Jeff Su — Director of Investor Relations, TSMC
Thank you, CC. This does conclude 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 so that we can allow all the participants an opportunity to ask their questions. Questions will be taken both from the floor and from the call. 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'd 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, then two. Now let's begin the question-and-answer session. I will take the first few questions from the floor, then we'll flip and alternate to those on the line. Maybe we'll go left, center, and then right, sort of a sequence. We'll start here with Gokul Hariharan from JPMorgan.
Gokul Hariharan — Analyst, JPMorgan
Thanks, Jeff. Good afternoon, CC and Wendell. First question on demand. I think, CC, you mentioned data center AI demand definitely looks better than maybe three months back. Last quarter, you also mentioned CoWoS capacity will probably come into balance by 2026. Is that still our view, or do you think that the capacity now starts to look tighter? Second, I think you talked about on-device AI as a potential future driver. Are you seeing more development on the on-device AI part?
Is it better compared to maybe three, six months back? Lastly, near term, your Q4 looks like you're expecting revenue to decline. Is that based on what your customers are telling, especially on the consumer side, or is it just TSMC being cautious and conservative in terms of the guidance?
Jeff Su — Director of Investor Relations, TSMC
Okay. Gokul, thank you. Again, for the benefit of those, of course, here in person and on the line, please allow me to summarize your questions. Maybe we'll take them one by one. His first question is on the demand, particularly data center and AI-related demand. As CC said in his remarks, it is certainly still even stronger. His question is about the advanced packaging and CoWoS demand into 2026. How do we see the supply-demand gap narrowing or becoming more balanced for CoWoS specifically?
CC Wei — Chairman and CEO, TSMC
Gokul, the demand from the AI is getting stronger and stronger.
If you pay attention to what the four trading companies' CEOs said. The makeup trend for the AI continues to be strong, and so is the CoWoS. Now we are, again, we are in a mode trying to narrow the gap. I do not want to use the balance. The last time you guys misunderstood what I said, you thought it is a bad wording. I will say we try to narrow the gap. All right. Momentum is still there and very healthy.
Jeff Su — Director of Investor Relations, TSMC
Okay. The second question or second part is on on-device or edge AI. Gokul wants to know how is the development of customers to work on on-device AI compared to maybe three to six months ago? What is the interest or activity level, and how do we see this?
CC Wei — Chairman and CEO, TSMC
As the last time I said, I said that it takes one to two years for my customer to complete their new design on the product. The momentum is still going. They are still continuing to, as time goes by, as I said, the increase on the edge device, the number of the units is actually mild. The die size increased. We continue to see that. The die size increased by about 5%-10%. That kind of trend continued.
You have to wait around probably six months or to one year to see an explosion.
Jeff Su — Director of Investor Relations, TSMC
The second question or the final part on the near term. I think, Gokul, your question is with our third quarter guidance implies fourth quarter. Is there any particular reason or any comment that we want to make about the implied fourth quarter business momentum?
CC Wei — Chairman and CEO, TSMC
I think you did not mention Gokul's comment there, or you will become conservative. That comment is more real. We are a company that what we say we will achieve and achieve the high target. Your calculation, I think Charlie also is nodding. A lot of you are calculating our reported numbers so that you can easily see that our fourth quarter is decreasing. We take into consideration the possible impact of tariffs and a lot of other uncertainties. We become more conservative. That is our current attitude. I guarantee you, with our technology leadership position and excellent manufacturing, if there are any opportunities, we will catch and be expected that we will achieve our high-end target. Okay.
Jeff Su — Director of Investor Relations, TSMC
Okay. Thank you, CC. Thank you, Gokul. Sorry. Let's go one by one. Second question, then maybe Charlie Chan from Morgan Stanley.
Charlie Chan — Analyst, Morgan Stanley
Thanks for taking my question.
Good afternoon, CC, Wendell, and Jeff. First of all, congrats for really strong results, especially on the gross margin side. Very good execution indeed. My first question is really also on the gross margin because the accumulated effect input is almost like 4.4 percentage points. It is too big to ignore. When TSMC considers your 2026 delivery pricing, so-called reflecting your value, would you consider this effect impact and is conveying confidence to keep your margin similar to this year's level? I feel like 53% is a low bar. I just want to make it a little high and see if that effect impact can be considered to reflect your value. Thank you.
Jeff Su — Director of Investor Relations, TSMC
Okay. Charlie's first question is on margin and, I guess, pricing.
He notes, obviously, as Wendell said, the big move in the exchange rate and therefore a big impact to our profitability and gross margin. His question is, looking ahead to 2026, can we reflect or earn our value from including the effect impact into the pricing? Therefore, what is our confidence level on the gross margin for next year? Can it keep a similar level as this year?
CC Wei — Chairman and CEO, TSMC
Let me assure you that, yes, the impact from the exchange rate is huge. You try to imply that whether we are still our value. Let me answer that. We are working on it. We have confidence that the 53% gross margin and higher, I still want you guys to pay more attention to and higher. Okay. Thank you.
Charlie Chan — Analyst, Morgan Stanley
Okay. Thanks. Hopefully, it will work out well.
My second question is also a very hot topic recently about the H20 chip shipping to China. I remember three months ago, there was another question on this matter, right? Meaning that back then, I believe the chip was suspended, but you're still very confident about your mid-40% CAGR for cloud saving growth in the coming five years, right? Now, China becomes your addressable market again. Do you think that mid-40% CAGR target can be revised up?
Jeff Su — Director of Investor Relations, TSMC
Okay. Thank you, Charlie. Charlie's second question is around the AI accelerator demand. He notes, of course, our customers' product, H20, recently now seems to be able to ship to China versus three months ago, it was not. His question really is, our long-term AI accelerator growth CAGR to grow close to mid-40%, can it be higher? Do we think it will be higher? Is there upside to this?
CC Wei — Chairman and CEO, TSMC
Charlie, the H20 now is, again, according to the four trading companies' CEO, we did not receive the signal yet. It is too early to give you an estimate. Certainly, it is good news, right? I mean, China is a big market, and my customer can still continue to supply the chip to the big market. It is very positive news for them. In return, it is very positive news to TSMC. Whether we are ready to increase our forecast, not yet. Another quarter probably will be more appropriate to answer your question.
Charlie Chan — Analyst, Morgan Stanley
Sure. Thanks for your comments. Very helpful. Thanks.
Jeff Su — Director of Investor Relations, TSMC
Thank you, Charlie. Okay. We will move on to the right side of the room for us. Bruce Lu from Goldman Sachs.
Bruce Lu — Analyst, Goldman Sachs
Thank you for taking my question. I think Charlie already asked the profitability question already. I will just move on to your N2 REM.
What's the revenue contribution we can expect for the N2 ramp for next year? I'm a little bit surprised to hear that the N2 ramp is a similar rate with N3. With N2, you do have both HPC and smartphone customers ramping up at the same stage or in the first year, one or year two, right? Can we expect 15% revenue contribution coming from N2 for next year or a similar level compared to N3, which is around 10%-11% for the year two?
Jeff Su — Director of Investor Relations, TSMC
Okay. Bruce's first question is around the N2 ramp. His question is about the ramp and the ramp profile because we said the N2 ramp profile is similar to N3. What does that mean? Also, of course, what revenue contribution do we expect or can we share for N2 in 2026?
CC Wei — Chairman and CEO, TSMC
Bruce, you have a good argument.
Usually, we ramp up a new node using the smartphone. You knew that. Everybody knew it. Now, it's not only a smartphone, but also HPC product. However, the ramping profile I just reported is similar to 3-nanometer. It's limited by our capability to build a new fab to ramp it up. Also, a little bit, straightforward, it's constrained by the capacity. We say the ramp profile is similar to N3, but the revenue contribution certainly will be bigger because you don't expect our N2 is the same price as our N3, right?
Bruce Lu — Analyst, Goldman Sachs
Of course.
CC Wei — Chairman and CEO, TSMC
Good. Thank you.
Bruce Lu — Analyst, Goldman Sachs
If that is the case, we should assume that in 2027, the N2 ramp will be faster, right? Because you take 12-18 months for you to build new fab. You should be able to achieve even higher growth in N2 in 2027, right?
Jeff Su — Director of Investor Relations, TSMC
Bruce is asking if the revenue contribution is higher in 2026, then should it be even greater in 2027?
CC Wei — Chairman and CEO, TSMC
We will answer that question in 2026. Thank you.
Bruce Lu — Analyst, Goldman Sachs
The next question is for N5 and N3, right? I want to understand the supply demand for N5 and N3 in the coming two years. Most of the AI will migrate to N3 next year. It seems to me that the N5 conversion is mostly done. N5, N3 conversion is mostly done. We do not really see greenfield capacity expansion from N3. It becomes super tightness for the N3 for the coming years. Does that mean that N5 will be lower utilized, or we try to build more N3 in the future, or what kind of, or we should see we can sell out more value for N3 and N5 next year?
Jeff Su — Director of Investor Relations, TSMC
Bruce's second question is around N5 and N3. He wants to know what is the outlook, the supply demand at these two advanced nodes in the coming two years. His observation, AI products will migrate to N3, and the N5, N3 conversion is mostly done. His question is, will 3-nanometer supply be very tight the next few years? I think the last part, therefore, can we earn our value or price for that tightness? On the flip side, what about 5-nanometer? Will it become lower utilization?
CC Wei — Chairman and CEO, TSMC
I like your comment on that we have to share our value because of very tight in N3 capacity. It will be continued for a couple of years, very tight. In fact, N5 also very tight. The demand is high because of a lot of AI products still in the 4-nanometer technology node.
They will transition to 3-nanometer probably in the next two years. In the meanwhile, N5 is still very tight in capacity. N3, even tighter. We are working hard. One of TSMC's advantages is that we have a gigafab cluster. We have between N7, N5, N3, even the future N2, we have almost for each node about 85%-90% common tools. It is not free, but it is much easier for TSMC to adjust or convert the capacity between those nodes. Today, let me share with you, we are using the N7 capacity to support N5 because N5 is too tight. We are converting N5 to N3, as you just pointed out. We will continue to do that. Today, our leading-edge technology is a capacity. We define N7 and below are all very tight.
Seeing that, we are working very hard to, again, using my sentence, narrow the gap between the demand and the capacity.
Jeff Su — Director of Investor Relations, TSMC
Okay. Thank you, Bruce. Let's go to the participants online. Maybe we'll take two questions from the online, and then we'll come back to the floor. Thank you. Operator?
Brett Simpson — Analyst, Arete Research
Yes. Now asking question, Brett Simpson, Arete.
Yeah. Thanks very much. I had a question for Wendell on gross margins. And it's always helpful. You've laid out a framework for some of the puts and takes to TSMC's gross margins. But my question is really, some of these headwinds like FX and the dilution from overseas fabs are more structural cost increases. To what extent can TSMC adjust wafer pricing higher to neutralize these cost increases in your business?
I guess, secondly, on this point, how much economic benefits are you seeing from applying AI across the fabs? I mean, I think NVIDIA has mentioned that they're working with TSMC closely strategically in areas like computational lithography to try to drive further fab efficiencies. Can you maybe just give us some examples where you're seeing real gains in your cost structure? Are we at a point where you're starting to see several points of gross margin benefit from AI efficiencies? Thanks.
Jeff Su — Director of Investor Relations, TSMC
All right, Brett. Brett's first question is a little bit involved, but looking at our gross margin and profitability, he notes that the unfavorable exchange rate and the dilution from the overseas due to the higher costs, these are structural headwinds.
His question is, how can we or can we earn our value or adjust our wafer price to help offset some of these? Also, how much we've talked about before about using AI ourselves in our operations, how much economic benefit are we deriving from things such as CU Lethal with our customers? Are there other examples of using AI where it's helping our cost structure? Can we quantify that? Quantify the benefit, sorry. That's your question, right? Brett?
Brett Simpson — Analyst, Arete Research
That's right. Thanks, Jeff.
Wendell Huang — SVP and CFO, TSMC
Okay, Brett. The first question, gross margin. That's the reason why we've been talking about these six factors affecting our profitability. I don't think I need to repeat those six factors. Whenever, for example, using foreign exchange rate as an example, a few years ago, there were also periods of time that foreign exchange rates were against us.
We are able to lean on the other factors to help us mitigate the negative impact from certain factors and therefore still achieve our gross margin targets. You specifically asked about ASP, raising the price. The price is just one of the factors. I believe CC just elaborated a lot on earning our value. At the same time, there are other factors that we can leverage on. All in all, that is why we are saying 53% and higher, and higher gross margin is still achievable. Your second question on AI benefits, I think we also talked about that before. We use that in operation, in manufacturing. We also use that in R&D. Just think about if we are able to produce 1% of productivity gains in a company of our size, that equals to $1 billion.
That's the number we can share with you without going into too much other details.
Jeff Su — Director of Investor Relations, TSMC
Does that answer your question, Brett?
Brett Simpson — Analyst, Arete Research
Okay. Yeah. Yeah, that's great. Thanks very much, Wendell. I guess my follow-up question, just digesting your prepared remarks, CC, you mentioned 11 fabs in Taiwan. I think I count eight fabs for overseas that you're planning that aren't commercially online yet. Can you maybe just talk a bit about, I mean, I've never seen that type of construct, that type of roadmap before from TSMC. It's quite big. Can you maybe share with us if you're planning a bigger expansion of new capacity next year? I say this because in the last few months, we've seen so many gigawatts at data center announcements. I think this week we had one from Meta that was significant. The demand looks very strong.
I'm just wondering whether you have enough capacity to satisfy demand next year, whether you plan to convert further 5-nanometer to 3-nanometer, and how you see the N2 capacity plans for 2026. Thank you.
Jeff Su — Director of Investor Relations, TSMC
Okay. Brett, second question, he notes that we are building many fabs both in Taiwan and also overseas. He's never seen this size or scale of capacity expansion from TSMC before. It is very—and he also notes that the demand from data centers continues to be very strong. His question is basically very simply, do we have enough capacity to support the strong demand specific to next year and also very specifically to 2-nanometer? And will we further also convert more 5-nanometer to 3? That's, I think, all of his question.
CC Wei — Chairman and CEO, TSMC
Brett?
Brett Simpson — Analyst, Arete Research
Thank you.
CC Wei — Chairman and CEO, TSMC
Your observation is right.
Recently, we saw a lot of announcements of the AI data center all over the world. The demand on 3-nanometer, actually on 5-nanometer, 3-nanometer, and the future 2-nanometer are very high. We did not see this kind of strong demand for a long time. Will it be enough to support them? I still want to use my word, say that we try very hard to narrow the gap between the demand and the supply. We're working very hard.
Jeff Su — Director of Investor Relations, TSMC
Okay. Thank you, Brett. Let's go to the next participant on the call.
Arthur Lai — Analyst, Macquarie
Hi. Thank you, CC, Wendell and Jeff, as always on Macquarie. Again, congrats on a strong result. I would like to follow up on the N2.
I think, as CC highlighted, this is a very exciting node we all heard. I want to follow up on the return on investment. Can you compare the N2 and N3, the return on investment, and then give us more color? The reason we ask this question is because the CapEx per area, actually, N2 is higher, right? We also heard from industry that the company's yield on the N2 is also pretty good. Can you give us some put and take on how we think of the N2's future development? Thank you.
Jeff Su — Director of Investor Relations, TSMC
Okay. Arthur's question, both questions, I think, are around N2. N2, as he said, is a very exciting node. He wondered, would like to know, understand what is the return or the return on investment that we see from N2 compared to N3.
Can we talk a little bit into the CapEx is higher, but the yield is still very good? What is the developments that we're seeing for 2 nanometer? I think that is that your question, Arthur?
Arthur Lai — Analyst, Macquarie
Yes, yes. Yeah. Exactly.
Wendell Huang — SVP and CFO, TSMC
Okay, Arthur. N2, return. As we said before, N2, the profitability is better than N3. Now, there were questions asking how many quarters to catch up with the corporate before. And N3 was it took N3 longer. For N2, we think it will be back to the old days. Having said that, I need to remind everyone that in the old days, we're talking about corporate average of, say, 50% gross margin. Nowadays, we're talking about 53% gross margin. It becomes less meaningful to talk about how much time it takes to catch up with corporate nowadays.
Having said that, structurally, N2 does have a better profitability than N3. Okay. N2 development is right on track. We're ramping it. In the second half of this year, we expect the revenue to come up in the first half of next year. Okay.
Jeff Su — Director of Investor Relations, TSMC
Thank you, Wendell. Thank you, Arthur. Let's come back to the floor. We'll go left, middle, right. Maybe Sunny Lin from UB.
Sunny Lin — Analyst, UBS
Good afternoon. Thank you for taking my question. Very good results and congrats. My first question is to follow up on CapEx. Obviously, full year sales guidance is stronger. You are turning more constructive on high-performance compute and AI. Yet you are keeping your CapEx guidance. Is it fair to assume that you are considering some conservatism for CapEx for this year given the ongoing macro uncertainty?
Is it because in the short term, your capacity expansion is somewhat constrained by the ability that you can ramp up more capacity? Therefore, maybe in 2026 and 2027, we should expect some accelerations of your CapEx spending.
Jeff Su — Director of Investor Relations, TSMC
Okay. Sunny's first question is around CapEx. She notes, of course, that we raised our 2025 revenue guidance this year, and we certainly still see a very robust demand from AI, yet we kept our CapEx guidance in the same range of $38 billion-$42 billion. She wants to understand why. Is it because of macro uncertainty? Is it because of constraints in the construction? Her other part of this question is, what is the CapEx outlook for 2026 and 2027, I guess?
Wendell Huang — SVP and CFO, TSMC
Okay. Sunny, the CapEx, as we said before, the CapEx invested in a given year is for the business opportunities in the following years.
As long as there are business opportunities, we will not hesitate to invest. Having said that, nowadays, as CC also said, with all these macro uncertainties, we are mindful of these uncertainties. We also take that into consideration in our capacity and CapEx plan. Going forward, it's too early to talk about future years' CapEx, but I can share with you, a company of our size, it's unlikely that you see CapEx dollar amount suddenly drop a lot in any given years. That's all I can share with you.
Sunny Lin — Analyst, UBS
Got it. Sounds like CapEx could be going higher in the coming years. My second question is on cloud AI. Since, like earlier, you attribute most of the sales upside for 2025 to cloud AI.
Therefore, I wonder if you have an update on the cloud AI growth in 2025, which you guided before to be about 100% for 2025. The implication to your CoWoS capacity expansion, would you be, are you able to maybe expand a bit more CoWoS capacity for this year to support a stronger cloud AI growth for this year? Any early insight that you could share with us for your CoWoS capacity expansion for 2026? Thank you.
Jeff Su — Director of Investor Relations, TSMC
Okay. Sunny's second question is asking about our AI, she says cloud AI, basically our AI accelerator growth in 2025 and related to the CoWoS capacity. Her question is, what is the AI accelerator revenue growth we expect in 2025? What is our CoWoS capacity expansion plan for 2025? She asked a similar question to Charlie or someone earlier.
What is the plan for CoWoS capacity in 2026?
CC Wei — Chairman and CEO, TSMC
My answer stays the same. We are trying very hard to narrow the gap for now and for 2026. The demand, the momentum are very healthy and very strong. We are building many new facilities in the backend to increase the AI, increase the CoWoS capacity to support our customer. AI demand is very strong. The CoWoS capacity, the demand is very strong.
Jeff Su — Director of Investor Relations, TSMC
Thank you, Sunny. We will move on to Laura Chen from Citi.
Laura Chen — Analyst, Citi
Thank you for taking my question and congrats for the good result and outlook. My question is also about the AI chip. CC, you mentioned that the AI chip is getting bigger and bigger, and also the power consumption is getting much more.
I'm just wondering that among your advanced technology, including the advanced node, we also noted that during the symposium, TSMC also announced some of the new technology in advanced packaging as well. I'm just wondering, how do you kind of prioritize your leading-edge advanced packaging? During the symposium, we see that CoWoS now wafer, that kind of a new design. Do you have any plan or timeline for the new technology? Should we think about that, that should be kind of aligned with our most advanced node process, like N2 or A16 going forward?
Jeff Su — Director of Investor Relations, TSMC
Okay. Laura's first question is on advanced packaging. Notes AI, the die sizes are increasing. The need for power consumption or energy efficiency is rising. She wants to understand how our strategy for advanced packaging along with the advanced node development. Are there any specific packaging solutions that we're prioritizing?
What about the timeline and roadmap? How does that match up with our advanced node roadmap?
CC Wei — Chairman and CEO, TSMC
Laura, I think TSMC's philosophy to develop a technology is working with customer. The customer has such demand. We develop the technology. We increase the capacity for them. Priorities, every customer is important to TSMC. In the advanced packaging side, a lot of customers are using the different approaches. We are developing a variety of different backend packaging advanced technology for all the customers. Whether it's related to the advanced leading-edge technology, the answer is yes. We have a system integration. We have CoWoS, that's a terminology. We have a lot of different names that I cannot even remember, but it's a lot of varieties, and we work with our customer to meet their demand. That I can answer you.
Laura Chen — Analyst, Citi
Is that easy to kind of leverage or transfer different kind of technology from your perspective?
Jeff Su — Director of Investor Relations, TSMC
Laura's asking how fungible are these different packaging technologies? How interchangeable or easy to transfer the technology between different packaging solutions?
CC Wei — Chairman and CEO, TSMC
Of course, there are some similarities in between. Otherwise, we are going to take too much of the effort and then did not get the return. Yes, there's a lot of similarities, but a lot of varieties also.
Laura Chen — Analyst, Citi
Okay. Thank you. My second question is, we know that obviously the AI demand, advanced node, advanced leading-edge packaging is very tight. I'm just wondering that industry-wise, we still see probably overcapacity in mature node. Yet TSMC also has more mature 16 or nanometer or more above that kind of process.
Can we kind of consolidate our mature nodes to kind of make better efficiency and probabilities to enhance those capacities to fulfill the demands across the board?
Jeff Su — Director of Investor Relations, TSMC
Laura's second question is on mature nodes. She notes there is overcapacity industry-wide in older nodes. She wants to understand for TSMC specifically, if we take, for example, 16-nanometer and older nodes, what is our strategy? Can we consolidate amongst the different nodes? How do we protect our profitability?
CC Wei — Chairman and CEO, TSMC
Good question. If you read the newspaper, there is so much mature node capacity. TSMC's strategy actually is on the mature node technologies. We develop kind of specialties. For example, that is RF technology or CMOS image sensor or the high voltage. We develop the technology at the request of our customer. We do not worry too much about what you say, the overcapacity.
If it is really overcapacity, we will not build a fab in Japan. We will not build a fab in Germany. It is not overcapacity. It is all related to customer's need, customer's demand, and those are all specialty technologies. Did I answer your question?
Laura Chen — Analyst, Citi
Thank you.
Jeff Su — Director of Investor Relations, TSMC
Thank you, Laura. I think in the interest of time, we will go to Brad Lin from Bank of America. We will take one more from the line. If there is one more from the floor.
Brad Lin — Analyst, Bank of America
Thank you for taking my question. I have two questions. My first one is on the human robot. We have learned that human robot started to contribute to TSMC, and it is gaining momentum as the next frontier of the AI hardware.
How does TSMC evaluate the market size of human robot in the semiconductor and in terms of the potential market compute and also sensor requirements? Thank you. Do you think that might be another driver potentially for mature nodes too? Thank you.
Jeff Su — Director of Investor Relations, TSMC
Thank you, Brad. Brad's first question is around humanoid robots. We're starting to see some contribution. He wants to understand how do we evaluate the market size, what is the addressable opportunities for TSMC in the long term at the leading edge, and also on the mature nodes with certain type of specialty.
CC Wei — Chairman and CEO, TSMC
Brad, it's too early. Actually, it's too early to say the humanoid robot will play a role in this year. Next year, probably still too early because it's so complicated. You know that the humanoid robot will be most of the time will be used.
I think the first one will be used in a medical industry to take care of the people getting old like me. Probably someday later, I need some humanoid robot to help me. It's very complicated because we are talking about the brain only. Actually, you're talking about a lot of sensor technology. That's the image sensor, the pressure sensor, the temperature sensor, and all the feedback to the CPU. It's very complicated. Since it's dealing with human being directly, it has to be very, very careful. Once you start to fly, you are a big, big plus. I talked to one of my customers, and he said that the EV car is nothing. His robot will be 10 times of that. I'm waiting for that.
Okay. Did that answer your question?
Brad Lin — Analyst, Bank of America
Yes. Yes.
I believe the client definitely owns EV cars and robots too, so he knows me well. My second question would be on the potential point ahead of the so-called reflecting the value into 2026. We know, normally we continue to reflect the value into our pricing. Given the potentially higher pricing into 2026, are you observing any signs of demand pooling from the customers in the second half of the year? Potentially, given the tight pipeline of N3, N5, would we see a continuous strength into FOCU, even though we already guided potential decline? Any point potentially. Thank you.
Jeff Su — Director of Investor Relations, TSMC
Okay. Brad, second question, very specifically. He is asking, as CC talked about, that we will continue to earn our value. Do we see any customers trying to pull in their demand ahead of 2026 into the second half of this year?
Do we have any additional comments to offer on the fourth quarter besides what we have already shared?
CC Wei — Chairman and CEO, TSMC
The answer is no. We did not see any different customers' behavior so far. Okay. Let me share with you in more color. If you are talking about the 3-nanometer demand, for example, the cycle time is still what takes about four months. There is no way you can pull in anything. I mean, that is a—yeah. We have, as I said, our capacity is very, very tight. We already have all the schedules. There is very little room for pulling, let me say that, even if they wanted. No, the answer is no. 2026 is 2026. We will share with you.
Brad Lin — Analyst, Bank of America
Yeah. Thank you very much.
Jeff Su — Director of Investor Relations, TSMC
Thank you, CC. Thank you, Brad. All right.
Operator, let's go to the—we'll take questions from the last participant on the line.
Mehdi Hosseini — Analyst, SIG
Thanks for taking my question. The first one has to do with capital intensity. When I look at the past five years when you were ramping N3 and N5, the capital intensity was at or above 40%. And you also highlighted how N2 takeout is tracking better than N3 and N5 combined. Does that imply that the capital intensity would need to go back up to 40%? In other words, an initial investment for N2 to accommodate these takeouts? Is that the driver of thinking about how investment drives are going to play out and how to follow up?
Jeff Su — Director of Investor Relations, TSMC
Okay, Mehdi. Sorry, we couldn't hear you exactly clearly. Let me try to summarize your question, all right? Mehdi's question is around capital intensity.
He notes that in the past, when we invest in new nodes or structure the megatrends like we have with N3 and N5, our capital intensity has jumped up to greater than 40%. If I heard you correctly, Mehdi, your question is this time we talked about the strong demand for 2-nanometer multi-year upcoming, what is our expectation on capital intensity? Is that correct, Mehdi?
Mehdi Hosseini — Analyst, SIG
Yes, that's correct.
Wendell Huang — SVP and CFO, TSMC
Okay. Mehdi, let me answer this question. As we just said, the capital expenditure invested in any year is the future growth opportunities. If we do our job right, the growth in the next few years is likely to exceed the growth in CapEx dollars, even though, as I said, the CapEx dollars is unlikely to drop significantly in every given year.
You see a higher growth in revenue than the growth in capital expenditure, then you do not have such a high capital intensity. We actually demonstrated that in the past few years. Also, let me just share with you that because of this, we are not operating setting a capital intensity as a goal. The dollar amount invested is really on the structure demand growth in the following years. Talking about capital intensity is also less meaningful than before. Thank you.
Mehdi Hosseini — Analyst, SIG
Thank you. A follow-up for me, you highlighted N2P. I am sorry. You highlighted N16, which would be very applicable for high-performance compute. Is that the node where AI and HPC would actually be at power with a smartphone as an end market that would drive demand for the most leading-edge node?
Jeff Su — Director of Investor Relations, TSMC
Sorry, Mehdi. Again, I apologize.
I could not hear you clearly, but I think his question is about on A16, where we said it's more for specific HPC-related offerings. His question is, I think, Mehdi, your question is, is that where the AI demand also comes in for the two-nanometer family?
Mehdi Hosseini — Analyst, SIG
Yes. I answered because so far, AI has been N+1, N+2. Is that node A16 the first node where AI would move to the leading edge?
Jeff Su — Director of Investor Relations, TSMC
Okay. Maybe let me rephrase it. I think I understand better. His question is really about AI adoption of the leading-edge node, the end node. We see smartphone. We see HPC. His question very specifically, how do we see the AI adoption of the most leading node for TSMC? He observed in the past, it has generally been one node behind. How do we see that going forward with things such as A16?
CC Wei — Chairman and CEO, TSMC
Mehdi, you are right.
Mehdi Hosseini — Analyst, SIG
Thank you, Jeff.
CC Wei — Chairman and CEO, TSMC
Usually, the HPC customer is always one step behind using N+1 or N+2 technologies. Now, because of AI demand, it is so strong. That is one thing. The most important thing is they need some kind of performance, but the power consumption is very, very important. When we talk about A16, we have another power efficiency improvement close to 20%. That is a big value for all the AI data centers applications. That helped my customer moving faster because every time when we talk about the AI data center, if you notice that the first thing they talk about is power supply, electricity, right? They did not tell you, say, the power efficiency is very important, but they tell you that we had to build a very big electricity power plant to support the AI data centers.
That tells you how important it is. And TSMC's technology, by the way, on A16 is a further improvement of the N2 node. It is not a surprise for TSMC to expect for those people in AI data centers industry, they want to use in A16. Okay.
Jeff Su — Director of Investor Relations, TSMC
Okay. Thank you, CC. Thank you, Mehdi. We'll take the last question from the floor. We have one participant here, Felix Pan from KGI.
Felix Pan — EVP, KGI
Hi. Thanks, Jeff. Good afternoon, CC and Wendell. I only have one question about the overseas expansion. I think CC earlier mentioned that the second fab for the N3, there is a strong demand. You guys need to speed up several quarters for that. Together with, I think, the US government also raised the investment tax credit cap for next year.
I wonder how this shape or how this speed up your ramping schedule for the second fab and how this impact to the overseas fab dilution for the guidance windows given earlier. Yeah. Sorry. I think the follow-up question will be, if you guys speed up the US investment, how is that impact to other regional investment like Japan and Germany as well? Lastly, is that possible to break down the overseas CapEx and domestic CapEx going forward? Yeah. That's all my questions. Thanks.
Jeff Su — Director of Investor Relations, TSMC
Okay. Yeah. That's pretty much two questions. Okay. Felix's question on our overseas expansion plans, he notes that, yes, CC said we're speeding up the schedule for the second fab in the US. He also notes the recent passage of the US ITC bill. How does this impact or affect our ramp schedules in our US expansion?
What is the implication or impact to the overseas dilution? That's number one.
Yeah.
CC Wei — Chairman and CEO, TSMC
Wow. That's two questions. Okay. Let me share with you about our ramp-up schedule. It is totally because of our customer's demand. We appreciate the U.S. government increased the ITC from 25%-35%. We appreciate that. It helped. The real schedule is because of our customer's demand. We have to prepare the capacity to meet their demand. That is the number one consideration. Margin? Yeah.
Wendell Huang — SVP and CFO, TSMC
The margin impact, it is positive, although not that significant in the five-year period. Think about this. The ITC is used to offset the asset value. The benefit comes when depreciation starts. It gets amortized.
Felix Pan — EVP, KGI
Okay.
Jeff Su — Director of Investor Relations, TSMC
Felix's second question is, how does the ramp and speed up of the US cluster expansion, how is this impacting our expansion plans in Japan and Europe if it does at all?
CC Wei — Chairman and CEO, TSMC
You think about the TSMC expansion, the overseas fab. In the US, it is leading edge. In Japan, it is on specialty technology. To be specific, most of the time, it is for the CMOS Image Sensor. For Germany, it is automotive industry. They are all not in the same field. Actually, it does not affect the investment in the US or investing on the leading edge does not affect the investment in Japan or in Germany.
Jeff Su — Director of Investor Relations, TSMC
Thank you, CC. Thank you, Felix. Okay, everyone. This concludes 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. Both are going to be available through TSMC's website at www.tsmc.com. Thank you very much for joining us today. We hope everyone continues to stay well. We hope you will join us again next quarter. Goodbye and have a good day. Thank you.