Taiwan Semiconductor Manufacturing Company Limited (TSM) Q3 2026 Earnings Call Transcript

Jeff Su: Good afternoon, everyone, and welcome to TSMC's third quarter 2026 earnings conference call. This is Jeff Su, TSMC's Director of Investor Relations and your host for today. TSMC is hosting our earnings conference call via live audio webcast through the company's website at www.tsmc.com, where you can also download the earnings release materials. If you are joining us through the conference call, your dial-in lines are in listen-only mode. The format for today's event will be as follows. First, TSMC's Senior Vice President, third quarter 2026, followed by our guidance for the fourth quarter 2026. Then Mr. Huang and TSMC's Chairman and CEO, Dr. C.C. Wei, will jointly provide the company's key messages. Then we will open the line for the Q&A session. As usual, I would like to remind everybody that today's discussions may contain forward-looking statements that are subject to significant risk and uncertainties, which could cause the actual results to differ materially from those contained in the forward-looking statements. Please refer to the safe harbor notice that appears in our press release. Now, I would like to turn the call over to TSMC CFO, Mr. Wendell Huang, for the summary of operations and the current quarter guidance.

Wendell Huang: Thank you, Jeff. Good afternoon, everyone. Thank you for joining us today. My presentation will start with financial highlights for the third quarter of 2026. After that, I will provide the guidance for the fourth quarter of 2026. Third quarter revenue increased 17.6% sequentially in NT, supported by strong demand for our leading-edge process technologies. In U.S. dollar term, revenue increased 16.2% sequentially to $46.7 billion, above the high end of our third-quarter guidance. Gross margin decreased by 1.3 percentage points sequentially to 66.4%, primarily due to margin dilution from 2 nm ramp. Total operating expenses accounted for 7.8% of net revenue. Operating margin was 58.7%. Overall, our third-quarter EPS reached TWD 29.43 and ROE was 45.5%. Now, let's move on to revenue by technology. 2 nm process technology contributed 8% of wafer revenue in the third quarter, while 3 nm, 5 nm, and 7 nm accounted for 33%, 28%, and 10% respectively. Advanced technologies, defined as 7 nm and below, accounted for 79% of wafer revenue. Moving on to revenue contribution by platform. HPC increased 22% quarter-over-quarter to account for 68% of our third-quarter revenue. Smartphone increased 10% to account for 21%. IoT increased 5% to account for 4%. Automotive increased 23% to account for 4%, while DCE declined 12% to account for 1%. Moving on to the balance sheet. We ended the third quarter with cash and marketable securities of TWD 4 trillion, or $126 billion. On the liability side, current liabilities increased by TWD 110 billion quarter-over-quarter, mainly due to the increase of TWD 46 billion in current portions of bond payable and bank loans and TWD 26 billion in dividend payable. In terms of financial ratios, accounts receivable turnover days decreased by 1-28 days. Inventory days decreased by 5-82 days, primarily due to the increase in wafer shipment. Regarding cash flow and CapEx, during the third quarter, we generated about TWD 920 billion in cash from operations, spent TWD 443 billion in CapEx, and distributed TWD 156 billion for fourth quarter 2025 cash dividends. Overall, our cash balance increased TWD 488 billion-TWD 3.6 trillion at the end of the quarter. In U.S. dollar terms, our third-quarter capital expenditures totaled $13.8 billion. I have finished my financial summary. Now, let's turn to our current quarter guidance. Based on the current business outlook, we expect our fourth quarter revenue to be between $52.3 billion and $54.3 billion, which represents a 14% sequential increase or a 58% year-over-year increase at the midpoint. Based on the exchange rate assumption of one U.S. dollar to TWD 31.8, gross margin is expected to be between 65% and 67%, operating margin between 56.5% and 58.5%. This concludes my financial presentation. Now, let me turn to our key messages. I will start by talking about our third quarter 2026 and fourth quarter of 2026 profitability. Compared to second quarter, our third quarter gross margin decreased by 130 basis points sequentially to 66.4%, primarily due to the margin dilution from the initial ramp-up of our 2 nm technology. We have just guided our fourth quarter gross margin to decrease by 40 basis points to 66% at the midpoint, primarily due to the continued margin dilution from the steep ramp-up of our 2 nm technology, as well as dilution from overseas fabs. These impacts are expected to be partially offset by cost improvement efforts, including productivity gains and across node capacity optimization. While the cost of overseas fabs remain higher, thanks to the company's overall larger scale and cost improvement efforts, we now expect the gross margin dilution from the ramp-up of our overseas fabs to be closer to 2% more effort to support the strong structural demand from our customers. More fabs will be ramping up in the U.S. Thus, we expect the dilution from our overseas fabs to widen to 3%-4% in the next few years. We will leverage our increasing size in Arizona and work on our operations to improve the cost structure. We will also continue to work closely with our customers and suppliers to manage the impact. Overall, with our fundamental competitive advantages of manufacturing technology leadership and large-scale production base, we are confident that TSMC continues to be the most cost-efficient manufacturer in every region that we operate. Next, let me make some comments on our 2026 capital budget. At TSMC, a higher level of capital expenditures is always correlated with higher growth opportunities in the following years. Given the continued robust multi-year demand from our customers, including the emerging agentic AI market, we reiterate our full year 2026 capital budget to be between $60 billion and $64 billion. We continue to expect our CapEx investment from 2026 to 2028 to be even more significantly higher than the past three years as we continue to invest heavily to support our customers' growth. Additionally, while the tool supply is tight, we always collaborate closely with our tool suppliers well in advance to prepare the capacity, just like our customers collaborate with us well in advance to plan our capacity. Even as we invest for the future growth with this level of CapEx spending in the next few years, we remain committed to delivering profitable growth to our shareholders. We also remain committed to a sustainable and steadily increasing cash dividend per share on both an annual and quarterly basis. Now, let me turn the microphone over to C.C.

C.C. Wei: Thank you, Wendell. Good afternoon, everyone. First, let me start with our near-term demand outlook. We concluded our third quarter with a revenue of $46.7 billion, higher than our guidance in U.S. dollar terms, driven by strong demand for our leading-edge process technologies. Moving into fourth quarter 2026, we expect our business to be supported by continued strong demand for our leading-edge process technologies, including the steep ramp-up of our 2 nm technology. Looking ahead, we continue to observe consumer and price-sensitive end market segment are being challenged due to the impact of rising component prices and macroeconomic uncertainties. As such, we are being prudent in our business planning going into 2027 while focusing on the fundamentals of our business to further strengthen our competitive position. Having said that, demand for our leading-edge technologies continue to be extremely robust, supported by our strong technology differentiation and broader customer base. Therefore, we now expect our full year 2026 revenue growth to be around 44% year-over-year in U.S. dollar term. Next, let me talk about the AI-related demand outlook. The AI market continues to be very strong. Recent developments are very positive to AI's long-term demand outlook. We observe increasing AI demand fueled by the rising adoption of agentic AI as we transition from chatbot to AI assistance. In addition, we are observing great adoption of open source AI models. This will drive greater efficiency and help lower the barrier to future AI usage, and result in more and more demand for leading-edge silicon. Thus, AI-related demand is being supported by strong underlying demand for AI accelerators. In addition, the emergence of agentic AI CPUs and the continued demand for networking chips are driving further silicon demand. Therefore, semiconductor will remain essential in the AI mega trend. The value of semiconductors in the global supply chain is set to continue to increase in our view. This will create more value opportunities for our customers to capture and more value opportunities for TSMC as well. No matter which approach is taken, you will require use of the most advanced semiconductor process technology. With our technology leadership, manufacturing excellence, global capacity expansion, and customer trust, we are well-positioned to capture the full multi-year structural AI-related demand. Therefore, all the AI players are working with TSMC. Next, let me talk about TSMC's capacity expansion strategy. As a foundry, our first responsibility is to provide our customer with the best technologies and effective capacity support to unleash their innovations. To address the structured increase in the long-term semiconductor market demand profile, TSMC collaborate closely with our customers, yet our customer's customer to plan for our capacity. In addition, we are now going one step further and sending our own teams of people to check on the data center build-out progress and power supply situation in the U.S. The feedback is so far so good, and the CSPs has been well-prepared. Internally, TSMC employed a disciplined capacity planning system to assess the market demand from both a top-down and bottom-up approach. This is a continuous and ongoing process. Given the fundamental complexity of leading-edge technologies and the design in and lead time involved, we already have a very good idea of customers' multi-year product roadmap and production plans. Thus, we continue to hold strong conviction in the AI mega trend. We are moving full speed ahead to increase our capacity to meet our assessed demand, and stepping up our CapEx investment to support our customers' future growth. TSMC is expanding at unprecedented pace with a nearly 20 fabs worldwide in progress. We are building 13 leading-edge and advanced packaging fab in Taiwan over the next several years, and another five to six fabs overseas. Our new node, N2, has already entered high-volume manufacturing in the fourth quarter of 2025 with good yield. N2 is ramping successfully in multiple phases at both Hsinchu and Kaohsiung site. We expect our N2 family technologies annual capacity to more than double in 2027, supported by strong demand from both smartphone and HPC AI applications. We are also leveraging our manufacturing excellence to generate more N2 wafer output, further increasing our capacity support to customers. In addition, we have already begun preparing the capacity for our A14 technology. Risk production is well on track for 2027, and volume production will start in 2028. Thus, TSMC's semiconductor technologies and manufacturing will continue to play a pivotal role in supporting the global semiconductor industry while unleashing our customers' innovation. Now, let me talk about TSMC's global N3 capacity expansion update. As we said before, historically, we do not add additional capacity to a node once it reaches its targeted capacity. However, based on our assessment, we are executing well on our global capacity plan to add three new fabs to support the robust multi-year pipeline or demand for 3 nm technologies. In Taiwan, volume production for our new 3 nm fab in Tainan Science Park is well on track to begin the first half of 2027. In addition, driven by the strong demand from our customer, we have decided to add one additional 3 nm fab in Tainan. Volume production is scheduled for the first half of 2028. In Arizona, we have completed the construction of our second fab and are in the process of tool moving and installation. Volume production will begin in third quarter 2027, slightly ahead of our previous schedule. In Japan, we are progressing well to our plan to utilize 3 nm in our second fab, and volume production is scheduled in 2028. In addition to all the new fabs, we continue to convert 5 nm tool to support 3 nm capacity in Taiwan. We are also leveraging our manufacturing excellence to greater productivity across our fab in all locations. To generate more wafer output, we are also focusing on capacity optimization across nodes, which including flexible capacity support among N7, N5, and N3 node. Thus, we are using multiple levers to do everything we can, whenever we can, as fast as we can, to maximize the support to our customers. We have no intention to leave food on the table for others, and we want to capture every piece of business possible. This concluding our key message, and thank you for your attention.

Jeff Su: Thank you, C.C. This concludes our prepared statements. Before we begin the question-and-answer session, I would like to remind everybody to please limit your questions to two at a time to allow as many participants as possible the opportunity to ask their questions. Should you wish to raise your question in Chinese, I will translate it to English before our management answers your question. For those of you on the call, if you would like to ask a question, please press the star then one on your telephone keypad now. If at any time you'd like to remove yourself from the questioning queue, please press star two. Now, let's begin the Q&A session. Operator, can we proceed with the first caller on the line, please?

Operator: Yes. First one, Laura Chen, Citi. Go ahead, please.

Laura Chen: Thank you very much. Good afternoon, and appreciate C.C. for sharing with us the TSMC capacity expansion strategy. May I follow up more on that? Given the increasing aggressive AI semiconductor demand, how does TSMC now see the AI revenue CAGR growth, and how does TSMC reconcile customer the capacity requirement with some of the potential concern on the data center power availability, and also maybe funding recently and other critical components such as the HBM, APS, advanced packaging, and also in the networking side, we see the tightness of laser as well. If these kind of trends create a mismatch between semiconductor supply and also actual AI development, how would TSMC adjust your capa-minimize the potential overbuilding in the supply chain? Thank you. That's my first question.

Jeff Su: Okay, Laura. Laura has a very involved first question. I think she is asking how do we see the AI demand outlook. The AI demand, as C.C. said, seems to be very strong, but she is worried or concerned, how do we reconcile the strong demand of chips, AI chips, versus risk related to whether it be power supply, I think also different bottlenecks within the supply chain, et cetera. How do we make our assessment of what is the appropriate CapEx and capacity we need to prepare? Is that correct, Laura?

Laura Chen: Yes. Thank you.

C.C. Wei: That is a long, long question. Laura, let me share with you. We have a very sophisticated capacity planning process, and we also have a very disciplined on our consideration. Let me share with you, we put every possibility we can, talking about the CSPs, CapEx, data center build-out, financial, power supply, and also other component supply, et cetera, into our consideration. We make our own judgment, and then based on our assessment the demand, we still think the demand is higher than the supply, and we are working very hard to narrow the gap. Did I answer your question, Laura?

Laura Chen: Sure. Thank you. Thank you very much. So far from our comprehensive study of the supply chains, we are still quite positive on the demand in the longer term run. In that case, can you also give us more color about that? Any update on your AI CAGR growth? I recall previously we talking about like a high 50%.

C.C. Wei: We-

Jeff Su: Sorry. Sorry. Yeah. Part of Laura's question is then based on our assessment, we had previously guided our AI accelerator CAGR, which is a narrow definition to-

Laura Chen: Correct.

Jeff Su: grow mid to high 50s percentage from 2024 to 2029. Laura wants to know, do we have any update to that number today?

C.C. Wei: Laura, all I can say right now is the demand is extremely robust, okay? We did not see any sign of slowdown. Whether it is a high 50 CAGR, it does not matter. It probably will be higher.

Laura Chen: Sure. Okay. Yes.

C.C. Wei: Okay.

Laura Chen: Thank you very much. My second question is also about, given the very strong demand, the CapEx cycles, we see that definitely this round will be much higher than before. TSMC also announced that earlier, you will adapt a High NA EUV around 2030. Should we think about that will be further improve the productivities, thus will be more meaningful for further accelerating revenue growth in the longer term? Also given that the CapEx intensity are also getting much higher, so how should we think about the pricing power and also the long-term gross margin and the capital efficiency return?

Jeff Su: Okay. That is actually two questions, but okay. We will just very quickly, so Laura's part of it is on, with the strong demand on our CapEx cycle, how do we view the adoption of High NA, which we said we will in 2030, and then her question is on profitability, long-term outlook.

Laura Chen: Yeah. Thank you.

C.C. Wei: Let me answer that question. Yes, for the long-term productivity, we continue to improve it, and High NA definitely is a very good tool. So far, I think our plan is to use it in 2030. Again, our CapEx will be much higher than today. You are talking about the gross margin or the profitability can be increased or not. Let me say that, there is no ceiling for TSMC's profitability, but a dramatic increase will not happen overnight either. Our philosophy or our strategy always are working with our customer, and we want to make them be successful, and then we can gain our value. Yeah.

Laura Chen: Sure. Thank you. Very helpful. Appreciate it.

C.C. Wei: Okay.

Laura Chen: And congrats for the great result. Thank you.

C.C. Wei: Thank you.

Jeff Su: Thank you. Operator, can we move on to the next participant, please?

Operator: Next one, Gokul Hariharan, JPMorgan. Go ahead, please.

Gokul Hariharan: Hi, good afternoon, C.C., Wendell, and Jeff. My first question, again, on capacity addition. Looks like the shortage. Just wanted to hear, C.C., your view on this. I think it's been quite some time since we have seen TSMC add capacity at more than 10% every year. I think if you look at your annual report, I think it'd be about 10 years that you've added capacity of more than 10%. Given this accelerated build-out, can we start getting wafer capacity growth closer to 10% or even higher than 10% in the next couple of years? Any comment on what is your planned wafer capacity growth in the next couple of years? That's my first question.

Jeff Su: Okay. Gokul's first question is regarding the capacity addition and the demand. His question first is, the tightness in the wafer supply, do we think that will be even more or greater in 2027 than 2026? Along with that, how do we see our overall capacity addition? Although you say 10%, I think every year we disclose in the annual report, it has been about mid to high single-digit, and sometimes even lower. Anyway, regardless, those are the two questions. First, on just the demand outlook and then the supply. Will next year be even more tight than this year?

C.C. Wei: I just mentioned the demand is extremely robust and getting higher and higher. Of course, we have our own judgment, and we look at it every angle. As I said, we talk to customer, customers are customer, and then we do our own one more step to check the data center build-out. All the result we put into our judgment and our assessment, the result is, the demand is very, very high. The gap between demand and supply are still quite big, and that is why we put all our strengths to increase the capacity, as I just mentioned. Whether it is bigger or the gap become bigger or the gap become narrow, we are working very hard. I cannot answer this question right now because of my customer continue to increase their demand.

Wendell Huang: Okay. Gokul, I think it is a bit too early to talk about capacity increase percentage going forward. Let me just remind you that the increase every year, even though the number may be small, but the increase in advanced nodes are really the main driver for the capacity increase. Okay?

Gokul Hariharan: Fair enough. Okay. Thank you. Secondly, I think you alluded to more aggressive overseas expansion plans in the U.S. especially, and I think recently there has been a lot of discussion with potentially a new fab cluster in Texas, as well as, I think one of your customers also talked about discussions with TSMC as well. C.C., could you talk a little bit about your overseas plan, like the more accelerated kind of capacity build-out? Also any engagement with SpaceX or Tesla, given your customer did call you out last week in terms of some engagements here. What are the broad contours of this kind of engagement likely to be? Is it similar to what you have done with some of the customers in Japan and ESMC as well?

Jeff Su: Okay. Thank you, Gokul. Gokul's second question, he notes that C.C. said we are expanding at an unprecedented place, doing everything we can, as fast as we can to increase the capacity. His question is particularly to the U.S., what is the status or update? There have been reports that we are looking to expand in Texas. There have been reports that one of our customers, we are engaging in discussions. He wants to know what kind of framework or parameters of any cooperation may be, or any comment on either of those two things.

C.C. Wei: Okay. Gokul, we work very hard to expand our capacity in U.S. This is necessary because of all our customer, most of our large customer in U.S., and they request that TSMC to increase the capacity. Also U.S. government are very helpful. That is a place we want to expand. Recently, we did expand our capacities consideration to see outside Arizona whether we can get another chance to expand our capacity. Texas is one. We send the teams to Texas to do all the evaluation. We also talk to one of our customer who try to build a fab, Terafab, let me be more specific. We are discussing with them, see whether that we can be utilizing that actual capacity also. We are still in discussion. If we have any solid conclusion, we will share with you. But today, that is it, yes, from Arizona to Texas, yes.

Gokul Hariharan: Okay. But all these are going to be TSMC-run fabs, right? I think you are not becoming part of the Terafab project or anything like that. Could you just help us understand what are the principles that you are using here? Is it a similar principle that you use for JASM or ESMC?

Jeff Su: Okay, so Gokul's question is not in particular to the potential cooperation with this customer. Would it be TSMC's own? What are the principles that we have, if anything?

C.C. Wei: We are still in discussion, let me say that. It's not either way, it would be possible, but the fab has to be run by TSMC.

Jeff Su: Thank you. Operator, can we move on to the next participant, please?

Operator: Yes. Next one, Krish Sankar, TD Cowen. Go ahead, please.

Krish Sankar: Yeah, hi. Thanks for taking my question. I have two of them. First one, C.C., some customers talk about a secondary wafer market where one customer passes their wafer allocation to another. Is this a very small portion of the business? Is it customer specific, or do you see that as an emerging trend where customers whose demand is weak pass some of their allocation to stronger customers? Then I had a quick follow-up.

Jeff Su: All right, Krish. Thank you. Krish's first question is asking, some of our customers are talking about a resell or a secondary wafer market where they can pass their allocation on. Is this a very small portion of our business, or is this more of a trend for us to do, and is this something that we allow our customers to do?

C.C. Wei: Well, let me answer that. To say whether this is a new strategy, new methodology, the answer is no. The principle of TSMC is we do not allow our customer to sublet or auction off the capacity. But in certain cases, customer can collaborate through a mask licensing agreement, but this is not our common practice. The amount is very, very small. Krish, did I answer your question?

Krish Sankar: Yes, you did. Thank you very much for that. Then a quick follow-up. You did touch upon it, the High NA agreement with ASML from 2030 onwards. I'm just curious, historically, TSMC was more ahead of the curve with the EUV adoption. It looks like by the time you adopt High NA EUV, it would be 2030, 2031, but some of your competitors would already been using it by then. Do you see that as a disadvantage, or you're not worried about competition getting ahead with High NA EUV? Thank you.

Jeff Su: Okay, so Krish's second question is looking at High NA. He notes some of our competitors are both using, already adopted, or will adopt High NA earlier than TSMC. His question, does this put us at a competitive disadvantage or risk compared to our competitors in terms of technology leadership?

C.C. Wei: Krish, let me answer that. We always evaluate a new tool, and if we want to put into the mass production, it got to be the best technology with a very reasonable cost, and most important is the best maturity. Okay? We say in 2030 we will start to use it because we also think at that time, High NA will be mature, and we are also working very closely with our ASML to establish a 12 in photomask by 2031. At that time, we think that High NA's cost will be reasonable and for TSMC to make it into the mass production. This is a very good tool, let me emphasize that. In the smaller geometry, it definitely has some advantage. Whether that our technology, because we are using it later than our competitor, so disadvantage to TSMC? No, because we have confidence that we are still the best company to offer the most advanced technology to our customers.

Jeff Su: Krish, does that answer your question?

Krish Sankar: Thank you very much.

Jeff Su: Yeah. Okay.

Krish Sankar: Yes. Thank you very much, C.C. Thank you, Jeff.

Jeff Su: Okay. Operator, can we move on to the next participant, please?

Operator: Yes. Charlie Chan, Morgan Stanley. Go ahead, please.

Charlie Chan: Hi, gentlemen. Thanks for taking my question and also glad to hear very strong AI semi demand amid some recent stock market correction. My first question is really about your unprecedented capacity expansion strategy. As far as I know, one of your IDM customer is expanding capacity aggressively. There are some concern about this customer may insource more than outsource to you. First of all, I am not sure how you are going to deal with this sort of business opportunity. Secondly, I remember last time that C.C. mentioned about A14 could be a bigger node than N2. May I know whether that assume the big IDM company will outsource more CPU, like server CPU, at A14? Thank you.

Jeff Su: Okay. Charlie's first question is around IDM competitor. He wants to know very specifically, if we see this competitor moving to insource more. His question is then, our prior comment of A14 being a larger, bigger node than N2 has even been, does this include the server CPU from this IDM competitor?

C.C. Wei: Well, Charlie, we treat every customer fairly. Whether it is IDM or it is purely fabless, we treat them the same. But of course, we take into consideration of their internal capacity. We calculate our support, we calculate our capacity build-up according to our own judgment. Actually, in our A14, we have seen more and more customer actually higher than the 2 nm activities so far. We expect that the A14 will be a bigger node than our current N2 technologies.

Charlie Chan: I see. It doesn't need to consider this IDM will also outsource more. We can still have A14 to be a bigger node than N2. Is that the right understanding?

Jeff Su: Well, again, Charlie wants to know very specifically this customer's outsourcing plan, which you probably may need to ask them, but yeah.

C.C. Wei: Well, Charlie, as I said, to be frank with you, I don't comment on a specific customer, but let me guarantee you that we take everything into consideration. Okay?

Charlie Chan: Okay.

Jeff Su: Okay. Do you have a second question, Charlie?

Charlie Chan: Thank you. Yes, I do. Thanks, Jeff. The second question is a kind of follow-up about your advanced packaging. From what we are seeing that the reticle size is getting much bigger. C.C., you said last time, or even two quarters ago, about your response, SoW, 3.5D, so lots of initiative is going on. My question is that, do you think those are kind of a purpose SoW 3.5D can really catch customers who have very rapid developments of their larger reticle size chip, like 14 reticle? Secondly, if some customer, they initially choose to use Intel's EMIB-T, and for some reason cannot get to sufficient demand, are you also support customers for the CoWoS version or capacity. Thank you.

Jeff Su: All right. Charlie's second question is focused on the advanced packaging. I guess, Charlie, if I understand your question, summarize it is, of course, customers are moving to larger and larger reticle size 14x. He wants to know TSMC's roadmap and solutions, and how competitive we can be against, I think he mentioned specifically our competitors, EMIB-T. How do we view our solution? Is it as competitive as our competitor? That's the first part of his question.

C.C. Wei: All right, Charlie. Actually, I still think that TSMC's advanced assembly, or they say packaging technology, are very, very competitive. But the most important thing is that we using that one to help us help our customer's product can be introduced in the market smoothly. That's our number one priority. We have to offer them the best technology with enough capacity. So far, let me share with you today's situation. We are in the real-world production of 5x reticle size. I don't believe that anybody can do that today except at TSMC, and the yield is excellent. So our customer are very satisfied with the result. We are working on the 9x reticle size and moving into the 14x reticle size. Of course, we are also working on the CoWoS, and to make sure that in the future we can offer a reasonable cost reduction. But so far, the CoWoS is still the number one choice for our customer. As time goes by, yes, our competitor offer a very competitive solution. Still, I want to say that today's situation, we are the only one that makes sure that the advanced packaging is a 5x reticle size, and still, we are very competitive. I want to emphasize that.

Jeff Su: And then Charlie's question, if the competitor cannot deliver on their solution, Charlie, I think you're saying, will you be willing to be something like a backup plan?

C.C. Wei: I try to avoid to answer it directly, but let me say that, yes, we will be. Number one priority I already indicate that TSMC's customer's product can be introduced into the market smoothly. That's a number one priority. Okay? And that give you enough hint.

Charlie Chan: Yes. Yes, indeed. So thanks for all the answers. I hope you don't give any food to others, including the advanced packaging. Thank you. It's 50.

Jeff Su: Okay.

Charlie Chan: Thanks, everyone.

Jeff Su: Thank you. All right, operator, let's move on to the next participant, please.

Operator: Next one, Sunny Lin, UBS. Go ahead, please.

Sunny Lin: Good afternoon. Thank you very much for taking my question. My first question is to have a bit more color on how we should think about CapEx beyond 2026. Recall a few quarters ago, you mentioned, if you are doing your job well, then you may be able to drive CapEx to grow a bit less than top line. But now obviously the overall call activity may have further strengthened, and you are also willing to support your clients with more substantial expansion, and therefore wonder at this point, when you look at maybe in line with sales or even a bit faster, just like for 2026. Meanwhile, obviously the tool price is also increasing, given a very tight supply, and therefore, going to 2027, will the higher tool price also be a driver for higher CapEx as well? Than you.

Jeff Su: Higher, sorry. What was the last part of your question, Sunny? Higher tool price, higher what?

Sunny Lin: Well, so yeah, when we think about 2027 CapEx, should we also consider tool price inflation to be a meaningful variable?

Jeff Su: Okay.

Sunny Lin: For a lot higher CapEx compared with this year?

Jeff Su: Understood. Okay, sorry. Okay, thank you. So Sunny's first question is about our CapEx beyond 2026. Really, she's asking 2027, 2028, the next two years. With the strong, extremely robust demand that C.C. indicated, question is around our capital intensity the next two years. How do we see the capital intensity trending revenue growth versus a CapEx growth? Part one. And then also the impact of tool inflation on the CapEx outlook as well.

Wendell Huang: Okay, Sunny. This is Wendell. It's a bit too early to share the CapEx number beyond this year's. As we said in the prepared remarks, we have very strong conviction in the AI megatrend. Therefore, our CapEx from 2026 to 2028 is expected to be significantly higher than the past three years. This is a step up of our investment to support the customers' growth. We have no intention to leave any food on the table for others. The tool price inflation is real. The tool price, material, labor cost, et cetera. However, the majority of the CapEx is always related to the long-term demand. That will be the composition going forward. Capital intensity-wise, as we always said, if do our job right, the increase in revenue will outpace the increase in CapEx, and that is still the case from what we are looking at.

Jeff Su: Okay.

Sunny Lin: Sunny, thank you very much. If I may have a very quick follow-up on the CapEx side. Earlier, Wendell, you raised the margin dilution for overseas to, I think, 3%-4% starting from 2027, given faster overseas expansion. For the U.S., you're still so far focused on Arizona. Does that imply for Arizona, the expansion for the second phase will happen very quickly, or you are actually accelerating the expansion for the third phase or even the other phases as well, as therefore the result for higher margin dilution for next year?

Jeff Su: Okay, Sunny's question is about the puts and takes to next year's margin. She does note, as Wendell said in our prepared remarks, that the overseas fabs will widen a bit to 3%-4% in the next few years. She wonders what's driving that wider dilution. C.C. did say we are pulling in the schedule for the second fab for 3 nm. Her question is, are we also pulling in third fab? What's leading to this wider dilution? Maybe, yeah.

Wendell Huang: Okay, Sunny.

Sunny Lin: Thank you.

Wendell Huang: As we mentioned, we are doing everything we can wherever we can to increase our capacity to meet the strong demand from the customers. We are also accelerating everywhere, including in the U.S. We did mention before that the latter part of the five-year period, the margin dilution from overseas fab will be 3 percentage point to 4 percentage point, and that still remain the case. So it's 3 percentage point to 4 percentage point in the second part of the five-year period.

Sunny Lin: I see. Thank you very much for the clarification. My second question-

Jeff Su: I am sorry.

Sunny Lin: I want to follow up on the point.

Jeff Su: That was two questions, Sunny. That is already two questions. Sorry, but we probably just to accommodate as many as we can, we are going to move on.

Sunny Lin: Sure.

Jeff Su: Thank you for your understanding. Operator, can we move to the next participant, please?

Operator: Next one, Haas Liu, Bank of America. Go ahead, please.

Haas Liu: Yes. Hi, C.C., Wendell, and Jeff. Thanks for taking my questions, and congrats on the great results. My first question would be on your view regarding the potential shift in AI demand mix. You mentioned in the prepared remarks that the AI accelerator chipset demand is proliferating into agentic AI demand, including CPUs and networking switches. We understand the overall AI investment is still pretty strong and even trending stronger across the board. Can we get more clarity that are you seeing your customers' demand shifting toward more efficient compute solutions, especially monetization gets more important? How should we think about the incremental opportunity for your manufacturing and packaging business? Thank you.

Jeff Su: Haas' first question is on the AI-related demand. His question is, are we seeing a shift in the AI-related demand more and more to CPU or networking switches and away from AI accelerators? Any type of comment, yeah.

C.C. Wei: Well, all I can say is AI's demand is extremely robust. Whether it's come from accelerator or is more from the CPU side, it's not for me to make a comment. Although we did see the demand from agentic AI CPUs continue to increase. Yes, that is a trend today. Other than that, everything is very robust. I want to emphasize again, in our judgment, we consider all the parameter, and the demand actually is still far above our capability today to supply. We are working very hard to narrow the gap. That's all I can say.

Haas Liu: Okay. Thank you. That is very clear. My second question is just regarding the yield learning curve going forward for your dual technology tracks strategy going forward. Historically, I think you have been ramping up the yield of a new node through producing the chipsets with smaller die size and simpler design to improve the learning curve of a new node. But for the future node families supporting both a conventional power delivery for the applications like mobile and consumer applications and Super Power Rail technology for HPC and AI chipsets, would operating two process architectures at the same time make the learning curve more challenging or increase the manufacturing complexity? How should we think about the implication to your gross margins from the yield and also cost control perspective? Thank you.

Jeff Su: Okay, so Haas' second question is really about ramping a new node, the learning curve of yield and cost, and therefore the implication to gross margin, really, I think is what he is driving at. His question is that in the past a new node is mostly ramped with smaller die size and simpler design. But today we have both the smartphone and also with the AI and HPC applications coming in. Does that increase the complexity of the node process architecture, make the learning curve more challenging or difficult, and therefore the implication to the margin?

C.C. Wei: Well, today's situation actually is a little bit different from previous years. Yes, we now have a very strong demand from smartphone and also AI HPC applications. So the smaller die size per se or the big die size in the accelerator, they happen at the same time. This is to TSMC's advantage, let me say that. Because we are good. We have been prepared for this kind of thing, and happy to working with us. So the learning curve, yes, but we are doing a reasonable job, and so the customer are very happy today.

Haas Liu: Okay. Thank you so much.

Jeff Su: Okay. Thank you. Operator, let's move on to the next participant, please.

Operator: Arthur Lai, Macquarie. Go ahead, please.

Arthur Lai: Hi. I really appreciate management taking my question. I have a question on the power as a gating factor. I want to understand, have the power grid constraint led you reduce your customer forecast in the history? Also, could this tighter and tighter power grid surprise make your customer actually accelerating adoption of your Super Power Rail product for the better performance power? Thank you.

Jeff Su: Okay. Arthur's first question is about the power supply or electricity supply. Is it a gating factor, and have we seen that lead to a reduction in customers' forecast? On the other side or on the other hand, with the tighter power electricity supply, have we seen more customers adoption of Super Power Rail technologies?

C.C. Wei: Well, let me update that. Definitely power supply situation is one of the consideration factor that we are looking into. That's why also we do one more step to double check whether the power supply situation in the U.S. And we have done all the study, and so far so good, let me say that. Whether this kind of situation change our customers' behavior and move more into the backside power rail? Answer is no. Whether you use the backside power, those kind of metallization scheme is based on your need for your product. Okay? Some of the accelerator might be using it, some of the product like mobile phones, they don't use it. So it depends on customer's product. Did I answer the question?

Arthur Lai: Yeah.

C.C. Wei: Okay.

Arthur Lai: Yes. Thank you, C.C.

Jeff Su: Do you have a second question?

Arthur Lai: Yeah. Thanks. Yes. So just recall the previous discussion on the agentic AI demand on the CPU. So I wonder, according to our previous framework, demand forecast framework, the mid to high 50s growth including the CPU or not, and if we heard our clients increase the CPU forecast significantly, when is the next time we refresh our AI demand growth numbers?

Jeff Su: Okay. So Arthur's second question is on our, we already stated, but anyway, our AI accelerator definition. He wants to know with the emergence of agentic AI and CPU, does that AI accelerator definition include CPUs or not, and when we will provide inclusion for him.

C.C. Wei: Let me answer the question. It is not included in our AI demands calculation. The reason is very simple, because we cannot identify which CPU user in the conventional server CPU or AI server CPU. Also, it depends on our customer's application and their own allocation. So, no. So far today, we stick to our initial very narrow definition on the AI's demand.

Arthur Lai: Understand. If we have seen a very strong CPU-driven demand, when will we revise up the-

Jeff Su: Arthur, I think we already said very clearly it is not part of the AI accelerator definition. It is a narrow definition, but it is certainly part of our business. Okay?

Arthur Lai: Yeah, sure.

Jeff Su: Okay. Operator, in the interest of time, can we take question from the last participant on the line, please?

Operator: Yes. Sahej Singh, Stifel Financial. Go ahead, please.

Sahej Singh: Hello. Thank you for the question. Into 2027, you have N3 turning accretive, and now that is crossed over. N2 is still in its diluted ramp, and the overseas dilution is widening. On the offset side, there have been reports of significant price increases across your mature and advanced nodes. Could you provide some color into perhaps some of that conversation? Where pricing does move, does it typically apply across the existing contracted book, or is it mainly being applied to incremental capacity above what customers have already committed to?

Jeff Su: Yeah. Okay. The first question is, I guess you are asking about the potential puts and takes for next year, or you are asking specifically on pricing? Sorry. For pricing.

Sahej Singh: Specifically on pricing.

Jeff Su: Okay. It is a very specific question that there are reports that we are increasing the price. He wants to know, does it apply across the existing contracted commitments from customers, or is it mainly for the incremental capacity?

C.C. Wei: Well, we have discussed this kind of thing before, that our pricing strategy is very strategic. It also depends on the market segment, depends on the location, and depends on the technology and something. Once we offer the flexibility of the physical location, like in the U.S., it will be a little bit different from the one we produce in Taiwan. No matter what, we try to get our value to be recognized, and we will give our shareholders the maximum value of our technologies.

Sahej Singh: Understood. Thank you. For the second question, something you said in your prepared remarks which stuck out to me was, we will not be leaving food on the table. Earlier, I think it was maybe one or two weeks ago, there was the announcement via GlobalFoundries that they would be providing silicon interposers to TSMC. I would love some color into how are you thinking through these strategic decisions in terms of procuring from some of the specialty foundries versus providing in-house. Thank you.

Jeff Su: Okay. His second question is, again, he notes that recently we had an announcement with GlobalFoundries on interposers. How do we view these partnerships, and what is our approach strategically?

C.C. Wei: Well, let me answer the question. We designed a deal with GlobalFoundries, and this is part of our overall strategy in the back-end supply. Actually, our number one priority is to help our customer to be successful in their market. We want them to be fully supported by all the supply, including the front end and back end. Definitely, we have a little bit shortage on the back end, and we would like to work with our so-called partner with GlobalFoundries so that both TSMC and GlobalFoundries can support TSMC's customer for their product to be introduced with enough support on the capacity.

Sahej Singh: Understood. Thank you very much for my questions.

Jeff Su: Okay. Thank you for your questions. Thank you for all the participants. This does conclude our question-and-answer session. Before we conclude today's conference, please be advised that the replay of the conference will be accessible within 30 minutes from now. The transcript will become available 24 hours from now, and you can find both will be available through TSMC's website at www.tsmc.com. Again, thank you for joining us today. We hope everyone continues to stay well. Have a good autumn, and we hope you will join us again next quarter. Thank you and have a good day.