[Foreign language] Welcome to TSMC's second quarter 2018 earnings conference and conference call. This is Elizabeth Sun, TSMC's Senior Director of Corporate Communications, and your host for today. Today's event is webcast live through TSMC's website at www.tsmc.com. If you are joining us through the conference call, your dialing lines are in listen-only mode.
As this conference is viewed by investors around the world, we will conduct this event in English only. The format for today's event will be as follows. First, TSMC's Senior Vice President and CFO, Ms. Lora Ho, will summarize our operations in the second quarter 2018, followed by our guidance for the third quarter. Afterwards, Ms. Ho and TSMC CEO, Dr. C.C. Wei, will jointly provide the company's key messages. TSMC's Chairman, Dr. Mark Liu, will host the Q&A session where all three executives will entertain your questions. For those participants on the call, if you do not yet have a copy of today's press release, you may download it from TSMC's website at www.tsmc.com. Please also download the summary slides in relation to today's earnings conference presentation.
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 on our press release. Now, I would like to turn the microphone to TSMC CFO, Ms. Lora Ho, for the summary of operations and current quarter guidance.
Thank you, Elizabeth. Good afternoon, everyone. Thank you for joining us today. I will first summarize our second quarter financial results, and then provide the guidance for the third quarter. Our second quarter revenue in US dollars was $7.85 billion, a decrease of 7.2% sequentially, but an increase of 11.2% year-over-year. In TWD, revenue declined 6% sequentially, mainly due to the impact from mobile product seasonality. Gross margin was 47.8% in the second quarter, a decrease of 2.5 percentage point versus first quarter. About 2 percentage point of the decrease was attributable to the absence of the favorable inventory valuation that I had mentioned in April, while the remainder was attributable to a lower level of capacity utilization, partially offset by cost improvements and a slightly more favorable foreign exchange rate. Operating expenses ratio was 11.33% as our revenue decreased more than our expense decrease.
Operating margin decreased 2.8 percentage points quarter-over-quarter to reach 36.2% in the second quarter. As I said during the last quarterly conference, the corporate tax rate would increase to 17.5% in the second quarter as we accrued the 10% tax on undistributed retained earnings. The tax rate will fall back to 10%-11% level in the second half, and the full year tax rate will be about 12%. Overall, our second quarter EPS was TWD 2.79 and ROE was 18.7%. Let's take a look at wafer revenue contribution by application. During the second quarter, communication and industrial standard decreased 14% and 1% respectively, while computer and consumer increased by 34% and 23% respectively. Let's take a look at revenue by technology. We began volume production of seven nanometer process technology in the second quarter.
The revenue contribution was less than 1% in second quarter, and will ramp to above 10% in the third quarter. 10 nanometer contributed 13% of total wafer revenue during the second quarter, while the combined revenue from 16 and 20 nanometer accounted for 25%, and 28 nanometer was 23%. Advanced technologies defined as 28 nanometer and more accounted for 61% of total wafer revenue. Moving on to the balance sheet. We ended the second quarter with cash and marketable securities of TWD 749 billion, an increase of TWD 65 billion from the first quarter. On the liability side, current liabilities increased by TWD 121 billion as we accrued about TWD 208 billion for cash dividends, which will be paid out today. On financial ratios, accounts receivable turnover days decreased four days to 38 days.
Days of inventory increased 11 days to 74 days, primarily due to the ramp-up of 7-nanometer, which has a longer cycle time and a slight increase in raw wafers. Let me make a few comments on cash flow and CapEx. During the second quarter, we generated about TWD 130 billion cash from operations and spent TWD 60 billion in capital expenditures. As a result, we generated free cash flow of TWD 70 billion. Overall, cash balance increased by TWD 54 billion to TWD 632 billion at the end of the second quarter. In the US dollar terms, the capital expenditure spent in the first half of 2018 totaled $4.5 billion. I have finished my financial summary of the second quarter. Let me provide you the third quarter guidance.
Based on the current business outlook, we expect third quarter revenue to be between $8.45 billion and $8.55 billion, which is an 8.2% sequential increase at the midpoint. Based on exchange rate assumption of $1 to TWD 30.50, our third quarter gross margin is expected to be between 48%-50%. Our third quarter operating margin is expected to be between 36.5%-38.5%. This concludes my financial summary. Let me make remarks on capital expenditure and profitability. I will first talk about our capital expenditure. At our last investor conference in April, we stated our 2018 CapEx budget to be between $11.5 billion and $12 billion. We now plan to trim our CapEx budget by about $1.5 billion and expect our 2018 CapEx to be between $10 billion and $10.5 billion. The reduction of 2018 CapEx come from the following three factors.
The first one, about $700 million US dollars came from delay of payment to 2019 due to leading edge tools relocation schedule adjustment. However, the planned capacity remained unchanged. Second, about $600 million US dollars comes from efficiency gains that allow us to spend less on tools. Third, about $200 million comes from the US dollar appreciation against EUR and JPY. My second remark is regarding profitability. Now I will talk about third quarter 2018 gross margin and the overall outlook of our profitability. Our third quarter gross margin is expected to improve from second quarter by more than one percentage point. The increase mainly comes from better utilization rate, more favorable foreign exchange rate, and improved profitability of our backend business.
Offset, however, by the unfavorable technology mix, which includes the ramp of 7-nanometer that is expected to dilute our gross margin by more than one percentage point in the third quarter, and the lower contribution from 28-nanometer. The net gain in gross margin, therefore, is expected to be slightly more than one percentage point. That said, TSMC's financial objective remain unchanged. Our goal is to achieve revenue and net income compound annual growth rate in the next few years to be between 5% and 10% in US dollars. Gross margin to be about 50%, operating margin to be about 39%, and ROE to be above 20%. This ends my remark. Now let me turn the microphone to C.C. for his comments.
Thank you, Lora. Good afternoon, ladies and gentlemen. Let me start with our near-term demand outlook. We conclude our second quarter with revenue of TWD 233.3 billion, or $7.85 billion, in line with our guidance given three months ago. This result reflected mainly a strong demand from high-performance computing, including cryptocurrency mining, but was offset by seasonal decline in high-end smartphones. Moving into third quarter 2018, our business is expected to benefit from new product launches using TSMC's industry-leading seven nanometer technology, while cryptocurrency mining demand will decline due to weakening cryptocurrency prices. That being said, we do see slight improvement in smartphone demand in second half of this year as compared to our forecast three months ago. GPU demand for AI and gaming continue to increase.
For the full year of 2018, we forecast the overall semiconductor market, excluding memory, will grow by 5%, while foundry is expected to grow by about 7%. We forecast TSMC's 2018 revenue in US dollar will grow by a high single-digit rate rather than the previously stated about 10% due to general weakness in cryptocurrency mining demand. Now let me move to our long-term business growth driver. As we stated three months ago, that we are optimistic about the development of the industry's mega trend, particularly AI and 5G communication. Recently, we have observed more promising development. For example, we see AI continues to fast proliferate from data center to edge server and to end client devices.
As for 5G, major operator in several countries has rolled out a development schedule, while multiple ODMs and IC vendors have planned their 5G products, which are set to ramp in the coming two years. Despite the slowing unit growth in smartphones in the near term, we expect the development of 5G will fuel the next wave of smartphone growth, both in units and in silicon contents. In HPC, we expect the increasing workload in data center and complexity of AI will boost the demand for AI accelerator, GPU, and CPU in server. We also expect the introduction of next-generation video gaming will aid growth of HPC. We believe all our four growth platforms, smartphone, HPC, IoT, and automotive, are well-positioned to benefit from the longer-term mega trend of AI and 5G.
With our leading and comprehensive technology offering, our vast capacity, and our policy of not competing with customers, we will be able to support our customers to expand their markets and therefore fuel our future growth. Let me talk about the N7 ramp-up status. TSMC's seven nanometer technology is leading in the industry. It has the best performance, power, and area density, and its schedule is ahead of competition. For the tape-outs that we have completed for customers all have very good yield and performance. We forecast a total of more than 50 customer product tape-outs by end of this year from a wide range of applications covering mobile, server CPU, network processor, gaming, GPU, FPGA, cryptocurrency, automotive, and AI. Our seven nanometer is already in volume production and accounted for less than 1% of our total wafer revenue in second quarter.
It's expected to jump to more than 10% of our wafer revenue in third quarter, and is estimated to contribute more than 20% revenue for us in fourth quarter in this year. Now we talk about N7+ and EUV. Our seven nanometer plus, or N7+, can leverage the success of our N7 and enjoy 15%-20% better gate density and more than 10% power reduction. With a few EUV layers replacing certain immersion lithography process, we're able to have fewer masking layers, shorter cycle time. Less process complexity. Therefore, we expect to achieve better yield as compared to our N7. Furthermore, as we have fine-tuned all the advanced equipment to their optimum condition during the ramp-up of both our 10 nanometer and seven nanometer technologies, we believe we can leverage our production learning to 7+ and enjoy the industry's best defect density among our peers' comparable technologies.
The silicon result from our N7+ today are very encouraging. The lead N7+ product has taped out early this month, and we expect to receive a few more tape-outs by end of this year. Volume production will start Q2 next year. That is Q2 2019, which will be the world's first EUV foundry production by that time. We have made ready multiple EUV scanners to support not only the N7+ development, but also N5 development. Our silicon data have proved all the benefits we expect from process simplification with EUV. In addition, we have also started our N3 technology development using EUV. Now let me talk about the EUV status. Good progress continued to be made in the EUV infrastructure in the last few months. They include photoresist, mask defect and yield, particle defects and transmission. Besides the silicon development, EUV technology continues to mature toward high-volume production.
We have achieved 250 watts source upgrade in April, and the tools are running smoothly with minimal degradation and high level of uptime. In summary, we started EUV development work early, and we have secured the largest number of EUV tools among our peers to be ready for 2019 volume production for N7+, and 2020 volume production for N5. Let me talk about N5. Our five-nanometer technology, N5, is progressing well. The 256 megabits SRAM yield is one quarter ahead of schedule, and the device performance is well on track. TSMC's N5 will begin risk production in first half 2019. We believe it will be the most advanced technology in the foundry industry by that time. We are actively engaging with several lead customers, and we are running their test chip now. We expect to receive first customer product tape-outs in first half 2019.
Volume production is expected to start in first half 2020. I'm talking about now with our specialty technology. We work closely with our customer to accelerate our specialty technology roadmap. By leveraging our largest capability, we develop specialty technology features such as MEMS, CMOS image sensor, high voltage, power management IC, emerging embedded memories and analog to more advanced node, including 55, 40 nanometer and 28, 22 nanometer. These efforts bring scaling benefit to our customers. Our 22 ULP process is qualified and on track for risk production in August. That is next month. About 40 customer product tape-outs using our N22 are planned in the next few quarters, covering a broad spectrum of applications from digital TV, consumer electronics, to IoT and RF connectivity. N22 have 15% performance gain, 25% power reduction, and 5%-10% die area shrink when compared with N28.
With this improvement, we expect our N22 will extend our leadership at 28 nanometer node and enjoy a long and successful economical life. Since we have already built a broad technology capacity at 12-inch for various specialty technologies, we are migrating some of our customers' products from eight inch to 12 inch with seamless transition, where our customer can enjoy flexible capacity support and further scaling benefit. That's all. Thank you for your attention.
This concludes our prepared statements. Before we begin the Q&A session, I would like to remind everybody to limit your questions to two at a time to allow all 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 questions in Chinese, I will translate it to English before our management answers your question. For those of you on the call, if you like to ask a question, please press the star, then 1 on your telephone keypad now. Questions will be taken in the order in which they were received. If at any time you would like to remove yourself from the questioning queue, please press the pound or the hash key. Now, let's begin the Q&A session. First question will be coming from Credit Suisse, Randy Abrams.
Yes. Hi, thank you. The first question I had on the 28 nanometer, you talked last quarter about it being a bit underutilized. Could you talk about now your confidence to backfill that node? Last quarter, you mentioned a big mask investment. How much of that is tied to some of these backfill activities, versus the mask writer investment for some of the advanced applications?
You have two questions.
Two parts.
The first one is 28 nanometers are loading.
Yeah. The 28 loading, your confidence to get that loaded, the type of applications.
Okay.
Second part of that is the mask investment. How much of that is tied to these backfill applications versus advanced capacity?
All right. In my statement, I say that we improved 28 nanometer to 22. Now a lot of 28 nanometers are tape-out, has been changed to 22 nanometer. We start to ramp up 22 nanometer next month. It will take probably a few quarters so that the 22 nanometers volume will be high. I expect that after a few quarters, the 28 nanometer node, the confidence to fully utilize is high. Now you're talking about we invest.
Can I add this one?
Yeah.
Regarding your question, by the way, thank you and good afternoon, everyone, to join this conference and also the online participants, particularly on this hot summer afternoon. On 28 nanometer, this is the eighth year of our 28 production. Inevitably, the competition is coming gradually, and with some expectation, they build capacities. The 28 nanometer has a lot of nuance in it. You have high performance. There are different grade of speed. Now C.C. is talking about additional sub-node, 22 nanometer. I think the underloading will be temporary, and we intend to backfill this capacity based on our technology differentiation.
Randy, your second question is we invest on-
Yeah. It was a-
mask as a capacity.
Yeah. Very high CapEx last quarter for mask. Just what was the purpose or what applications you were tied to this high capacity relative to history that you called out the mask CapEx?
Actually, we invest on mask capacity because of a leading-edge node. Most of it is because of leading-edge node. Because it's very complicated in making the mask, and many layers. For related to 28, it's part of it. Part of it, because we still have a very high activity in 28 nanometer node. Actually, still that's the number 1, number 2 tape-outs account for TSMC's total tape-outs.
Okay. It sounds like it's both 28 plus for the seven nanometer for the mask.
The 16.
16. Okay. The second question I have on cryptocurrency. In the past, it was on a lagging node like 28 and 16, where it was filling capacity as a second wave application. How do you view crypto? It's slowing in the second half, how do you view devoting new capacity as crypto to get the best performance and power one to move to 7? How do you see that market and also devoting capacity if it becomes more of a first wave application?
Well, cryptocurrency all related to the hash rate. Naturally, they will move to very high-end leading-edge technologies to improve the performance and lower down the power consumption. Whether that it will be that at the same time as a high loading when we move into the leading edge, that probably we cannot say that. It's volatile in the business unit, and it continue to depend on the cryptocurrency's pricing. We don't pin our capacity because of that.
All right. Thank you.
Next question will be coming from UBS, Bill Lu.
Hi there. Thank you very much for taking my question. Lora talked about the slightly lower CapEx and TWD 600 million of that is from the efficiency gains. Can you talk a little bit more about that? Is that mostly from seven nanometers or where is the gains coming from?
Okay, the TWD 600 million efficiency gains actually covers various area. Number 1 is process simplifications. If you have a simpler process, you don't need to buy that much tools, number 1. We also share tools between R&D and operations. If we can find more opportunity to share, then we don't have to buy that much tool, okay. The third one is relating to the back-end equipments. We do see with the effort from TSMC and customer jointly, we do see a test time reductions. So those are the few areas that we classify as efficiency gain. Of course, you know we are always doing that, and we just have results coming right now.
Lora, I can add more color to that. What we call the efficiency is that. Let me tell you, we plan customers' product out, so you need some kind of a processing cycle time. You need to buy the tool. Tool has a lead in time. Now we improve our process cycle time quite a lot. Now you don't have to buy the tools so early, 1. The 2 is, equipment vendor work with us, so they also shorten their lead in time, so we don't need to buy so early, and that's what we call part of the reason of the core efficiency improvement. It's continued to improve. That's a progressing every day, every month.
Yeah, I guess I'm just wondering how much of it is one-time, and how much of it can I translate into lower CapEx going forward?
It will be continuous effort because it's our job to shorten the cycle time. Usually, if we ramp up a new technology, the cycle time is a little bit longer. Now we are getting better. We are getting very aggressive and very competitive cycle time, that's why we can cut it.
Sorry, one last follow-up on that. I don't know the exact number, but I think the company has said, CapEx next several years, if I'm not mistaken, TWD 10 billion-TWD 11 billion. Do you want to comment on that, whether there's any new thinking or changes?
Of course, number is a kind of rolling forecast. As far as we can see, it's still within that range. The CapEx intensity, as I said earlier, is somehow between 25%-30%. We're still with that view.
My second question is on the seven nanometers. It sounds like it's progressing quite well, and maybe slightly ahead of plan in terms of yield improvement and such. Can you just talk about the yield and the improvements that you're making there versus the previous couple of nodes, maybe 10 and 16?
The yield progressing very well. Performance is good. As compared with previous node, it's comparable, probably a little bit better. I cannot give you exact numbers.
Basically, seven nanometer, we just have a better architecture and make the yield improvement easier.
Would that imply that you get to corporate average gross margin a little bit earlier as well?
I think the rule of thumb, I said about eight quarter from mass production to quarterly average is about the same. We have seen that for every node have that kind of pattern. There's no exception this time.
Thank you.
Next question will be coming from Deutsche Bank, Michael Chou.
Thanks for taking my question. The first question actually is for 28 follow-up. You mentioned after a few quarters, your UTR for 28 overall should improve. Is that under the assumption of the same capacity this year? Do you think that next year, total 28/22 nanometer sales will be up year-on-year? That's my first question.
Michael, let me tell you that TSMC continue to improve the productivity. Actually, even we don't spend the CapEx, the capacity continue to increase, so that every year when we're talking about our business versus the utilization, actually, the capacity continue to increase. Your question is the revenue?
Yeah. Would 28/22 nano together be up year-over-year in 2019? Do you maintain the same I shouldn't say maintain the same. Given the underutilized 28 nanometer, would you consider convert some 28 nano capacity to advanced node going forward? You will just maintain the same capacity?
Well, converting the capacity is the last resort, okay? We want to fill the capacity based on technology. This is a constant effort. 22 nanometer is one example. There are other specialty technologies on the pipeline to offer, that is our first resort, hopefully to get to fill it up as much as we can. If by then, we'll have other capacity adjustment option. At this time, it's not on our plan.
Do you expect the-
The revenue?
Yeah, the revenue.
I would expect in 2019, probably drop a little bit, because we are ramping up, and after that, we'll start to increase.
You mean 2020?
Yes.
Okay. Sure. Thank you so much. Second question regarding your 7-nanometer progress. You mentioned, you will do servers CPU, if I hear it right. Would that be ARM-based or x86, or you cannot give color for that?
It's too specific, so it's a CPU.
Thank you.
Next question will be coming from Citigroup Roland Shu.
Good afternoon. Thanks for taking my question. You talk about the 7-nanometer will be more than 10% of the total revenue in Q3, versus less than 1% in Q2. The increase for the 7-nanometer is more than 10%. The Q3 overall revenue guidance growth is less than 10%, means that the overall revenue from 10-nanometer and above are declining. Can you give us more color for this technology node? Which node is strong, which node is weak in Q3? Thank you.
Which node is strong, which node is weak?
Yeah. On the overall 10-nanometer, 16, 20, 28, and above, the overall revenue in Q3 will be decline sequentially. I would like to know for which node specifically is strong and which node is weak in Q3.
I cannot comment on that, right?
We know 28-nanometer actually will be still decline. Is that right?
Right.
Means 16 will be still fully loaded.
You want to ask it one by one, huh? Starting from 0.15 micron or something like that?
Yeah. That would take about 10 minutes.
That's good enough. Actually, you know that our 7-nanometer is very strong, 7 and 10 very strong. 16-nanometer is fully loaded. Okay? That's good enough information for you to estimate.
Okay, thank you. Follow-up question for this 7-nanometer ramp up, because now we think 7-nanometer will be a major node, like 16 and the 28-nanometer. Question is that for next year, for 7-nanometer revenue ramp up, will it be similar as 16-nanometer in 2016 or 28-nanometer in 2013, which is the second year for you ramp up this
Stronger.
Technology node.
Stronger than any node we have in history.
Okay. That means that for next year, for maybe on the single quarter, I think that for 7-nanometer to reach 30% of the total revenue will be a reachable target?
Oh. I see. I don't want to go to quarter, Roland. I would say, we just said that 7-nanometer will account for 10% third quarter, more than 20% in fourth quarter. For next year, 7-nanometer will account for more than 20% for the whole year. That's all I can say.
Okay. Thank you. Second question is for June monthly sales unusually declined by double-digit percentage point year-on-year and quarter-on-quarter. Was that because the customer did not take the wafer shipment in their quarter end, or was that purely because of the demand weakness?
Don't look at the monthly revenue. Look at the quarterly revenue, which we have just provided guidance.
I still have to ask, how about the Q3, the monthly sales linearity in Q3?
The monthly linearity is not really in our control. The customers shipping, they have their schedules. Sometimes within a couple of days range, they want to adjust their inventory, and there are many factors. It's not our target to hit target each month, have a financial goal. Rather, it would be a quarterly goal.
Okay. Thank you.
Next question will be coming from CLSA Securities's Sebastian Hou.
Thank you. My first question is on N5. Regarding the tape out, early mention that you expect to receive the first tape out in first quarter or early next year. How many tape out do you expect to receive by the end of next year before the mass production begin in first half 2020?
Actually, we don't know at this time. 5-nanometer is a node, of course, is in the development. Right now, we know that we have customer test chips, put it in our test vehicles, several of them. Those are the sub-functional block of their products. At this time, most of them haven't commit the tape out date. We really cannot summarize the number and particular time. The engagement activity is very active.
How do you compare the engagement activity or customer interest to N7 at the same stage?
Okay. At this time, probably less than N7 in the beginning, because this is a big investment for our customers, and I think they are planning in a later time when N7 gets to ramp up in fluidity.
How do you see the ROI on N5 versus N7? If less interest right now, and the investment probably larger for you, how in terms of the payback ROI, how do you compare N5 and N7 at this moment?
It will be very similar.
My second question is on the advanced packaging that I think TSMC has continued to roll out new or launch new packaging offerings almost every year. Can you give us some update on that and how many product offering you have on this? Do you see TSMC become even more serious in packaging rather than just in the past to facilitate the big customers?
We have been very serious all the time. In fact, to add some color to it, in the future, we see the high performance computing is very important. TSMC's advanced packaging help the customer to improve the system performance. Now more and more activities to engage with the customer for their high-end computing devices.
Can we assume that a big portion of the new tape out you're receiving on 16, 12, 7, and potentially 5 nanometers will adopt your packaging solution, whatever is InFO, its variants or CoWoS?
Sebastian, the advanced packaging is a major thrust of our technology development. The reason is, the Moore's Law over the years is indeed slowing down. That means that we have to create more value to our customer on their product to allow the technology migration continue. Advanced packaging is, you can almost say, is a parallel thrust together with Moore's Law to develop our customers' product. Advanced packaging development encompass our customer's architecture development and algorithm development. In that, I'm talking about to go into the 3D IC. That is the purpose, to add to the Moore's Law development to maximize our customer's product. This is not a typical packaging business. It is a major technology development for TSMC. Although today it is indeed more expensive, some of the market sector cannot afford it.
Today we see the affordable segment will be the high performance computing, particularly the server and also the high speed networking area. Of course, the current biggest customer is in the mobile, I believe the mobile sector will gradually coming in when they see the value. It is not a typical Moore's Law pace. Rather, it is an additional development effort to augment it to the Moore's Law. That's my response in the total picture.
This part of the things are now, before the InFO launch, now you have more offerings. Presumably, we can assume you can generate more packaging revenue going forward, that is incremental compared to the past. Can we assume that your packaging revenue may potentially? I know you reported non-wafer revenue part of business and amounts, that is about over 10% last year. Can we assume that those part of the business can grow faster than your wafer revenue going forward?
Of course. Of course, we expect that. Again, in this semiconductor packaging in the past hasn't been playing the role of increased system performance. This is new and we indeed first see in our biggest customers' verification products. There are new customers are interested in that, including some of them are mobile, some of them are high-performance computing. Cost is always a factor in determining when do they move in to this technology. That varies segment by segment.
Thank you.
Next question will be coming from J.P. Morgan's Gokul.
Thank you. My first question is on N7+ and N5. Given N5 is a pretty high layer count for EUV. In your early engagement with customers, are you seeing customers trialing N7+ at least for some small volume products and then going to N5? Or are you seeing more customers directly going to N5? Could you also give a view on how big N7+ is likely to be? Is it going to be a small volume node while N7 is still going to be majority of the seven nanometer family? That's my first question.
Gokul, there is no correlation that whether customer need to go to N7+ and then go to N5. Customer choose working with TSMC for their product development. They choose the best technology at the time that fits their product well. No correlation. You are talking about whether they need to go through this? No.
You feel that customers are already comfortable with the EUV throughput at that level of layer count to directly go?
They are, because we communicate with them our progress and our status. Actually, we believe we have a very good progress, and they feel comfortable with.
Okay. Just one clarification on the smartphone commentary in terms of slightly better demand that you're seeing in smartphones compared to the last three months. Could you classify it? Is it more coming in the mid to low end, or is it really at the high end? Second question is it because TSMC is gaining more share, compared to what you expected to, or your share gains are coming faster? Or is it a view on the overall smartphone market itself?
Well, mostly it's on high-end smartphone. All right? The mid low end, the industry-wise is weak. Although it's gradually coming back, its pace of coming back is slower than expected.
Next question will be coming from Goldman Sachs, Donald Lu.
Okay. Yeah, good afternoon. I have two questions. First question is about China. Here I have two specific questions. One is that, earlier this year when the ZTE saga starts to play out, MediaTek initially said it cannot ship to ZTE. My understanding is TSMC has never turned down ZTE's wafer order. I'm wondering what's going on here. Is there a law or a rule or something going forward, going to dictate how you can serve Chinese customer? Because Chinese customer now accounts almost a quarter of your revenues. Second on China is, UMC is going to list its China entity in China, and arguably can get a lot of capital for R&D and growth. Will TSMC one day consider that? I'm sure it will be very welcomed. Yeah, I have another question, but I will wait.
First on ZTE. Yes, when U.S. government put a ban on the shipment to use ZTE, indeed, most company, ZTE supplier stopped the shipment. However, TSMC is not a direct supplier to ZTE. Indeed, they do have a subsidiary of ZTE. According to their rule, you need to have a certain percentage of value added from the U.S. For that particular subsidiary, the value added is mostly from China and from TSMC. That also is beyond the restriction scope.
Sorry, can you explain a little more here? I understand ZTE also makes chips and that's manufactured at the TSMC. That's not correct?
No, no. ZTE do not make chips and send order to TSMC. They do have a subsidiary and a small design house and very small volume, have some business with us. However, we talk to our Taiwan government and outside counsel. The value added, it doesn't occur at U.S., so it was not in the restriction scope.
Hypothetically, if HiSilicon have a problem similar to ZTE, would that be a problem?
I don't want to answer the hypothetical question, okay?
Understand.
Thank you. Yeah. You have another question?
Yeah. It's on the potential A-share listing of your.
Okay. This goes to Lora. Lora told me we're not short of capital. Okay.
Yeah, company go for IPO, main purpose to source the funding. We have enough funding to support our growth. There is no plan for us to do that.
Second question is on seven nanometer. The first is about China. seven nanometer, I have a question. What was TSMC's market share at 16 nanometer and at seven nanometer foundry market? Also, how compare the TAM, how much is seven nanometer TAM versus 16 nanometer TAM in terms of market size?
Our seven nanometer of market share and 16 FinFET. Okay, 16 FinFET market share is very high, and seven nanometer is even higher.
How much higher? 10% more?
Cannot be so specific, right? High is high.
It's the TAM of seven nanometer
The TAM of seven nanometer is bigger.
Bigger than 16?
Yes.
Why is that? Historically, that seems to be decreasing. Why seven nanometer?
Well, there are some reason that we are talking about the AI, we're talking about 5G, high-performance computing, all add together.
Talk about dollar of TAM, right? Thank you.
Let's actually go to the line for the first caller on the line. Operator, please.
Thank you. We have a question from HSBC . Your line.
Apologies about the background noise. In the fourth quarter conference call, you disclosed last year, 7% of revenues came from back-end services. Now you're talking about also improved profitability in that area. Can you help us understand your outlook for total back-end services contribution this year and how significant the profitability improvement is?
Back-end contribution this year and profitability.
The back-end contributions to total revenue is increasing. It's slightly higher than last year. We see more advanced packaging coming on the line. I think back-end and EBO adds up is slightly more than 10% of our revenue versus about 10% last year.
Okay, thank you. One more question for you, Lora. I remember last year you also had 10 nanometer going to greater than 20% of revenues in the fourth quarter. This year, you have seven nanometer going to greater than 20% of revenues in the fourth quarter. The guidance that you just gave for the third quarter suggested 100 basis points of headwind from the initial ramp of a seven nanometer. I'm curious, as seven nanometer goes to more than 20% of revenues in the fourth quarter, does the volume start to offset and you won't get as much of a headwind? Or the fact that it's such a large percentage of mix, will seven nanometers still be a significant headwind to the potential margin progression going into the fourth quarter?
Steven's question is, if we ramp seven nanometer to more than 20% revenue in the fourth quarter, what will be the margin headwind brought by this ramp?
In the first year of production, usually the margin for seven nanometer is lower than corporate average. We do see, with the third quarter's guidance I just gave you, which include 10% contribution from seven nanometer, this is going to dilute corporate margin by one percentage point, slightly more than one percentage point. Going forward to the fourth quarter, where seven nanometer will account for more than 20% of total wafer revenue, we expect the dilution will be about 2.5 percentage point to corporate gross margin.
Okay, great. Thank you. If I could just sneak one last one in. I'm surprised we haven't talked much about feedback from your customers on potential trade war impacts. Does TSMC have a viewpoint, and what are you hearing from your customers on potential impacts for the rest of this year?
On the U.S.-China trade tension, they issued tariffs on the three waves. First wave is on the $34 billion products. Second wave on $16 billion, third wave is on the $200 billion.
200.
Only the first wave has been executed today. In that, we see very minimal impact for our business and our customer. Those are the mostly related to the IC, related to the semiconductor, it's only discrete devices in the first wave. On the second wave, although it's still in the comment stage, we do a thorough check on that, in effect. It does include integrated circuits. We look at it's still a very minimal impact effect in our business so far. Even though the SIA in the U.S. is still protesting about including the IC into the second phase, what I'm talking about, even though they put it into effect, it's still a very minimal effect.
Great. Go ahead.
Thank you.
Thank you.
Next question will be coming back to the floor.
Next question.
Sorry. Let's come back to the floor first, and will be from Morgan Stanley's Charlie Chan.
Thanks for taking my question. My question is actually a follow-up to previous questions. First of all, is the smartphone semi better in third quarter? Do you see an upward revision of smartphone semiconductor, or you mean it's better than second quarter? Can you comment on that?
Of course, it's better than second quarter.
Okay.
Three months ago, we are a little bit more conservative. Recently, the development in the smartphone units actually recovered. It's better than what we forecasted three months ago. It's not a big deviation from our forecast at the beginning of this year.
Okay, thanks. I guess another question that people also care about is your revenues exposure to the crypto semiconductor. After two quarters, what is the exposure in first half, and what do you think the % of crypto semi will be in second half?
We already mentioned that because of our cryptocurrencies pricing, we forecast a softer demand from the cryptocurrency mining in the second half. As what the ratio between the first half and the second half, I cannot say specific, but with one of the reason is very uncertain, right? It's volatile.
Our own forecast, we lower it down.
Okay.
Yeah. I think the full picture is, currently, we at our best effort to forecast a weaker cryptocurrency. However, it's largely compensated by the increased strength of the smartphone. That's why the total number in our forecast is not that different. However, cryptocurrency still have some uncertainty. Right? That's why we adjusted the 10% number. Still, we're going to fight for that to reach close to that, yeah.
Right. Thanks. My next question is regarding your AI semiconductor. It has been a very strongly growing sector over the past two years, right? It's going to be another key growth driver for coming years. Can you give us some numbers? For example, the AI semiconductor revenue contribution this year, last year, and the growth rate trajectory in the coming years, especially next year, are going to see AI semiconductor double. I want to define this AI semiconductor a little bit. That should exclude the AI features, including the smartphone chipset. I'm referring to those discrete chips, no matter it's for cloud AI or the edge AI. Thanks.
Wow. That's a good question, but very hard to answer because most of the AI's functionalities today is embedded into the existing product. For example, you look at the smartphone. In this case, the application processor embedded a lot of AI functionality inside, so you can have a lot of new features, right? Look at the face recognition, the voice recognition, and then et cetera. If you want to specifically identify what is the increase of last year, what is increase of this year is pretty hard for us to do. We can see the activity going up because you look at networking processor keep coming, application processor for all the smartphone keep increasing functionality. You can estimate what kind of a die size they increase, so the silicon content. By doing that, you probably can figure out what is the contribution.
Why TSMC making it so big deal, because we saw the trend from AI and 5G all combined together. That will be a mega-trend.
Let me add some for the future communication purposes. AI is in every growth segment we have. Okay? In the smartphone, in the automotive, in IoT, AI increase the silicon content. For the high-performance computing, indeed, there are a lot of discrete AI chips. With those discrete AI chips, that include XPU, accelerator GPUs, and network processors, and some of them even gains in that. Those are discrete GPU. As we speak today, indeed, high-performance computing growth is pretty fast. Last year was 40%, this year is 25%? Around that. The reason it's slowing down, because cryptocurrency is included in the high-performance computing. Going up is still the highest growth sector today we have. Next year, we're just trying to help our customer to expand their market as quickly as possible.
I think we still need to go back to the line for the next caller. Operator, please.
Yes. Next question is from Agency Partners. Douglas Smith, your line is now open.
Hi. Thank you. A couple quarters ago, Mark Liu said that there were no plans to expand manufacturing capacity in the U.S. Because of current events, has that been rethought? The second question is, what steps is TSMC taking to protect its intellectual property, given there's been a lot of allegations of IP theft recently?
First question is, we have mentioned that we have no plan to expand manufacturing capacity in the U.S. Have we changed our mind?
No. We did not. What? It's working. No, we did not change our plan. Only when it's necessary we will do it. So far, we do not see the need to establish a new fab in the U.S.
The second question is, how do we protect our intellectual property against theft?
Yes. Actually, this is the core of China-U.S. trade tension, we take it very seriously, too. Theft, we call it espionage, can come from everywhere, not just from one country, or there are sometimes indirect, come to getting our proprietary information. I can only say that we are fiercely protecting our IP and information, because those IP information is origin developed in TSMC over the past more than 30 years, and that is ours. No other states can take it or putting the effort espionage. That is the main focus among all these tension.
We come back to the floor. The next question will be coming from UBS, Bill Lu.
Hi, thanks very much. There's been quite a few questions today on advanced packaging and how that's sort of helping with Moore's Law slowing down. If you look at these HPC applications, the move to parallel compute, the requirement to access high bandwidth memory, that I think is a big part of the packaging trend. I just feel like packaging is going up in value, but so is memory, right? Is there a case to be made for TSMC doing more there, either in terms of partnerships or in terms of own efforts or just comment on that space?
You're right. We don't produce memory, neither DRAM or flash. We work very closely with three memory houses, including Samsung and Hynix, and Micron. All the engineering work together very closely. At this point, the engineering work is further intensified. This is our strategy, is that we still want the memory supply can come from multiple sources for our customer, so that they can most freely develop their product. Yes, among the three, some of the memory company work closer with us, but all three are working quite close with us to help our customer's product to get to the market.
Second question is on your ASP. If you look at last several years, CapEx going up, now the CapEx intensity is coming down a little bit. Yet you're going to have big market share at seven nanometers and the big market that is upcoming. What is the thinking here? Should we think that with lower cost, that you're going to pass it on to the customers, given that you are going to get to your ROE goals? How do we think about that?
Bill, I thought you were asking about ASP, and then you are also asking about how we deliver value to our customers?
Sorry. I guess my question is, if you think about ASP with your last several nodes, ASP has been going up, right? Now that your cost structure is coming down, does that mean we should expect flatter ASP going forward?
Okay. Do we share our cost reduction benefit with our customers more than before?
Which is reflected in the price. That's what you are asking.
I just feel like, last several notes, the leading edge actually has been shrinking, right? Now you've got a big market, your cost is coming down. If you're not going to raise price, you potentially benefit more because the size of the market might grow. Just how do you think about that problem, I guess?
I still don't understand that you are talking about that it's getting more mature, the cost is down, we have to lower down our price.
I'm sorry. I'm not being very clear. Just comment on ASP next couple of years, I guess.
We sell the value.
Yeah.
We don't sell by the cost. That's the first rule. We work with the customer to make sure that their product can be sellable, feasible in the market. That's all the rule of thumb that we consider.
Okay. Thank you.
Next question will be coming from Credit Suisse, Randy Abrams.
Yes, hi. Thank you. First question I want to ask on the four growth platforms, if you could give an update. Mobile, I think in the past, early in the year was flat, or that was the expectation, then I think came down. If you could give your view now, that mobile platform expectation year-over-year. Then for the other three, an updated snapshot, like how large HPC, IoT, and auto are for TSMC now.
Okay. Let me answer that. As I just mentioned, I think that because of introduction of the AI and 5G, even we start to forecast that the mobile smartphone, the contribution to our growth will continue. There's one surprisingly that good result from the HPC. HPC's contribution now it will be comparable to the mobile phone, then followed by IoT and automotive. That now is a two big contributor. We used to say mobile is 50%. Now it's a little bit less than 50%. HPC will be closing to the mobile's contribution now in the growth, per se.
Okay. I guess, I think HPC has been running or growing from 20%-25%, or just kind of the range now for HPC. I think the last metric, auto, was about TWD 1.4 billion. I'm not sure if you've given the IoT, if you have a rough, just so we have a rough picture, since that's the way we're looking at the business now.
Well, you're asking a question through lens of the future. Look at it, high performance computing today you said is 25%. We expect that percentage will increase slightly. Mobile, the percentage will decrease slightly. IoT and automotive, those are roughly about 6% or 7% of our business. That growth rate is pretty fast. We are talking about more than 20% growth year after year. That is our current picture of the growth, and that is also why we see these as four growth drivers for TSMC.
Okay. In the 6%-7%, is that combined or each one?
Each one.
Okay. The second question I have on gross margin, following up to Steve's question. There is about 1.5% headwind in Q4 from seven nanometer. You are also guiding growth, again in fourth quarter, based on the full year guidance. Do you expect you can offset the depreciation, or sorry, the seven nanometer headwind with growth to keep margins at near similar levels?
Actually, the headwind for seven nanometer in the fourth quarter is 2.5, not 1.5 percentage point, because volume is pretty big.
2.5 over third quarter. Okay.
Oh, over what third quarter? Yeah, you're right. 1.5 more than third quarter. Well, there are many factors. 1.5, it's not that big. There are things can be done if we improve the product mix. If the utilization gets better and we have better cost efficiency, then you can get there.
Yeah. Okay.
I'm not projecting that, but that's all the effort we've been exercising in the company.
Thank you.
Next question will be from CLSA Securities, Sebastian Hou.
Thank you. My first follow-up is to clarify what Mark just said on the HPC revenue growth. Was it 40% last year, 25% this year?
Yes.
Okay.
Including the crypto.
Okay. If we look at the full year guidance for this year, it's high single digit rate. If we take cryptocurrency out of it this year and last year, apple-to-apple comparison, without cryptocurrency, what's the growth rate of TSMC? What's the growth guidance for TSMC for this year? Still in the range of 5%-10%?
We don't have the number readily, but I think that for this numerical answer, maybe Elizabeth Sun can relay to you after the meeting.
Okay. Right. Going forward, I think you maintain the CAGR, compound annual growth rate, of 5%-10%, and that's including crypto or without crypto?
That's including the crypto. As we look at today, the cryptos will be not a factor if you talk about five-year horizon. Things can change. At this time, we don't count on the crypto.
Which means the non-crypto part of the high-performance computing segment, you're becoming more confident. Can we say that?
Yes.
Okay. Second follow-up is on the mature nodes. If I calculate your 0.11 micron process nodes above, assuming that's on 8-inch fab, but I know some of them you use 12-inch fab to manufacture those node. Presumably, 0.11 micron above nodes in the first half this year versus the first half of last year, the YOY is down by 8%, 9% U.S. dollars terms. There's been a lot of industry saying about that 8-inch foundry is very tight, demand very strong, price high, et cetera, but we don't see that reflected on TSMC in the first half this year. What can you tell us what's going on?
It's actually, you're talking about our 0.11 micron and above's revenue decreasing. Actually, we are fully loaded in that 8-inch wafer fab. That's why I mentioned in my statement that we are transferring some 8-inch wafers product into 12-inch wafers so that we can have more capacity to serve the customer. As for the revenue, I did not have exact number in my hand. The wafer pricing is dropping.
Well, it looks like 8-inch is 0.15 and below, and you're talking about is 0.11 and below. I think that's the differences. There are some under-utilization in that part.
Okay.
Yeah.
Okay. You're saying the wafer pricing is declining. Is due to the product mix or like-to-like comparison? Because we heard like some other, your peers are raising price since the beginning of this year, and maybe another wave of the price hike second half.
I don't comment my competitors' behavior. For TSMC, we actually work with our customer, and once we settle down the wafer pricing, we have a commitment. We don't easily change that. Okay? Even we observe the raw wafer pricing increase, we stick on our commitment to our customer.
Okay. Thank you.
Next question will be coming from Deutsche Bank's Michael Chou.
Thanks. For follow-up. The first question is that you mentioned the high-end smartphone units are stronger than expected than three months ago. Can I say that?
Yes, you can say that.
Okay. Is that broad-based or customer specific?
It's broad-based.
Okay. Second question is more housekeeping. What is the outlook for the other segment, consumer, industrial? Because you mentioned crypto and smartphone.
You're asking the third quarter segment? Okay. In the third quarter, we see communication improves, computer decline the most, consumer slightly decline, industrial about flat.
Next question will be from Citigroup's Roland Shu.
Thank you. Just one question. You have a sufficient capital to fund your growth, also, you are generating more free cash flow. Are you considering to do the share buyback? By what kind of like criteria you are considering to do share buyback? Thank you.
Currently, it is not in our plan. Actually, we have done a very thorough study of how do we return cash to shareholder. I think the general feedback is our shareholder prefers cash dividend than buyback.
Now we need to go to the line. Operator, please have the next caller on the line.
Yes. Our first question comes from the line of Mehdi Hosseini. Your line is open.
Yes. Thanks for squeezing me in. A couple of follow-ups. I noticed your wafer shipment in the March and June quarter has been rather flattish, but revenues down by high single digits both in March and June quarter. Should we expect wafer shipments in the second half to be flattish? Or how should we think about the trend in the second half? Again, the trend between wafer shipment and the revenues, Q3 and Q4. I have a follow-up.
Wafer shipment in second half will be increasing, will be higher than the first half, so is the revenue. Did I answer your question?
Does that mean, would your inventories remain the same?
You mean our own inventory?
Yes.
Oh.
Your days of inventory has been going up over the past six months.
Inventory mainly for TSMC's working process. We had 73 days inventory at the end of second quarter. With the ramping of seven nanometer, we expect the inventory will go up a few days in third quarter, will come down as we ship more seven nanometer by end of fourth quarter. Our inventory are mainly working process.
I have a question regarding the longer term trend, specifically on artificial intelligence. Can you provide some qualitative assessment how you see the mix between GPU and ASIC evolving for the AI application?
All right, Mehdi, your question is with respect to future artificial intelligence related ICs. You want us to give you some sort of qualitative descriptions about the breakdown between GPU and ASIC.
Yes. The purpose of asking this question is, I'm just trying to get a sense of how the market trends are evolving and also how I should think about the difference in the die size. I'm under assumption that GPUs generally are bigger dies compared to ASIC. How should I think about this looking forward?
Well, we can only say that AI will be implemented in the GPU, CPU area. Comment on customers or die size, no, we cannot comment on that. It's increasing, we can say. AI going to be used-
Sure
in a lot of functionalities anyway.
Sure. Let me rephrase the question. Would the wafer capacity requirement for AI be different between GPU and ASIC?
The wafer requirement? Or
Wafer capacity requirement.
Wafer capacity.
Does that make a difference for you?
Well-
Yes.
Okay. Let me give you probably not exactly what you want. The biggest portion of high-performance computing today in TSMC is XPU, followed by GPU. Okay?
Great. That's helpful.
purpose.
All right. Now with this very bright long-term outlook of artificial intelligence, bigger die size, a lot of wafers, we will conclude today's conference. Please be advised that the replay of the conference will be accessible within three hours from now. Transcript will be available 24 hours from now, both of which will be available through our website. Thank you for joining us today. We hope you will join us again next quarter. Goodbye and have a good day.