Prepared remarks
Hello. I am Ken Hsiang, the Head of Investor Relations at ASE Technology Holdings. Welcome to our second quarter 2026 earnings release. I'm joined today by Dr. Tien Wu, our COO; and Joseph Tung, our CFO. Thank you for joining us today. Please refer to our safe harbor notice on Page 2. All participants consent to having their voices and questions broadcast via participation in this event. If you do not consent, please refrain from asking questions or leave the session now. I would like to remind everyone that the presentation that follows may contain forward-looking statements. These forward-looking statements are subject to a high degree of risk, and our actual results may differ materially. For the purposes of this presentation, dollar figures are generally stated in New Taiwan dollars, unless otherwise indicated. As a Taiwan-based company, our financial information is presented in accordance with Taiwan IFRS. Results presented using Taiwan IFRS may differ materially from results using other accounting standards, including those separately presented by our subsidiaries. For today's presentation, Dr. Tien Wu will begin with a midyear business update. I will then walk through the Q2 results, and Joseph will close with our third quarter outlook. With that, let me hand the presentation over to Dr. Tien Wu.
Good afternoon. I would like to give you the first half 2026 recap and also the full year outlook. For my presentation, I will be all in U.S. dollar terms. Consolidated revenue grew 24% year-on-year in the first half of 2026, with ATM revenues up 35% year-on-year for the first half. Leading-edge advanced packaging and overall testing outpaced growth. For ATM business, we expect to maintain the same growth momentum into the second half. For full year, LEAP services revenue is tracking ahead of prior guidance of USD 3.5 billion, while general segment is expected to grow by 20% year-on-year versus previous guidance of 13%. So for the full year, we expect the ATM business revenue to grow by 35%. Machinery CapEx was USD 2.7 billion. Building facility automation was USD 1.4 billion in the first half. Joseph will give you more detail for the full year. We are stepping up investment in R&D, human capital, advanced capacity and also automation and smart factory infrastructure to support multiyear growth. On the second page, I would like to give you some highlights on market dynamics and positioning. There are many moving parts in the market today. You're reading the same newspapers and watching the same news as I do. What I'm trying to present to you is the company view. We will try to present the logic behind why we are making particular decisions at this particular junction in time. AI-enabled new applications have larger scale and potential. I'm not going to articulate the details here; if you're interested, we can talk more in the Q&A. That's our current view. I think very few of you will disagree with the statement that AI is a paradigm shift, and we do have potentially larger scale and multiple applications. In terms of when and how, we are monitoring the progress. I believe we are at the beginning of the AI paradigm shift. There will be multiple stages of transition. We can talk more in the Q&A. The second comment is how we feel. AI demands new hardware that did not exist previously in size, complexity and integration. You can argue about the competition intensity, the memory, the power, the linkage, the bandwidth. All in all, what we're trying to develop right now is a brand-new platform to support potential AI applications. The data center, agentic systems and future physical AI such as humanoids — all of the hardware will be different than what we have been producing so far. There is a growing need for industrial power, connectivity and storage devices because of the AI transition and evolution. The company is seeing all three right now from our customers on a multiyear basis. Let me talk about ASE's strategic priorities. Again, this is the company view. We have to have a blueprint, a plan, a vision in order for the 100,000 employees to act on. What I'm trying to present to you are the highlights of that blueprint. Hardware infrastructure is a bottleneck. With AI, the hardware requirement is new, insatiable and more complicated. Today, there are very few manufacturers capable of producing the required hardware. Therefore, it is a bottleneck today for capacity, for automation and, more importantly, from an innovation perspective. We can talk a little bit more about panels, CoWoS, glass substrate, VRM, silicon photonics; I could go on. But all of these are tied to the infrastructure and your capability to ramp with the complexity, integration and the design blueprint customers are asking for. All in all, I call all of this 'hardware infrastructure,' and that is the new bottleneck. We have not experienced this for the last 40 years. Packaging is moving up in the system architecture value chain. For me, as a designer by training, system architecture is always at the top of the value chain. Packaging is approaching system architecture value by providing the new complexity and integration capability with the variables I just mentioned. Now, if you believe hardware is the new bottleneck and packaging is moving up in the value chain, then ASE has a unique position to support the AI migration, evolution or paradigm shift and align with our customers' long-term objectives. ASE's competitive position — I have discussed this many times — includes the ecosystem position in the Taiwan cluster, scale for AI data center and other digital scale, and efficiency. I want to single out the pure-play aspect. Being pure-play enables seamless cooperation with all supply chain players. In the future, this could be a competitive advantage for ASE as a pure-play OSAT; we have no conflict with foundries or substrate providers. We have no conflict with anybody. Therefore, we have a good way not only to collaborate with our customers long term, but also to collaborate with all ecosystem players that will be critical for the overall complex, integrated nature of the AI transition and evolution. The first-mover advantages are specific: technology, speed, capacity and, most importantly, trust. Everything the company does circles around long-term business objectives as well as seamless integration with ecosystem players to deliver speed, efficiency and earn customer trust. So that's the highlight for the second half. Thank you.
Thank you, Tien. As Dr. Wu highlighted, our businesses performed extremely well throughout the first half of the year. For the quarter, we saw strong growth within both our LEAP and our general businesses. Non-LEAP capacities like wire bond and traditional advanced packaging were also tight. For test, our wafer sort and final test capacities were also running near full. Effectively, outside of equipment lines being placed into service, capacities were generally very close to being full. Our blended utilization rate for the quarter was between 80% to 85%. Our capability for near-term incremental growth is being gated by our abilities to install capital equipment and build out our buildings and facilities. From a financial perspective, second quarter ATM revenues came in ahead of our original expectations driven by higher loading from both our LEAP and general services. From the profitability perspective, we also saw our gross margin outpacing our original expectations, driven by higher loading, resulting in more operating leverage. Revenues for our EMS business were generally in line with our expectations. EMS profitability was slightly below our expectations due to product mix and higher component prices. Please turn to Page 5, where you will find our second quarter consolidated results. For Q2, fully diluted EPS was TWD 4.61, and basic EPS came in at TWD 4.80. We believe our earnings reflect strong core profitability and a few nonoperating items I'll outline shortly. Consolidated net revenues reached TWD 191.1 billion, up 10% sequentially and 27% year-over-year. Gross profit was TWD 40.2 billion with a gross margin of 21.0%, up 1 percentage point sequentially and 4 percentage points year-over-year. The sequential improvement stems from higher operating leverage in our ATM business. The annual improvement reflects both structural efficiency gains and a more favorable NT dollar environment. Operating expenses totaled TWD 19 billion, up TWD 3.5 billion year-over-year, primarily driven by higher labor costs and further R&D spend to support LEAP initiatives. As a percentage of revenue, operating expenses remained flat at 10% and declined 0.3 percentage points annually. This delivered an operating profit of TWD 21.1 billion, up 21% quarter-over-quarter and 107% year-over-year with an operating margin of 11.1%, expanding 1 percentage point sequentially and 4.3 percentage points annually. Nonoperating income totaled TWD 4.6 billion, compared to TWD 0.7 billion in the prior quarter. While this amount appears relatively elevated versus prior quarters, the increase primarily reflects some potentially nonrecurring gains. Our net nonoperating income includes mark-to-market equity gains related to our corporate investments of TWD 4.2 billion, foreign currency hedging gains of TWD 1.5 billion and other items, including equity method investment income, government grants, rent income and dividends totaling TWD 0.8 billion. These gains were partially offset by TWD 1.9 billion in net interest expenses. Tax expense was TWD 4.2 billion. Our effective tax rate came in at 16.4%. We saw a lower effective tax rate due to higher R&D credits generated during the quarter. We continue to expect an effective tax rate of 18% for the year. Net income for the quarter was TWD 21.1 billion, up 49% sequentially and 180% year-over-year. On Page 6 is a graphical view of our consolidated quarterly performance. For the second quarter 2026, our ATM business represented 66% of our consolidated holding company revenue while representing 94% of our operating profit. This is compared to 61% of consolidated holding company revenue, while representing 87% of operating profit in the second quarter last year. We see this being primarily driven by the growth of our ATM LEAP services over the last few years. On Page 7 is our ATM P&L. The ATM revenue reported here contains revenues eliminated at the holding company level related to intercompany transactions between our ATM and EMS businesses. For the second quarter of 2026, we recorded record revenues for our ATM business of TWD 126.1 billion, up TWD 13.7 billion sequentially and TWD 33.6 billion annually, representing an increase of 12% sequential and 36% annual growth. Gross profit for our ATM business was TWD 34.5 billion, up TWD 5.3 billion sequentially and up TWD 14.2 billion year-over-year. ATM gross profit margin was 27.3%, up 1.3 percentage points sequentially and 5.4 percentage points annually, driven by higher operating leverage and a more favorable revenue mix, particularly a higher mix of LEAP. During the second quarter, operating expenses were TWD 14.7 billion, up TWD 1.4 billion sequentially and TWD 3.3 billion year-over-year. The sequential and annual increases in operating expenses are primarily related to higher overall labor costs and general R&D expenses. Our operating expense ratio was 11.7%, down by 0.1 percentage points sequentially and 0.6 percentage points annually. We expect our ATM operating expense ratio should continue to improve during the back half of 2026. During the second quarter, operating profit was TWD 19.8 billion, representing a sequential 25% increase of TWD 3.9 billion and a 124% annual increase of TWD 11 billion. Operating margin was 15.7%, up 1.6 percentage points sequentially and up 6.2 percentage points year-over-year. The NT dollar had a positive 0.1 sequential and a 0.6 annual percentage point impact to our gross and operating margins. On Page 8, you'll find a graphical representation of our ATM P&L. The chart highlights the improvement in our gross profit margin. It should be noted here that our second and third quarter 2025 margins were heavily impacted by NT dollar strengthening. Over this time frame, our margin improvement has been largely driven by the recovery of our utilization rate related to our general manufacturing capacities and increasing LEAP product mix. On Page 9 is our ATM revenue by the 3C market segments. LEAP services are primarily included within our computing applications, with a lesser amount being included in the communications applications. As can be seen here, the computing application percentage continues to grow steadily. At this time, we see this trend continuing into 2027 and 2028. On Page 10, you will find our ATM revenue by service type. Despite the overall growth in our business, we did not see substantial shifts in service types during the quarter. All business lines appear to be keeping pace outside of small adjustments between materials and others. At the beginning of the year, we believed that our test business, led by rapid expansion of our wafer sort business, would outpace growth in our assembly business. While our test business has shown the strong growth we initially expected, our assembly business has been showing even stronger-than-expected growth, especially as it relates to legacy wire bond services. We now believe that both our assembly and test businesses will grow at similar rates during the year. On Page 11, you can see the second quarter results of our EMS business. EMS revenues grew 6% sequentially and 12% annually to TWD 65.8 billion. Sequentially, our EMS business' gross margin decreased by 0.6 percentage points to 8.9%. This change was principally the result of product mix differences. EMS operating expenses increased by TWD 0.3 billion sequentially and annually. Our second quarter EMS operating expense ratio of 6.5% was flat sequentially and down 0.4 percentage points annually. Operating margin came in at 2.4%, down 0.6 percentage points sequentially and 0.2 percentage points year-over-year. The sequential margin decline is the result of product mix and a higher component cost environment. Our EMS second quarter operating profit was TWD 1.6 billion, down TWD 0.3 billion sequentially and up TWD 0.1 billion annually. On the bottom of the page, you will find a graphical representation of our EMS revenue by application. Generally, the moves in the consumer and communication categories are related to the seasonality of the underlying products we service. The growth in the computing category is largely driven by business related to our AI accelerator products. From the holding company perspective, we continue to pursue synergies between our ATM and EMS businesses to co-develop system-level solutions, particularly in key areas such as optical interconnects, power delivery and thermal management. By integrating critical EMS competencies with ATM technologies, we have the potential to optimize entire systems end-to-end, applying the same co-engineering disciplines that successfully scaled SiP architecture. On Page 12, you will find key line items from our balance sheet. At the end of the quarter, we had cash, cash equivalents and current financial assets of TWD 107.4 billion. Our total interest-bearing debt increased by TWD 40.9 billion to TWD 306.2 billion. Total unused credit lines amounted to TWD 396.2 billion. Our EBITDA for the quarter was TWD 45.8 billion. Our net debt to equity this quarter was 47%. On Page 13, you will find our equipment capital expenditures relative to our EBITDA. Machinery and equipment capital expenditures for the second quarter in U.S. dollars totaled $1.7 billion, of which $840 million was used in packaging operations, $804 million in testing operations, $49 million in EMS operations and $2 million in interconnect materials operations and others. In addition to spending on machinery and equipment, during the quarter, we also spent $658 million on facilities. It is worth reiterating what Dr. Wu spoke of earlier. We are investing in capacities and facilities because what we do has immediate impacts on key bottlenecks in semiconductor supply, performance and efficiency. The AI build-out, regardless of open or closed models, will require unprecedented hardware capacity and capability expansion. At this point, we are just trying to keep up. With that, I'll hand the presentation over to Joseph to walk through the company's outlook.
Okay. Now for third quarter 2026 outlook. Based on our current business outlook and exchange rate assumption of USD 1 to TWD 31.9 versus last quarter's TWD 31.6, management projects overall performance for the third quarter of 2026 to be as follows: On a consolidated basis, in NT dollar terms, our consolidated third quarter revenue should grow by 21% to 22% quarter-over-quarter. Our consolidated third quarter gross margin should be between 12.5% to 21.5%. Our consolidated third quarter operating margin should be between 11.5% to 12.5%. For ATM, in NT dollar terms, our ATM third quarter revenue should grow by 11% to 13% quarter-over-quarter. Our ATM third quarter gross margin should be between 28% to 29%. On EMS, in NT dollar terms, our EMS third quarter revenue should grow by around 40% quarter-over-quarter. Our EMS third quarter operating margin should be between 3.2% to 3.4% — with that is the '26 third quarter outlook. Now adding a bit of color for the full year and next. First, on CapEx. Given stronger demand for LEAP in 2026 and beyond, we will need to add another USD 1 billion each this year for facilities and equipment. So the total is adding another USD 2 billion for CapEx. While the additional investment for facilities and most of the equipment are for LEAP, we also need to add capacity for mainstream advanced packaging and testing to support the general market demand. On LEAP, well, this year's LEAP service revenue is tracking ahead of our prior quarter guidance of USD 3.5 billion — as business momentum continues to be very strong, we are aiming to double our LEAP revenue in 2027. Lastly, on ATM profitability, with expanding margin-accretive LEAP and test businesses, our second quarter ATM gross margin of 27.3% came in ahead of our guidance. We continue to expect sequential margin improvement with fourth quarter ATM gross margin likely to exceed our structural margin ceiling of 30%. At such point, we will start to review if we would adjust our structural margin range. With that, thank you.
During the Q&A session that follows, we would appreciate if your questions could be as clear and concise as possible and ask singularly. We will start by taking questions from live participants and then alternate in questions from our online participants. I, as the moderator, will be receiving each question and repeating and directing each asked question. After an initial question, the participant may ask a follow-up question, clarifications of the earlier question or another question entirely. Then we'll move to the next participant. Participants may return to the queue for any additional questions and/or clarifications. Thank you.
Questions and answers
Dr. Wu, just to follow our conversation. Your industry peers seem to be more aggressive in U.S. operations; not just Amkor, but also KYEC announced expansions. I want to ask you about your partnership with U.S. customers and also your operations in the U.S. I guess it's a bit of a separate question, but can you also comment a little bit about Intel EMIB involvement by ASE?
Charlie, you're asking about the competitive landscape that we're facing and secondly, EMIB. Dr. Wu?
Well, let me answer the U.S. operation first. What I'm going to talk about, I have repeated several times over the last two years. We do have U.S. operations, including R&D, test development and services in California. We have one factory in Fremont, California. We have one factory in San Jose. We're in the process of expanding to third and fourth factories. These are particularly required by our customers in the Bay Area for upfront chip design, test development and technology development, and that is ongoing right now. The agreement we have with our leading customers is the following: we will develop and build fully automated efficient lines in Taiwan. We are comfortable with our resources and efficiency. In due time, we will migrate and move some operations to other places in the world. It could be the United States, could be somewhere else. That has always been the case. So it's not like we're not supporting. We are supporting development, R&D and architectural design. In terms of manufacturing process development, for now, we are focusing on Taiwan. Until we build the appropriate scale, have the appropriate resources and know-how and efficiency, upon that time, we will work with our customers to move to other parts of the world for better logistics. The second comment is EMIB. EMIB, I'll repeat what the foundry has responded: right now, it's capacity constrained. If there's any alternative technology that can offer the same yield to relieve the bottleneck we have for the AI infrastructure, we welcome it. For ASE, we're also collaborating with customers along the same line, including EMIB. The question is whether EMIB poses a potential threat or competitive advantage over CoWoS. That's always a possibility. I have been working in packaging for 40 years and have seen many competing technologies. At the end of the day, only a subset become dominant. There's always competing technologies and alternative materials; that's always been the focus of R&D. For EMIB, we're happy to see if it can ramp up. In terms of efficiency and performance, that's up to the system and the market to decide. We're not making a judgement on superiority. Our focus is to support CoWoS and ramp up CoWoS scale and efficiency as fast as we can. If customers ask for alternative technologies, ASE will include them in our roadmap. I don't think it's a zero-sum game. If EMIB substrate becomes the right alternative and we do the assembly, there is no conflict. Anything is welcome. The world is a big place.
More questions from the floor.
My first question is: can you give us a little bit more detailed guidance for your Q3 ATM demand drivers for both LEAP and general packaging and testing — more detailed demand drivers across key applications or products?
Rick, you're asking for segmented drivers of what's helping the industry or our results pick up at this point. Is that correct?
The question is the Q3 driver. We are in an awkward and peculiar position because we're capacity constrained. So we talk about 11% to 13% growth; that means we have to add 11% to 13% capacity. Demand overall is strong. I'm not going to comment about memory pricing or consumer devices. But overall, all of our customers are asking for more devices for Q3 and Q4.
Rick, follow-up?
No follow-up. Second question: can you elaborate on your development of CoWoS? This year, roughly how much is still coming from outsourcing? And what's your development and progress into next year for your internal CoWoS development?
Rick, you're looking for the composition of our LEAP services between outsourced services and full-process internal work. Correct?
I think basically, we are on track with our full-process business development. This year, we said that we're going to have about TWD 300 million worth of revenue coming from that space, and things are on track. We are aggressively expanding that capacity. By next year, I think we will have pretty substantial growth in that area as well. This is under development. Margin-wise, it has not fully reflected yet. But going forward, I think full-process will also be another margin-accretive business for us with pretty substantial growth coming into next year.
We have our next question coming from online.
Next question is from Sunny Lin of UBS.
Could you hear me okay?
Yes, we can hear you.
Congrats on the very strong outlook. Sorry for not being able to attend in person. My first question is a follow-up on 2027 LEAP outlook. You are guiding for this year's LEAP to exceed $3.5 billion. Should we be looking at maybe $4 billion or somewhere between $3.5 billion to $4 billion? And then for 2027, for it to double, do you have an idea in terms of breakdown by full-process outsourcing, on-substrate work and test?
Currently...
Well, I think Tien mentioned that we are ahead on LEAP revenue this year. By ahead, I think we will be adding another couple hundred million dollars worth of revenue this year. And on top of that, for next year, we still see very strong momentum, and we believe at this point that we should be able to double the LEAP revenue next year. We are going full speed ahead with both our assembly as well as test. As Ken mentioned, assembly is also catching up in terms of its growth momentum with test. So for next year, I think the combination will be pretty similar to what we're seeing this year.
Let me clarify: for LEAP breakdown going into 2027, I assume full-process will account for a much larger portion in 2027. Any color would be helpful.
Sunny, you're looking for more color on LEAP guidance for this year and into 2027. Correct?
Probably two quarters from now we'll give you better color. Thank you.
Second question: Dr. Wu, you earlier commented that AI is driving many new applications with different complexities and scope. Based on your current engagement with clients, what types of devices are you seeing better visibility on, and which may have a meaningful ramp in the coming few years?
Sunny, you're looking for more color in terms of potential applications and devices we may see in the future tied to AI?
I would like to separate the technology statement versus the business statement. Technology development can take 15 to 20 years. The business statement typically focuses on the next four quarters. Let me comment on CoWoS and EMIB and alternative technologies. AI data center is driving computation intensity. Therefore, reticle sizes become larger, and everybody demands more bandwidth from logic, ASICs and memory. The immediate upfront need will be larger panels or more complicated CoWoS to accommodate bigger chiplets or reticle sizes. That's one dimension. We also expect that toward the end of the year, co-packaged optics (CPO) will start launching in small volumes. Initially it could be small, but it will provide critical benchmark information on bandwidth, system performance, and thermal dissipation. Depending on cost-performance ratio and yields, we'll learn critical information. When reticle sizes become bigger and optics enter a different hierarchy, power delivery will become the next immediate question; that will be deployed in the next two to three years. On the much longer term, there is significant activity to integrate sensors, analog and mixed-signal with digital. If you think about humanoid applications, other than the brain, the eyes, ears and fingers are all analog. So what technologies can provide low power, high bandwidth and fully automated, high-volume integrated capability for digital-analog mixed-signal is important. These are the kinds of infrastructure a pure-play OSAT supplier should develop with customers. Technology development takes 15 to 20 years; much of what we're deploying today is the result of past efforts. I want to make that separation. AI has a long runway. You can listen to many smart people about the infrastructure, AI data center, agentic systems and physical interfaces. My simple view is that AI is about new pattern recognition in familiar domains. When people start crossing domains — IT versus pharmaceuticals, IT versus medical — new patterns emerge and enable new applications. We're at the beginning of knowledge collection and early-stage pattern recognition. Today we mostly talk about single domains; eventually we'll bridge many domains. The technologies we're building now are the building blocks for that future. What we're doing is not just for AI data centers; engineering is about building elegant solutions for future demand regardless of difficulty. It takes 15 to 20 years. This is not a short-term trade. But we are first movers; we already have the cluster efficiencies in the early stage of this AI transition, and that matters if you understand the point.
Next online question is from Gokul Hariharan.
Dr. Wu and Joseph, maybe first to Dr. Wu: you've been working in very close partnership with a lead foundry partner for the last few years which has been helping LEAP revenue growth. They seem to be transitioning more aggressively towards 3D panel-level future technologies compared to the prior few years. As they embark on that transition, could you help us understand ASE's role? Is it getting elevated further in that partnership? Some of the full-process CoWoS is a result of them leaving some of these older areas to you as well. Could you outline how this technology partnership with the lead foundry will work in the next two to three years?
Gokul, you're asking about how our plans and role intermix with the foundry partner's roadmap and migration. Is that correct?
And do you get more value add out of it as they migrate to more complicated packaging?
The collaboration is long-term. Once again, I want to focus on the pure-play model. The foundry is ideally pure-play for wafers. Testing, bumping and packaging are enablers to deliver wafers to customers at the right time and with the right efficiency. Packaging pure-play develops packaging architecture and all the 'Lego' pieces to facilitate that. When a pure-play foundry and a pure-play packaging partner collaborate, the boundary is clear. Today, there are packaging areas done by foundries because wafer access or certain architecture IP is tightly controlled. That's a domain decision for the foundry and customers. But aside from that, collaboration covers all aspects of packaging. Where we stop or start depends on the management teams in the ecosystem. As you can see, OS, full-process and other collaborations span the full spectrum. Ultimately we all want to provide speedy, elegant solutions in the shortest time possible to our customers. That understanding is clear and will be obvious for the next three years as the ramp-up occurs.
Got it. My next question is about a couple of guidance increases you have: LEAP for 2027, where previously we thought an incremental ~$1.9 billion; now you're expecting doubling. Is that primarily coming from better line of sight into capacity availability given demand still exceeds supply? Secondly, mainstream now is expected to grow close to 20% this year versus 13% previously. Where is that upside coming from for mainstream growth?
Gokul, you're looking for explanation behind LEAP 2027 outlook and the mainstream upside. Correct?
We have clear line of sight on who needs what and on the buildings and facilities we're building today. When we commented that by year-end we're tracking ahead of the $3.5 billion guidance, that comment is made because we see yields and execution that give us confidence we'll achieve that target by year-end. At the same time, once that target is achieved, with the next 12 months of building new facilities and adding machines, we will be able to double that. That line of sight is there. The uncertainty is execution — our ability to execute. Business demand is not a concern; it's our capability to execute two lines of sight that are clear in front of us. For general devices, we see very strong demand in industrial, power, connectivity and storage devices. I don't have the overall semiconductor market number, but I believe we will outperform the general market because of our association with AI infrastructure development and our capability to run fully automated lines for many general devices. Customers building high-impact, high-risk devices for EV, AI data center or other applications often prefer our fully automated lines. That's why we're seeing good traction. So capacity expansion must include LEAP and the general market. That puts tremendous pressure on ASE, which is why we are building many facilities simultaneously this year. We're building, buying and spending CapEx. We're not happy about the pressure, but it's our obligation to provide critical capacity to satisfy customers' long-term objectives.
Charlie, do you want to ask another round of questions? Yes, go ahead. Microphone.
Ken, I have two questions and one clarification on the previous Q&A. About LEAP revenue next year: Joseph, you talked about the mix being similar to this year next year. But according to our analysis, your 2.5D end customer CPU parts are growing threefold and testing is also growing more than double. How can substrate outsourcing also double next year? NVIDIA Street consensus is next year GPU growth of around 50% and TSMC CoWoS expansion at 70%. How do we reconcile these views?
I was referring generally that assembly and test seem to have similar momentum. Assembly includes not just outsourced substrate services; there is full-process work and other process steps we are entering. So that explains some of the mix differences and why assembly growth can look similar to test growth overall.
Thank you. Second question: I notice your quarterly CapEx exceeding EBITDA — is free cash flow turning negative? Do you think the share price will react negatively tomorrow?
Yes, we will continue to have very heavy CapEx this year and into next year. The negative cash flow situation will remain for some time. At the same time, we maintain a very healthy balance sheet and have multiple cost-effective funding sources to fund upcoming CapEx requirements. We are confident we can support growth in a healthy manner. We're still in the early stage of this mega trend, and we will make necessary investments not just to support our customers but to maintain leadership.
Last question on pricing and margin: we keep hearing customers say you hiked prices aggressively in the second half. Can you confirm if this is the case, and whether price increases are more in traditional services or advanced packaging? Secondly, how does this translate into your long-term gross margin outlook — you had a 26% to 29% margin range for years; can you break 30% anytime soon?
I said earlier we are very likely to exceed our structural margin ceiling in the fourth quarter. At that point, we'll review whether to adjust our structural margin range upward. We're in a fairly favorable pricing environment, and while we've seen inflationary pressures in materials and components, so far we can pass these cost increases on to customers through pricing arrangements. We'll continue to pursue suitable pricing strategies considering the situation and our margin requirements.
Are you okay? Okay. We'll go back online for the next question.
Our next online question is from Haas Liu of Bank of America.
When you commented that gross margins are likely to exceed the ceiling of your structural gross margin range in the fourth quarter, can you share how much of that improvement would be from mature parts of the business versus being lifted by your growing mix in LEAP contribution?
Haas, I think you're asking which parts of the business are contributing to the margin expansion in Q4: mature products versus LEAP mix. Correct?
Margin improvement results from a combination of factors: margin-accretive business growth, improvements in efficiency, continuous expansion of our automated factories and operating leverage as volumes grow. On the operating side, our past investments are starting to pay off; OpEx ratio is improving yearly. Overall efficiency has improved and our revenue base is solid, which gives us confidence to continue sequential margin expansion this year and next. Once we pass the structural margin range, we'll review how far we can go.
A quick follow-up on CapEx: would you provide a breakdown? You mentioned the general market demand is strong; which parts of the business will see the most spending — advanced versus mature?
Haas, do you want a quantitative breakdown for this year's CapEx or a directional view of where we are spending?
This year we're raising CapEx by another $2 billion, bringing the total to about $10.5 billion. Out of this $10.5 billion, about $4 billion will be for new factory buildings and facilities and $6.5 billion for equipment. For the equipment portion, roughly 56% is for assembly, 40% for test and the remainder for EMS and materials. In terms of leading-edge versus mainstream, about 70% of equipment CapEx is for leading edge this year. We currently have 13 greenfield projects under construction and another eight brownfield projects where we're acquiring existing factories and renovating them. These projects should cover our needs into 2028 and part of 2029. Managing 20 projects simultaneously is challenging and requires efficient construction and execution; execution is critical.
Do we have another online question?
We have an online question from Gokul Hariharan of JPMorgan.
On CapEx: I know you're not guiding for next year, but given the gap between supply and demand you're facing, especially for LEAP, is it fair to expect CapEx will keep rising into 2027?
Gokul, you're asking for an indication about 2027 CapEx.
I think it's better if we wait another quarter to get more clarity on next year's needs. We expect it will be significant, but we want to have clearer visibility before giving guidance.
Could you also talk about CoWoS and panel-level packaging development? Based on your current assessment, when might these start entering production given varied views in the market?
Gokul, you're asking about our panel process and timing to production?
For ASE's panel process, our fully automated line will start production by Q1 of next year in a 310 by 310 form factor. Regarding CoWoS and glass substrate, we are evaluating many alternative materials with substrate suppliers, foundries and customers to assess feasibility and economics. But right now, we do not have glass substrate production planned in the next 12 months.
How does ASE's 310 by 310 solution differ — is it for a different market or customer compared to foundry offerings, or is it complementary?
They're quite complementary and target the same customer set with similar reticle and pitch size ranges. Which customers adopt which solution depends on capacity, performance and speed of execution.
To clarify CapEx equipment breakdown: this year, about 70% of equipment CapEx is leading edge. Equipment CapEx is roughly 56% for assembly, 40% for test and the remainder for EMS and materials.
Gokul, we got you covered there.
Next online question is from Sunny Lin of UBS.
My first follow-up is on CPO. Given the complexity of the technologies, what services will ASE be able to offer? Based on current development, when might CPO revenue become meaningful for ASE? Also, given your EMS capabilities, what synergies do you see across ATM and EMS for optical modules and interconnects?
Sunny, you're asking what ASE will offer for CPO and how EMS capabilities might provide synergies. Correct?
On CPO, I'd like to wait a couple quarters before giving more detail. Regarding optical hierarchy, the hybrid of electrical and optical signaling is a definitive direction. The question is when and how to execute at system levels from near-field to outer-field. Some optical devices are mature and used widely; near-field optical solutions at chip and substrate levels need more development and are difficult, which is why the industry has taken time. By the end of the year we'll have more data on how these behave and their benefits. We've been working on this area for about 20 years; two more quarters will give us useful datapoints.
Quick follow-up on your full-process CoWoS: given better visibility into 2027, should we assume CPU will be the major contributor, or will accelerators and other applications also provide substantial volume?
Sunny, you're asking about product-based diversification for full-process CoWoS revenue mix into 2027, correct?
For full-process, we should have better visibility in a quarter or two. Right now we have line of sight on capacity that we're developing and execution is tracking well. In terms of customers — GPU, CPU, ASIC — we need more time to digest how much we can share. In two quarters' time we should be able to provide LEAP services revenue outlook for next year and a clearer breakdown of OS, full-process and others. Our client portfolio covers these product types, and we're waiting for yield and market performance data.
Our next online question is from Michael Resnickers.
I just want to clarify: when you gave Q3 guidance, did you say EMS will grow 40% quarter-on-quarter in revenues? That's a pretty large sequential increase — what's going on there?
Michael, you're asking for clarification on the EMS Q3 growth of about 40% quarter-over-quarter.
The ~40% sequential growth for EMS is due to abnormal seasonality and component price effects, particularly in memory prices. If you take that pass-through component price effect out, Q3 EMS moves would reflect more typical seasonality.
Makes sense. How about Q4 — will EMS revenue continue to grow versus Q3 or will it shrink?
At this point, we expect a similar level of EMS revenue in Q4 compared to Q3.
Overall for the year, did you say you're looking for about 25% plus revenue growth?
Michael, are you asking for full-year growth expectations?
EMS for the full year is expected to be sub-20% growth. ATM business we expect to be 35% plus growth for the full year.
Got it. And the impact on EMS margins with larger memory-related revenues — is EMS margin flat year-over-year or does it increase or decrease?
For Q3 specifically, if you remove the component price hikes, operating margin for EMS would be in a typical seasonality range around 3.7% to 3.8%. Memory costs are largely pass-through.
Okay, thank you and congratulations on the strong performance.
Do we have more questions online? More questions on the floor? No? Very good. Thank you very much. I would like to thank everyone for attending our conference call today. Joseph, do you want to close up?
Well, I'm sure we're going to have another good quarter in third quarter, and we'll bring you some more good news next quarter. We'll see you next quarter.