what we're working on next.
This is going to be from Nomura's Kenny, KGI's Terry, and Yuanta's Bill, UBS's Ally, and of course, SinoPac's Vicky. The questions, new end applications, can you elaborate more on opportunities on humanoid robots and autonomous driving? What shall be the major products and content value in the competitive landscape? Which application does the firm identify that could be the next big theme? After HPC and RE humanoid robots, what's your expectation on cable shipment in the next five years? What's BizLink's strength in gaining shares? And let's see, the last one that's in English. What's your strategy to penetrate this market? Any difference between the US and China markets? And for this one, I'd like to hand it over to Roger.
We have another question here. Okay. So let us focus on the humanoid robot here since it's quite hot and new here.
So we have multiple areas that our two business groups could work on when it comes to humanoid robots. But some of you, the participants, know that many full-scale contributions will not occur this year, as in our opinion. And busbars for the integrating electronic, mechanical, and power component into the functional subsystem may be needed. Certain components may need clean rooms, possibly less than 10K or 100K units of production or assembly, such as robotics, in bulk precision, electro-optic, and high-performance actuator. And this humanoid robot will need to be charged, yeah, to provide power. So meaning it will be stationary and docked at one point, and it may use battery technology such as lithium-ion or solid-state battery or LFP, meaning the battery management system will be used. PCIe and low-speed Ethernet could be used for data communication.
Miniaturization and highly flexible cable are in our warehouse for high torsion, durability, and bendability. Silicone jacket high-flex cable could be in the joint. High torsion, durability, and bending capable cable will be needed in a wrist, less so in the fifth unless you need the robot to break dance. Finally, existing industrial dress pack and the healthcare robotic business will be instrumental, and we have all this in-house. Our strategic advantage, just like in HPC and in capital equipment, is that we can become the one-stop shop for humanoid robot. We will be a high-value partner, and we are working with some of the customers in this area on some of the early-stage projects. Okay? Now, back to Mike.
Thank you, Roger, and thank you for those questions from that group of analysts. Now, for the next question, I'd like to address J.P. Morgan's Bill.
Revenue exposure to IT and data comm went up 20% in fourth quarter 2024 versus 26% in third quarter 2024, but gross margins dropped quarter on quarter. What factors pressure gross margin "declined"? For this one, I'd like to hand it over to our CFO, Charles.
Okay, great. Thank you, Mike. And thank you, Bill, for this question. Actually, I would say this is more about our usual seasonality where we'll see an early season. So our costs tend to be higher for the post-quarter. That's held true to our past several years. You see this pattern. And also, there are always some end-of-year adjustment, accrual, and reversal during the end of the year. And actually, we also accrue some of our rebate to customers. That will be very usual end-of-year adjustment for that. So yeah, if you look at our fourth quarter, you will see this pattern.
So that is more or less why we'll see a little bit quarter on quarter margin drop. But we continue to see a favorable, I would say, favorable product mix in post-quarter and is even growing. So yes, I think that will more or less answer the post-quarter question.
Thank you, Charles. And thank you, Bill, for that question. Now, for the next one. Now, for the next question, thank you, Philip, for your two questions. I'd like to read this real quick for everyone. What's your view on the adoption and technical development of AEC versus ACC versus DAC in the next few years? The second question, how would these three compete if more GPUs are installed in a rack? Higher compute density. Can you talk about the pros and cons of using copper in a blade server structure?
For this one, I'd like to hand it over to Felix. And please go ahead.
Yeah. Yeah, for us, I mean, we see that actually with some more computing power required and also higher speeds, definitely we see that AEC will continue to grow, as we mentioned earlier. And also AOC, continuing. I think we'll try right now to stretch to seven figures, maybe even more. So of course, I mean, it's like the balance between speed and length, but we do see more room to grow for the AEC. And DAC, of course, for the very short cables within a rack or any of these, definitely they are still markets. They are considered cheaper compared to AEC. So I think that's a balance. About the second question, how would these three compete more GPU on?
So, in a rack, I think it's a, I mean, even with more computing power, I mean, we see that you will have a unique higher bandwidth and unique more ports. So I still think that, again, AEC will continue to, how do you say that, to overtake the DAC? We're not even talking about ACC here. Yeah. So that's our viewpoints.
Thank you, Felix. And thank you, let me see, Philip, for that question. Let me see for the next one. Thank you, Brian. Let me see this for this question. I think it's probably going to be the last one. How diversified is BizLink's AEC revenue across customers and projects within individual customers? Beyond Amazon, how do you see demand trends for other customers for the remaining of the year? For this one, I'd like to hand it over to Charles.
Okay. Thank you, Mike.
And also thank you, Brian, for this question. I think when we're talking about AEC, we need to bear in mind that we have, I would say, mutual exclusivity with Credo. We're partnership with them on this AEC product. And also, Credo actually is a relatively, I would say, new customer and new in the market. And it's very natural that their pattern, they tend to grow, I would say, they're still in their young period and their business pattern and the way they attract customers tend to be, I would say, a little bit more volatile. So yes, no, they talk about their large exposure to Amazon, but along the way, we work with them. We also see they work with different customers. And going forward, we cannot bet on them. But we believe that going forward, they will also diversify their customers among those hyperscalers you see.
So we're confident that they will gradually have a healthier diversification and exposure to all those big customers that you see.
Thank you, Charles, for that, and Brian for that question. I do see you have two more. We can address that separately as it's now 3:30 P.M. for this one-hour call. Thank you, Roger, Felix, Florin, 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 the BizLink investor relations team. We thank you very much for joining today's call. And once again, thank you to Yuanta. You may now disconnect.