Taiwan Semiconductor Manufacturing Co Ltd Q2 FY2023 Earnings Call

Good afternoon, everyone, and welcome to TSMC's Second Quarter 2023 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 Vice President and CFO, Mr. Wendell Huang, will summarize our operations in the second quarter of 2023, followed by our guidance for the third quarter of 2023. Afterwards, Mr. Huang, TSMC's CEO, Dr. C. C. Wei; and TSMC's Chairman, Dr. Mark Liu, will jointly provide the company's key messages. Then we will open the line for a question-and-answer session. As usual, I would like to remind everybody that today's discussions may contain forward-looking statements that are subject to significant risks and uncertainties, which could cause 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. And now I would like to turn the call over to TSMC's CFO, Mr. Wendell Huang, for the summary of operations and the current quarter guidance.

Thank you, Jeff. Good afternoon, everyone, and thank you for joining us today. My presentation will start with financial highlights for the second quarter of 2023. After that, I will provide the guidance for the third quarter. Second quarter revenue decreased 5.5% sequentially in NT or 6.2% in U.S. dollars as our second quarter business was impacted by overall global economic conditions, which dampened end market demand and led to customers' ongoing inventory adjustments. Gross margin decreased 2.2 percentage points sequentially to 54.1%, mainly reflecting lower capacity utilization and higher electricity costs, partially offset by more stringent cost control and a more favorable foreign exchange rate. Despite the industry's cyclical downturn, we continue to invest in R&D to support our N3 and N2 development. Thus, operating margin was 42%, down 3.5 percentage points sequentially. Overall, our first quarter EPS was NT7.01 and ROE was 23.2%. Now let's move on to revenue by technology. The 5-nanometer process technology contributed 30% of our wafer revenue in the second quarter, while the 7-nanometer accounted for 23%. Advanced technologies, defined as 7-nanometer and below, accounted for 53% of wafer revenue. Moving on to revenue contribution by platform. HPC decreased 5% quarter-over-quarter to account for 44% of our second quarter revenue. Smartphone decreased 9% to account for 33%. IoT decreased 11% to account for 8%. Automotive increased 3% to account for 8%, and DCE increased 25% to account for 3%. Moving on to the balance sheet. We ended the second quarter with cash and marketable securities of NT1.5 trillion or USD 48 billion. On the liability side, current liabilities decreased by NT62 billion, mainly due to the net decrease of NT87 billion in income tax payable as we pay NT120 billion for 2022 income tax, offset by NT33 billion accrued tax payables for the second quarter. Long-term interest-bearing debt increased by NT53 billion mainly as we raised NT41 billion in corporate bonds. On financial ratios, accounts receivable turnover days decreased 2 days to 32 days, while days of inventory increased 3 days to 99 days, primarily due to entry ramp during the quarter. Regarding cash flow and CapEx. During the second quarter, we generated about NT167 billion in cash from operations, spent NT251 billion in CapEx, distributed NT71 billion for the third quarter 2022 cash dividend, and raised NT41 billion from corporate bond issuances. Overall, our cash balance decreased by NT109 billion to NT1.3 trillion at the end of the quarter. Free cash flow was negative NT83 billion during the quarter as operating cash flow was more than offset by capital expenditures, partly due to the income tax payment of NT120 billion. In U.S. dollar terms, our second quarter capital expenditures totaled NT8.17 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 third quarter revenue to be between NT16.7 billion and USD 17.5 billion, which represents a 9.1% sequential increase at the midpoint. Based on the exchange rate assumption of USD 1 to NT30.8, gross margin is expected to be between 51.5% and 53.5%, operating margin to be between 38% and 40%. This concludes my financial presentation. Now let me turn to our key messages. I will start by making some comments on our second quarter ’23 and third quarter ’23 profitability. Compared to the first quarter, our second quarter gross margin decreased by 220 basis points sequentially to 54.1%, primarily due to lower capacity utilization. Compared to our second quarter guidance, our actual gross margin slightly exceeded the high end of the range provided 3 months ago, mainly due to more stringent cost control efforts and a slightly more favorable foreign exchange rate. We have just guided our third quarter gross margin to decline by 1.6 percentage points to 52.5% at the midpoint, primarily as the higher level of capacity utilization rate is offset by 2 to 3 percentage points margin dilution from the initial ramp-up of our 3-nanometer technology. Looking ahead to the fourth quarter, we expect the continued steep ramp-up of our 3-nanometer to dilute our fourth quarter gross margin by about 3 to 4 percentage points. In 2023, our gross margin faces challenges from lower capacity utilization due to semiconductor cyclicality, the ramp-up of N3, overseas fab expansion, and inflationary costs, including higher utility costs in Taiwan. To manage our profitability in 2023, we will work diligently on internal cost improvement efforts while continuing to sell our value. While we face near-term challenges, we continue to forecast a long-term gross margin of 53% and higher is achievable. Next, let me talk about our 2023 capital budget and depreciation. Every year, our CapEx is spent in anticipation of the growth that will follow in future years. Given the near-term uncertainties, we continue to manage our business prudently and tighten up our capital spending where appropriate. We now expect our 2023 capital budget to be towards the lower end of our range of between USD 32 billion and USD 36 billion. Our depreciation expense is now expected to increase by mid-20s percent year-over-year in 2023, mainly as we ramp our 3-nanometer technologies. Despite near-term inventory cycles, our commitment to support customers' structural growth remains unchanged, and our disciplined CapEx and capacity planning remains based on the long-term market demand profile. We will continue to work closely with our customers to plan our long-term capacity and invest in leading-edge specialty and advanced packaging technologies to support their growth while delivering profitable growth to our shareholders. Now let me make a few comments on our cash dividend distribution policy. The objectives of TSMC's capital management are to fund the company's growth organically, generate good profitability, preserve financial flexibility, and distribute a sustainable and steadily increasing cash dividend to shareholders. As a result of our rigorous capital management, in May, TSMC Board of Directors approved the distribution of NT3 per share cash dividend for the first quarter of 2023, up from NT2.75 previously. This will become the new minimum quarterly dividend level going forward. The first quarter ’23 cash dividend will be distributed in October 2023. For 2023, TSMC shareholders will receive a total of NT11.25 per share dividend and at least NT12 per share cash dividend for 2024. Going forward, as our capital intensity begins to decline in the next several years, the focus of our cash dividend policy is expected to shift from a sustainable to a steadily increasing cash dividend per share in the next few years. Now let me turn the microphone over to C. C. Wei.

Thank you, Wendell. Good afternoon, everyone. First, let me start with our near-term demand and inventory. We concluded our second quarter with revenue of USD 15.7 billion in line with our guidance in U.S. dollar terms. Our business in the second quarter was impacted by the overall global economic conditions, which dampened end market demand and customers' ongoing inventory adjustment. Moving into third quarter 2023. While we have recently observed an increase in AI-related demand, it is not enough to offset the overall cyclicality of our business. We expect our business in the third quarter to be supported by the strong ramp of our 3-nanometer technologies, partially offset by customers continued inventory adjustment. In the last quarterly conference, we said we expect fabless semiconductor inventory to rebalance to a healthier level exiting the third quarter. This statement continues to hold true. However, due to persistent weaker overall macroeconomic conditions, slower-than-expected demand recovery in China, and overall softer end market demand conditions, customers are more cautious and intend to further control their inventory into 4Q '23. Thus, while we maintain our forecast for the 2023 semiconductor market, excluding memory, to decline mid-single-digit year-over-year, we now expect the foundry industry to decline mid-teens and our full year 2023 revenue to decline around 10% in U.S. dollar terms. With such inventory control, we also forecast the fabless semiconductor inventory to exit 4Q '23 at a healthier and lower level compared to our expectation 3 months ago. Next, let me talk about the HPC and TSMC's long-term growth outlook. As we have said before, the massive structural increase in demand for computation underpinned by the industry megatrend of 5G and HPC continues to drive a greater need for performance and energy-efficient computing, which require use of leading-edge technologies. These megatrends are expected to fill TSMC's long-term growth. Even with a more challenging 2023, our revenue remains well on track to grow between 15% and 20% CAGR over the next several years in U.S. dollar terms, which is a target we communicated back in January 2022 Investor Conference. The recent increase in AI-related demand is directionally positive for TSMC. Generative AI requires higher computing power and interconnect bandwidth, which drive increasing semiconductor content. With using CPUs, GPUs, or AI accelerators and related ASIC for AI and machine learning, the commonality is that it requires leading-edge technology and a strong foundry design ecosystem. These are all TSMC's strengths. Today, server AI processor demand, which we define as CPUs, GPUs, and AI accelerators that are performing training functions, accounts for approximately 6% of TSMC's total revenue. We forecast this to grow at close to 50% CAGR in the next 5 years and increase to low teens percent of our revenue. The sensible need for energy-efficient computation is starting from data centers, and we expect it to proliferate to edge and devices over time, which will further drive long-term opportunities. We have already embedded a certain assumption for AI demand into our long-term CapEx and growth forecast. Our HPC platform is expected to be the main engine and the largest incremental contributor to TSMC's long-term growth in the next several years. While the quantification of the total addressable opportunity is still ongoing, generative AI and large language models only reinforce the already strong conviction we have in the structural mega trend to drive TSMC's long-term growth, and we will closely monitor the developments for further potential upside. Now let me talk about our N3 and N3E status. Our 3-nanometer technology is the most advanced semiconductor technology in both performance, power, and transistor technology. N3 is already involved in production with good yield. We are seeing robust demand for N3, and we expect a strong ramp of N3 in the second half of this year, supported by both HPC and smartphone applications. N3 is expected to continue to contribute mid-single-digit percentage to our total wafer revenue in 2023. N3E further extends our N3 family with enhanced performance, power, and yield and provides complete platform support for both HPC and smartphone applications. N3E has passed the qualification and achieved performance and yield targets and will start volume production in the fourth quarter of this year. With our continuous enhancement of 3-nanometer process technologies, we expect strong multi-year demand from our customers and are confident that our 3-nanometer family will be another large and long-lasting node for TSMC. Finally, I'll talk about our N2 status. Our N2 technology development is progressing well and on track for volume production in 2025. Our N2 top nanosheet transistor structure will provide our customers with the best performance, cost, and technology maturity. Our nanosheet technology has demonstrated excellent power efficiency, and our N2 will deliver full node performance and power benefits to address the increasing need for energy-efficient computing. As part of the N2 technology platform, we also developed N2 with a backside power rail solution, which is best suited for HPC applications. Backside power rail will provide a 10% to 12% additional speed gain and a 10% to 15% larger density boost on top of the baseline technology. We are targeting the backside power rail to be available in the second half of 2025 to customers with production in 2026. We are observing a high level of customer interest and engagement at N2 from both HPC and smartphone applications. Our 2-nanometer technology will be the most advanced semiconductor technology in the industry in both density and energy efficiency when it is introduced, and to further extend our technology leadership right into the future. This concludes my prepared remarks, and now let me turn the microphone over to Mark.

Thank you, C. C. And good afternoon, everyone. Today, I want to talk about TSMC's global manufacturing footprint, status update. TSMC's mission is to be the trusted technology and capacity provider of the global logic IC industry for years to come. Our strategy is to expand our global manufacturing footprint to increase customer trust and to expand our future growth potential and to reach for more global talents. Our overseas decisions are based on our customers' needs and the necessary level of government support. That is to maximize the value of our shareholders and to fulfill our fiduciary duty. In Arizona, we are building our first fab to provide the U.S. most advanced semiconductor technology in mass production to support the needs for U.S. semiconductor infrastructure. Our fab in Arizona started construction in April 2021 with an aggressive schedule. We are now entering a critical phase of handling and installing the most advanced and dedicated equipment. However, we are encountering certain challenges, as there is an insufficient amount of skilled workers with the specialized expertise required for equipment installation in a semiconductor-grade facility. While we are working to improve the situation, including sending experienced technicians from Taiwan to train the local skilled workers for a short period of time, we expect the production schedule of N4 process technology to be pushed out to 2025. In Japan, we are building a specialty technology factory, which will utilize 12, 16, and 22, 28 process technologies. Volume production is on track for late 2024. In Europe, we are engaging with customers and partners to evaluate building a specialty fab in Germany, focusing on automotive-specific technologies based on the demand from our customers and the level of government support. In China, we are expanding 28-nanometer in Nanjing as we planned to support our customer in China, and we continue to follow all rules and regulations fully. At the same time, we continue to invest in Taiwan and to expand our capacity to support our customers' growth. From a cost perspective, the initial cost of overseas fab is higher than TSMC's fabs in Taiwan due to: one, the smaller fab scale; two, higher costs throughout the supply chain; and three, the early stage of semiconductor ecosystem on those overseas sites, as compared to a matured ecosystem in Taiwan. In our recent meetings with senior government officials in the U.S., Japan, and Europe, we discussed our plans to expand our global manufacturing footprint with them. We also emphasized that one of our major responsibilities is to manage and minimize the cost gap to maximize return for our shareholders. Those discussions went very well. All sites understand the critical and integral role TSMC plays in the semiconductor industry, and we appreciate all the government's ongoing support in working with TSMC to help narrow down the cost gap. We will continue to work closely with all the governments to secure further support. Our pricing will also remain strategic to reflect our value, which includes the value of geographic flexibility. At the same time, we will leverage our fundamental competitive advantage of manufacturing technology leadership, large volume, and economies of scale to continuously drive our costs down. By taking such actions, TSMC will have the ability to absorb the higher costs of overseas fab while remaining the most efficient and cost-effective manufacturer no matter where we operate. Thus, even as we expand our capacity overseas, TSMC's long-term gross margin of 53% and higher and sustainable ROE of greater than 25% is achievable, and we will continue to maximize the value for our shareholders. This concludes our key messages. Thank you for your attention.

Thank you, Chairman. This concludes our prepared statements. Before we start the Q&A session, I would like to remind everyone to limit their questions to two at a time to give all participants a chance to ask their questions. If you wish to ask your question in Chinese, I will translate it to English before our management responds. Now let's begin the Q&A session. Operator, can we please proceed with the first caller on the line?

The first one – question, Gokul Hariharan from JPMorgan. Go ahead please.

Yes. Thank you. Good afternoon. And thanks for a lot of clarity on the AI-related exposure. My first question is on the AI front. A lot of TSMC's customers have been talking about capacity shortage and having to kind of queue up for capacity for AI accelerators, including GPUs and ASICs. Could TSMC talk a little bit about what TSMC is doing on the capacity side, especially on the advanced packaging, but also on other areas? And when do you expect to get back to some degree of demand and supply balance for these AI accelerators? Is it going to be only sometime next year? Or do you think it could happen quicker based on what you see on demand from your customers and the capacity plan?

Okay. Gokul, thank you. Let me try to - please allow me to summarize your first question. So first question from Gokul is that he notes that we are - customers are seeing strong demand from AI-related, but they're facing capacity tightness or shortage. So his question is, what are we doing in terms of the capacity side, maybe both in terms of the advanced packaging, as well as the logic? And then when do we see the demand-supply imbalance returning to a better, healthier balance level? Is it sometime next year? Is that correct, roughly, Gokul?

Yes. Right.

Okay. Gokul, this is C. C. Wei. Let me answer your question. For AI, right now, we see a very strong demand, yes. For the Tongan part, we don't have any problem to support. But for the backend, the advanced packaging side, especially for the cohorts, we do have some very tight capacity to - very hard to fulfill 100% of what customers needed. So we are working with customers for the short term to help them to fulfill the demand, but we are increasing our capacity as quickly as possible. And we expect these tightenings will be released in next year, probably towards the end of next year. But in between, we're still working closely with our customers to support their growth.

Okay. And C. C., maybe one follow up. Could you let us show what kind of capacity expansion, like how much capacity you're expanding on the cohort side? Like any kind of quantity, what kind of capacity you are adding?

Okay. So Gokul, just an additional to the first question, how much capacity are we going to increase in terms of cohorts?

Let me give you - I will not give you the exact number, but let me give you a rough probably 2x of the capacity will be added. Okay, Gokul.

Gokul? Okay, Gokul, are you there? If not, operator, maybe we move on to the next participant. Gokul, are you there? Okay. I think there was - disconnected. All right. Let's move on to the next caller, person, please.

Next one to ask question, Bruce Lu from Goldman Sachs. Go ahead please.

Okay. Thank you for taking my question. I still want to know about like, the TSMC maintained the 15% to 20% up in CAGR, when we cut this year's revenue to minus 10%, right. That's all – will you use that 15% revenue CAGR to 2026, that seems slightly above like 25-plus percent present data from the coming 2 years, which means that the overall semi growth is not increased like a lot for the next 2 years. And you just mentioned that the AI only accounts for 6% with low teens potentially. That is not big enough to get back to the trend. So what is the underlying growth you have with the global training in the coming years? And what are the key assumptions for the growth for each segment?

Okay. Bruce, let me try to summarize your first question. So Bruce's first question is on our long-term growth CAGR, which we have said is to be between 15% to 20% from '21 to '26 CAGR period. So Bruce's question, this year, C. C., just said we will decline around 10%. Under his calculation, I think he's saying, well, this implies you should grow 25% the next several years, which, of course, this is a CAGR, but nonetheless. And so Bruce's question is that, therefore, if that's the type of growth, then shouldn't that imply a much higher growth level for the overall semiconductor excluding memory industry. I think that is your question, Bruce. Am I correct?

Yes. What are the key assumptions for this quarter?

Okay. Let me handle this question. The - your rationale is correct. However, some of the factors may not be totally included. For one thing, in your model that the customer's gross margin, 60% plus, I don't think that will represent the average customer's gross margin and maybe some specific ones. However, the other one is the market share. The market share factor, you assumed the constant, that is not - one thing that could be different than in your formula. So the semiconductor growth, right now, we are forecasting 4% to 5%. It may increase, but definitely, as you said, it won't increase to 10%. But those longer-term semiconductor, excluding memory growth is still yet to be evaluated. Did I answer your question?

Yes. But the reason - the reason I do that is I'm assuming that you have like 1 million market share in the advanced node and also the growth is mostly coming from an advanced node, which customers growth lines is supposed to be higher. So I think that the gap is wide enough. That's what I'm wondering right, whether I miss anything, which might be big enough to move the needle that might - investments that the management might - can give us some color?

I don't - this is a factor. As far as the market share value, you might not totally included all the factors. That's my perspective. But I cannot dig into the numerical comparison at this point here. What I mean, the market share is not just the advanced leading-edge technologies, but also the share of the outsourcing.

So maybe, Bruce, if I summarize again, TSMC's growth is driven by both the underlying structural megatrends but also by our technology leadership and differentiation. So our CAGR is a combination of those two factors.

Okay. Thank you.

Do you have any...

Yes. My next question is regarding to the guidance changes. The previous guidance for the full year was long to mid-single digit, now is about like 10% decline. So the gap is like 5% of the total revenue, which is like quite sizable in terms of revenue. With that $3 billion, $4 billion highly concentrated in the second half. Can you give us like what are the changes in terms of this shortfall? Where are the weakness coming from?

Okay. So Bruce's, second question is looking at our 2023 full year guidance. He noticed last time, we had said low to mid-single-digit decline. This time, we have guided to around 10%. So his question is the delta of this seems to be all - a lot of it also in the fourth quarter. So what - is there any particular segment or market that is driving this? And what are the factors behind this? Is that correct, Bruce?

Yes. Thank you.

Okay. Bruce, let me answer the question. Yes, we did see something different. The first, the maker is weaker than what we thought. Three months ago, we probably - probably more optimistic, but now it's not. Also that is - for example, China economy’s recovery is actually also weaker than what we thought. And so the end market demand actually did not grow as we expected. So put all together, even we have a very good AI processor demand, it is still not enough to offset all those kinds of macro impact. So now we expect that the whole year will become minus 10%. That's what we thought.

And in terms of by particular segment or is there a particular market?

It's almost - well, thank you. You are asking me the question. It's almost intact...

I understand my question. Thank you.

Yes. It's overall market segment is being impacted because of - it's a combination of the macroeconomics.

So can we conclude that other than AI, almost every application sees some weakness in the second half?

You got it.

Okay. Thank you, Bruce. Operator, can we move on to the next participant, please?

Next one to ask question is Gokul Hariharan from JP Morgan.

Gokul, you are back. Okay.

Yes, sorry about that. So next question, just wanted to ask about TSMC management's view on the current inventory cycle. It looks like this cycle is taking much longer to get through the down cycle compared to '19 and 2015. When do you think we kind of bottom out? And do you feel that the recovery in next year is going to be a strong recovery? Or you think it's going to be a more gradual recovery? What are the kind of plans that we are putting in place as we think about next year's recovery once the inventory situation lasts? Thank you.

Okay. So Gokul's second question is about the inventory correction cycle. He notes this cycle seems to be taking much longer to get through as compared to 2019 and/or 2015. So his second question is when do we think this cycle can bottom out? What will 2024 next year look like? Do we expect a strong recovery? And what factors are we looking at? Is that correct, Gokul?

Yes. Thanks. Thanks a lot.

Okay. Gokul, this is a good question. Let me answer it in a short one sentence. All about the macro is not so - it's become weaker than we thought. In fact, higher inflation and interest rate impact end demand in all market segments, in every region in the world. As we said, under such situations, our customers are more cautious in their inventory control in the second half of this year. So while we expect the fabless semiconductor industry, their inventory to be cleaner and healthier existing this year, but we're much closer to the seasonal level. But our expectation for them, they will continue to manage their inventory. And 2024, it is still dependent on the macro situation. Gokul?

Okay. So, excuse me, so that - it sounds like you're still expecting at least early part of next year to still be a little bit challenging, similar to what it is looking like right now. Is that fair to say?

Look, if you are...

About next year...

Gokul, we will give you our comment next time for 2024.

Thank you.

Thank you, Gokul. Operator, can we move on to the next participant, please?

Next one, we have Charlie Chan from Morgan Stanley.

Hey, gentlemen. Good afternoon. Thanks for taking my questions. So my first question is about the overseas fab cost seems to get higher. So would TSMC consider to further adjust your pricing to absorb those increased costs? And also management mentioned that you're doubling or more than doubling your advanced packaging, given the AI rush orders. Would that give you a chance to reprice the back-end foundry service? Because I remember there was a kind of below company's gross margin average. Would that be a chance to bring that back to the corporate average? Thank you.

Okay. Charlie. Charlie's first question is, I guess, regarding pricing, two parts or two angles. First on the overseas fab, given that the costs are higher, would TSMC consider to further adjust our wafer price. And also along similar lines related to advanced packaging, given we are – C.C. said doubling roughly the capacity would we consider to also charge more or higher, given that the returns of the back end are lower?

Right. Let me answer the first question first. Yes, the overseas fab will cost higher, at least for the near future, where their supply ecosystem is not mature yet. And the labor cost is from our experience, actually it's a little bit higher than we expected. But to answer your question, yes, as we try to get the maximum government subsidy, and we also really look at how the price value for the overseas geographical flexibilities, these are all considered. The aim is to, one, to increase our customer trust, make them continue to work with us going forward under the geopolitical concerns. Secondly is to maximize shareholders' value. To answer your question on price is strategically yes.

All right...

C.C, can you answer the second question?

Okay. I think the second question is about the pricing or the - on the cohorts. As I answer the question, we are increasing the capacity as soon as possible manner, of course, not including actual costs. So in fact, we are working with our customers. And the most important thing for them right now is supply assurance, is supply to meet their demand. So we are working with them. We do everything possible to increase the capacity. And of course, at the same time, we share our value.

And then...

Thanks. So may I - may ask a second question, it's a different topic. Is that, okay?

Sure.

You get two questions. So, sure.

Thank you. Thanks, Jeff. I have another question regarding the demand for AI SMEs. You've provided useful insights into revenue contribution and growth assumptions, but I'm curious about how TSMC assesses AI demand. Currently, it seems like customers are aggressively booking capacity, so how can the company determine if the AI SME demand is genuine? Additionally, I'm interested in whether the shift towards ASIC solutions by customers is indeed outpacing that of GPUs. I believe the key concern, which investors are particularly interested in, is whether AI is affecting the demand for CPU servers. These are the questions on my mind. Thank you.

Okay. Let me summarize your second question, Charlie. Charges on AI demand. He wants to know how do we judge the demand properly because customers are very aggressive, but how in his words, how do we know that this demand is real. And then also, how do we see the demand specifically for ASIC as it relates to AI?

Okay. You want to answer?

Let me try first and you probably... This is a very deep question. Of course, we have a model, basically. The short-term fancy about the AI demand definitely cannot extrapolate for the long term. Neither can we predict the near future, meaning next year, how the sudden demand will continue or will flatten out. However, our model is based on the data center structure. We assume a certain percentage of the data center processor are AI processors. And based on that, we calculate the AI processor demand. And this model is yet to be fitted to the practical data later on. But in general, I think the - our trend of a big portion of data center processor will be AI processor is a sure thing. And we can cannibalize the data center processors. In the short term, when the CapEx of the cloud service provider are fixed, yes, it will. It is. But as for the long term, when their data service, when the cloud services having the generative AI service revenue, I think they will increase the CapEx, that should be consistent with the long-term AI processor demand. And I mean, the CapEx will increase because of generative AI services. Anything more for you?

Yes, Charlie, I think part of Charlie's question is also how do we see ASIC related in AI development?

Well, actually, the customer also have a high demand on the ASIC part for the AI application. And as Mark pointed out, a short-term sudden increase, you can extrapolate to be long term. But again, let me emphasize that. Those kind of application in AI, be it CPUs, GPUs or AI accelerator or ASIC, they all need leading-edge technologies. And they all have one symptom. They are using the very large die size, which is TSMCs strength.

Got it. Thank you. Thank you very much.

Okay. Thank you, Charlie. Thank you. Operator, we'll move on to the next participant, please.

Next one to ask questions. Randy Abrams, Credit Suisse.

Okay, yes. Thank you. I wanted to shift to the profitability, maybe more for Wendell. For - looking at the fourth quarter, you mentioned the 3 to 4 point dilution from N3. I think that is 2 to 3 points in the third quarter. Is that what you're suggesting, the directional change could be a little bit down margin profile? Or do you have positive offsets that could keep it more stable? And then a follow-up on the margin, where you discussed is a tough year for margins on these factors, like the energy ramp up 3. But could you discuss 2024, do you think we're going into a period of a bit more challenging profitability or you see factors that we could comfortably get back to the 53 and above next year?

Okay. Thank you, Randy. So Randy's first question is on gross margin. Fourth quarter with the N3 dilution of 3% to 4%, does that mean directionally, fourth quarter margin is sequentially down? Are there any positive offsets? And then for looking to 2024 for the full year, if Wendell can give some comments about 2024 gross margin, will it also be challenging? Or do we still feel confident in a 53% and higher gross margin?

Okay, Randy. Starting from the second half of this year, as we said, we faced certain cost challenges, including the ramp up N3 which will dilute about 2 to 3 percentage points in the third quarter and 3 to 4 in the fourth quarter, plus the higher electricity cost. But we're not giving our guidance on the fourth quarter at this moment. We're just spelling out some of the challenges that we're seeing. And of course, we are going to continue to drive down our costs and sell our value to ensure that we will have a good return on the node. That's for this year. For next year, we're seeing - we're not talking about the whole gross margin, but we still see that N3 will dilute about 3 to 4 percentage points of next year's gross margin. And although the yield will be better next year, at the same time, the percentage of revenue contributed by N3 will be bigger. So net-net, worth, we also see some dilution from the N3 next year. But the margin - the guidance will be given out next year.

Okay. A quick follow-up to the first question. I think the last few nodes was 2 to 3 points dilution in the first year or 2 of ramp. The factor for it larger, is it the higher capital intensity or something different versus 3 - versus 5 and 7 or it looks like a little bit more dilution?

Yes. The increasing process complexity does add on to the challenges of a newer node. However, the other important factor is that our corporate averages has become higher than before. We used to have 50% gross margin. We're now talking about 53% and higher gross margin, okay?

Yes. Okay. And the second question, I wanted to ask how you're thinking about CapEx, just netting a few things, the geographic expansion, the 3 and then the start of 2-nanometer, the first ramp up or tool move in versus the mixed outlook you're looking at for macro for a ballpark CapEx into next year? And if I could maybe within it, ask if the Arizona fab delays, does that push out where you mentioned the low end of guidance, push out some of this year to get some lift to next year?

Okay. So Randy's second question is on CapEx. He wants to know basically focusing on 2024 CapEx to some of the delays in the Arizona fab push out CapEx from this year to next year, as we expand overseas, as we invest in N2, but at the same time as the macro remains uncertain, how does this impact 2024 CapEx?

Yes. Randy, the push out of fabs does push out some part of the CapEx, but that doesn't affect a big part. For 2024, it's too early to talk about the overall CapEx. However, our CapEx, if you - as we said before, every year, we spend the CapEx to capture the future growth opportunities. And in the past few years, our CapEx has risen very fast to capture the megatrend. And going forward, the next few years, when we start to harvest those investment, we believe that CapEx will begin to level off in terms of dollar amount and that will lead to - start to lower the capital intensity in the next several years.

Sorry, Randy. Sorry, do... Yes, my quick follow-up. I think you mentioned that as you could use your 5-nanometer to support the ramp up 3. Given the AI and some of that pickup, do you still see that potential that could help optimize CapEx? Or do you need to keep it for existing node? And that's my final one. Thank you.

Yes. So Randy is just also asking then how does tool commonality play a role in our future CapEx?

Yes. We always build two commonality between nodes to provide greater flexibility. We mentioned last time the strong multiyear demand from N3, we are able to support that using some of the tools from N5. We're not going to comment on the CapEx beyond this year. However, as I just mentioned, every year, the CapEx is spent to capture the future growth opportunities.

Thank you, Wendell.

Okay.

Thank you, Randy. Operator, can we move on to the next participant?

Next one to ask questions, Laura Chen from Citi.

Thank you very much for taking my question. Good afternoon, gentlemen. I very appreciate C.C. and Mark sharing on TSMC's view on the longer-term outlook in AI. So I'm just wondering how does TSMC evaluate your second capacity expansion to think with foundry and wafer side? Since there is no problem in the foundry space, were you kind of concerned about potential overcapacity in the back-end side beyond next year? Or actually, we may see more upside at the foundry wafer side. So our - say, like advanced node utilization rate may go higher into next year? Thank you. That's my first question.

Okay. So Laura's first question is looking at our expansion of advanced packaging or back end versus the front end wafer as we are expanding the back end, but not the front end, does that imply that first, that our front-end wafer particularly leading node, we expect the utilization to increase next year? And then conversely, or is there a risk that we are over expanding or overcapacity risk for the packaging side?

Okay, Laura. Let me answer the question. AI today is a very hot topic. A lot of my customers right now increase their demand, and that will increase their front end demand, of course. TSMC almost have the major share or the largest share, let me say that in the front end wafer. According to the front ends adopting, we really work closely with our customer and to decide what is the back end that they need. And so on that perspective, we are planning our cohorts capacity, although probably still not enough, but we're working very hard to increase it. Overcapacity, not today is a concern, that it is concerns, it's not enough capacity to support all the very strong demand.

Okay. Thank you, C.C.

Yes. Thank you. That's very clear. Thank you very much C.C. And my second question is also about the gross margin outlook. If I am wrong, please correct me. I recall that the previous cycle, like 7 or 5-nanometer, the capacity usually will be three times in the third year of the new technology ramping. So I'm still - I'm wondering is still the case for N3. And particularly, we have seen that significant capacity intensity increase may lead to some margin pressure, particularly in the first few years. So I'm just wondering how does TSMC balance your technology leadership and also the margin situation? Thank you.

Laura, you said three times - sorry, are you referring to the revenue contribution? Sorry, you said N7...

The capacity, yes.

Okay. Laura. Let me answer this question. As I just mentioned, the N3, due to the increasing process complexity is becoming more challenging than the previous nodes. We - but at the same time, we will continue to sell our value and drive down the cost at the same time. But the - we still believe that N3 will be a long-lasting and large node for TSMC. With all the efforts, we still believe that the whole company's gross margin will be 53% and higher.

Okay. Laura, does that answer your question?

Thank you, really appreciate.

Yes. Okay. Thank you, Laura. Operator, let's move on to the next participant, please.

Yes, right now, we have Rolf Bulk from New Street Research. Go ahead please.

Thank you for taking my question. This quarter, you're also using the legacy node 16 and 28-nanometer in particular, were down around 15% to 20% Q-on-Q. And my question is, were there any particular end markets that caused this decline? And how do you think about the recovery of those legacy nodes? Should we still expect recovery in the fourth quarter of this year or is that more 2024 event? Thank you.

Okay. So Rolf's first question is looking on the mature nodes, such as 16 and 20, 18 nodes – he notes, sorry, that those all saw sequential declines in the second quarter. So his question is, what is driving this - what end markets are driving this decline? And what is the expectation for this in the second half?

Well, let me answer that question. The mature nodes wafer actually or the product actually try to be convenient for the smartphone or for the PC market or for the HPC. So while the total unit of smartphone become weaker and PC become weaker. So it's a high - the leading-edge technology node being also demand dropping and so the mature node. That's together. Did I answer your question?

Yes. Thank you. That's very clear. My second question, maybe you focus on cohorts and advanced packaging in general and also the weakness that you see in the remainder of your business. Could you maybe comment on the percentage of your CapEx spending that will go towards leading nodes, specialty nodes, and packaging this year compared to last year?

Okay. So Rolf second question is for 2023 CapEx, which our CFO has said towards the lower end of the 32 to 36% range, can we give a breakdown between leading-edge, specialty technologies and then the packaging testing, mass-making, and others.

Leading-edge technology accounts for between 70% to 80% of our total CapEx in the year, mature specialty technology between 10% to 20%, and the remaining are split between advanced packaging and EBO and some others, okay?

Thank you, Rolf.

Perfect. Thank you.

All right. Operator, let's move on to the next participant, please.

Next one, we have Sunny Lin from UBS.

Thank you very much. Good afternoon. Thank you for taking my questions. So my number one question is on 3-nanometer ramp-up. As we are going through third quarters on mass production, I think you must have a pretty good visibility for customer engagement for the coming few years. So I wonder now how should we think about the overall ramp of 3-nanometer if we compare with 5 and 7-nanometer? If we look at 5, you reach towards 18% of revenue in the second year of mass production and then about 24% of revenue in the third year. Whereas for 3-nanometer, I think the concerns by smartphone customers have been on cost. Then the question will be if HPC is significant enough to still drive a meaningful pickup of 3-nanometer. And so it will be greatly appreciated if you could provide us any kind of thoughts. Thank you very much.

Okay. So Sunny's first question is on the ramp of 3-nanometer. Her question really, I believe, is coming from a percentage of revenue contribution. She wants to know how is the ramp-up 3-nanometer? And then can it contribute to the revenue like N5, N7 in the past?

Yes. As I just said, we believe N3 will be a long-lasting and large node for TSMC. Now in terms of percentage, I think it's sometimes less important because our overall corporate revenue is much, much bigger these days than before. So I think you should also take that into consideration. But dollar amount-wise, it's a much bigger node.

And C.C. also said it's a multiyear strong structural demand.

Got it. Well, so I have a quick follow-up on 3-nanometer profitability. And so Wendell has provided a pretty good insight about the dilution for 2024. But historically, a new node would take about 7 to 8 quarters to get to corporate average after mass production. I understand now corporate average gross margin is also higher. But any expectations that N3 will become in line with corporate average gross margin?

Yes. So Sunny's question is looking at the 3-nanometer. Her question is really that with 3-nanometer and process complexities. Sunny, you're asking really kind of reach the corporate average over time?

Or is there a timeline that you are expecting?

Yes. Sunny, as I just mentioned, it's becoming more challenging for the leading nodes because of the process complexity increases a lot. It applies to N3. So it will be challenging for N3. And we actually mentioned that at the beginning of last year already, it will be challenging for N3 to reach the corporate leverage in the 7 to 8 quarters time frame like before. Yes. But however, part of it is really because of the higher corporate margin that we currently have.

Got it. Thank you. My second question is on 2-nanometer. And so if we look at your target for 2-nanometers improvement over 3-nanometer in terms of speed and power, the upgrade seems to be actually less than 3-nanometer over 5-nanometer. So I wonder what's actually the implication of GA transition to cost versus performance. Is the target somewhat conservative? Or are there other technological challenges that we need to consider?

Okay. Sunny, second was - it's really a third, but the question is on N2. She notes that the performance and the improvement seems to be less than 3-nanometer versus 5-nanometer. So could we talk more about that?

Yes, let me answer the question. Sunny, you have a very good observation. Yes, you are right. As I compare node to node from 5 to 3, the improvement becomes less from 3 to 2. But let me point it out; usually, we are talking about the performance, the speed, and also the density so that the geometry shrinkage. Now we focus on the power consumption reduction, which is still a poor node as a performance because as time goes by more and more customers, really, they are increasing towards greater power efficiency. This is very important for the data center; very important for servers. And that's what we are working on. So did I answer your questions, Sunny?

Got it. Thank you. It's good to know that you are on track to deliver a second generation of 2-nanometer in 2026. Thank you.

Thank you, Sunny. Operator, let's move on to the next participant, please.

Next one to ask questions, Brett Simpson from Arete.

Yes. Thanks very much. And the first question is for C.C. I was interested in getting a read on the customer reception you're getting for the new variants for N3, I think you talked about N3P and 3X. Are customers still as focused on N3E or are you seeing a preference for them to migrate to the new variants such as N3P, N3X rather than the N3E? And this as a follow-on for AI, when do we actually start to see N3 adoption for AI? Thank you.

Okay. So Brett's first question is looking at our 3-nanometer families and the continuous enhancements that we always have. He is asking what is the customer reception of N3P and N3X? How does this compare or cannibalize N3? And when do we expect AI related to adopt 3-nanometer family solutions?

Let me answer the last one first. AI application is already adopting our N3 technology node; we continue to improve our technology, as we always do. So we have N3E, N3P, and N3X. X is the actual performance that's for very high speed and very high, let me say, performance computing for some of the CPU applications. But N3Es are widely accepted by all my customers and the design starting from N3E, and we help them for some of them go to the N3P. So all together, every version, every variation, there's a lot of customer engagement right now.

Okay. Thank you, C.C.

Thank you, C.C. And good afternoon, everyone. Today, I want to talk about TSMC's global manufacturing footprint, status update.

This concludes our Q&A session. Before we conclude today's conference, please be advised that the replay of the call will be accessible within 30 minutes from now. The transcript will become available 24 hours from now, both of which you can be found and available through TSMC's website at www.tsmc.com. So thank you for joining us today. We hope everyone continues to stay well. Have a good rest of the summer, and we hope you will join us again next quarter. Goodbye, and have a good day.