Sumco Corp

TSE:3436

Q3-2025 Earnings Call

[Takayuki Komori, executive]

Thank you for your participation today. This is the results briefing for the third quarter of the fiscal year ending December 2025. Before starting the presentation, allow me to confirm today's materials which consists of 4 items. The consolidated financial results for the 9 months ended September 30, 2025, and -- the announcement concerning difference between forecast and actual figures for the 9 months ended September 30, 2025. The announcement regarding revision to dividend forecast and the presentation deck entitled Results for Q3 fiscal 2025, which we will use now.

Next, a disclaimer. The estimates, expectations, forecasts and other future information discussed here and shown in today's materials were prepared based on the information available to the company as of today and on certain assumptions and qualifications, including our subjective judgment. Actual financial performance or results may differ substantially from the future information contained in this material due to risk factors, including domestic and global economic conditions, trends in the semiconductor market and foreign exchange rates.

We will have presentations today from Representative Director, Chairman and CEO, Mayuki Hashimoto; and Representative Director and Vice President, CFO, Shinichi Kubozoe. Chairman and CEO, Hashimoto, will discuss our forecast and operating environment. to be followed by an explanation of the financial results by CFO, Kubozoe. We have set aside time for a Q&A session as well.

I will now hand over to Chairman Hashimoto.

[Mayuki Hashimoto, executive]

I am Chairman Hashimoto. I will start with an overview of the Q3 results. We did come in ahead of plan. Although it wasn't a big overshoot, the results were pretty good. One factor was a delay in incurring depreciation of JPY 0.6 billion. There was also a JPY 1.1 billion contribution from cost reductions, which was the aggregation of many different elements of around JPY 0.1 billion each, such as improvement in yields. Although sales fell short of plan, profits were not that bad.

On sales, and I think this will be covered later, our standard for recognition is not shipment, but arrival at destination. So when there are delays to sea freight, sales can easily vary by JPY 1 billion to JPY 2 billion in either direction. Basically, you should view sales as having been largely unchanged Q-on-Q.

Turning to the earnings forecast for the fourth quarter of 2025. We project sales to be generally similar Q-on-Q at around JPY 100 billion. Q4 includes the year-end holiday season, which has implications for shipments. Sales tend to fall slightly short every year. However, given some pushouts from Q3, we are projecting JPY 100 billion, basically flat Q-on-Q. On profitability, we are expecting a slight dip given our expectation of lower volumes on the back of regular maintenance during the quarter. Depreciation is expected to rise JPY 5.7 billion Q-on-Q, which is reflected in the projection of an JPY 8.4 billion Q-on-Q drop in OP. Production will also decline as a result of the regular maintenance for a negative impact of around JPY 3 billion. These 2 factors combined largely explain the Q-on-Q change.

Next slide, please. This slide shows shareholder returns. The fiscal year-end dividend guidance has been set at JPY 10 per share for a total annual dividend per share of JPY 20. With regard to determining dividend level, we consider a number of factors. We take into account the demand for cash for items such as free cash flow, EBITDA, the situation for funding for dividends to be paid as well as CapEx.

Of course, we take profitability into consideration. But given we are already incurring significant depreciation and expect to incur more going forward, we expect the profit levels may be challenging until depreciation rolls off. However, from next year onward, I expect cash flow should improve significantly. So from the perspective of cash flow available for dividends, I think the situation should get better.

When I became Chairman and CEO, retained earnings were a negative JPY 82.6 billion. But in terms of cash available for dividends, the combination of retained earnings and capital surplus is now around JPY 360 billion. So even in bad times, I would hope to be able to pay dividends at a level equivalent to bank interest rates, hence, the JPY 10 fiscal year-end dividend.

Next page, please. This is the trend for 200-millimeter wafers. The red line is 2025. Trends remain weak. I think that a recovery will be very difficult. I looked at many different factors, but certainly, the decoupling of the U.S. and China is a factor. China is home to a huge pool of 1.43 billion consumers, so consumption of conventional semiconductors is big. The development of a homegrown semiconductor industry has meant that the Chinese are now producing their own chips as well as packages. This has made it difficult for our customers to export to China and has limited their appetite for wafers. This is the single biggest factor. 200-millimeter wafer demand in China was not that large to begin with, so the impact was significant. I think the outlook for a recovery is not good. We will continue to monitor the situation and implement structural reforms as appropriate.

Next slide, please. This is the trend for 300-millimeter wafers. Despite the general conditions, we are seeing a gradual recovery in 300-millimeter wafers. The trend has caught up to the 2022 level, which was the peak level. At that time, with the previous production facilities, everyone in the industry was operating at full capacity. However, the incremental portion of capacity that was subsequently installed is currently excess to requirements in my view.

In Sumco's case, the capacity we added is all leading edge. So in our case, while new facilities are being utilized, increasingly utilization of our older facilities is declining. Therefore, I believe modernization of such facilities is an immediate priority.

Next slide, please. I find myself making the same comments with regard to market conditions every quarter. Third quarter market conditions were largely unchanged. However, if we compare 300-millimeter conditions to last year, there has been a clear recovery.

On the other hand, 200-millimeter market conditions are clearly depressed. LTA prices are being respected. The outlook for Q4 is similar. In terms of the outlook going forward, for now, we continue to expect strong growth for AI-related chips. The reason why I say this is because leading -edge production at our customers is running at full capacity regardless of whether we are talking about HBM or logic at 5 nanometers or lower. There are shortages in customer capacity here, so the drag on growth is the constraints to chip production capacity. Customers are very focused on increasing capacity. However, mature legacy products have been slow to rebound.

Some of this is a reflection of the general sluggishness of the overall market, but consumer and automotive applications are weak. The transition of 200-millimeter applications like CMOS sensors or IGBTs for automotive to 300-millimeter is also substantially impacting 200-millimeter demand. Of course, Chinese players have developed capabilities in producing relatively lower-end products, so there is competition with Chinese players as well.

Next slide, please. So what is our view of the market? We expect the semiconductor market to get to $1 trillion in 2028 with the growth in the market for AI-related chips accounting for the majority of overall growth. We don't expect to see huge growth for the market as a whole, but growth related to AI will be the driver for the semiconductor market, in my view.

There are 3 key applications using 300-millimeter wafers that are enabled for AI, smartphones, PC and tablets and servers. Starting with smartphones, we expect to see rapid adoption of AI functionality in smartphones, but it isn't necessarily the case that the handsets will incorporate AI chips. Instead, much of the AI functionality consists of transactions with the cloud. So this will not be a big driver for leading-edge wafers, although NAND memory will likely increase. So this only accounts for demand of less than 1 million wafers per month, and we estimated the increase from 2024 to 2028 is only around 100,000 wafers per month.

Next slide, please. For PCs as well, we don't expect much growth, although there is likely to be a dramatic increase in AI functionality. By 2028, the vast majority will be enabled for AI. However, here again, we don't expect this to be a big driver of increased wafer demand. This is because bit growth will be offset by scaling. So as a consequence, there won't be much wafer growth.

Next slide, please. Looking at servers, this is an area where bit growth is expected to be dramatic so much that scaling will not be able to keep pace. So we expect leading-edge logic to grow by around 30% and DRAM to grow around 17% to 18%. NAND, on the other hand, is not absolutely necessary for servers, but we do expect to see volume growth in absolute terms. We are hearing of shortages even in HDDs, so there is some slight growth in NAND happening.

More recently, we have started to see a pickup in activity levels for NAND, but customers are carrying significant levels of wafers. So I don't expect to see an immediate impact on wafer demand, although there are signs of an improvement going forward for memory.

Next slide, please. This is an image of total 300-millimeter wafer market demand by application. By 2028, we expect to get close to 10 million wafers per month. The driving force will be servers in our view. PC and tablet demand volume is likely to be flattish and therefore, not a significant growth driver. The same could also be said of smartphones as well. That said, for the foreseeable future, we expect AI will continue to be the market driver.

Next slide, please. This is the situation for customer inventories, which I have been mentioning. Inventory levels remain elevated at a plateau with no signs of dramatic decline. This suggests that a recovery in demand is not likely to lead to an immediate increase in wafers. However, versus where inventories have been, there has been some improvement.

Next slide, please. This is customer inventory split into logic and memory. Actually, in terms of inventory months, memory is stabilizing. Logic inventory months remain high. The reason for this is because while wafers for leading edge logic like GPUs for 7-nanometer or 5-nanometer and lower are selling like hot cakes, but wafers for commoditized conventional logic at 28 nanometers or 40 nanometers won't recover in the absence of a macro recovery. What's more, logic at these design rules are no longer the preserve of Western countries, but can now also be fabricated in China.

These factors are what is weighing heavily on inventory. Also, although we talk about wafers as if they are interchangeable, the production and degree of difficulty for wafers for 28 or 40-nanometer is very different from leading-edge use wafers. It is inventory of wafers for such conventional products that is proving difficult to work down.

So from our perspective, customers are working very hard to increase their capacity for leading edge by building fabs, but top line growth for us depends on this capacity coming online, boosting production and therefore, revenue. This completes my section of the presentation.

I will hand over to CFO, Kubozoe, to talk about details of our Q3 earnings.

[Shinichi Kubozoe, executive]

I, Kubozoe will present the earnings and outlook in more detail. The results for Q3 fiscal 2025 are shown in the third column from the right and are as highlighted earlier by Chairman Hashimoto. Sales were JPY 99.1 billion, operating profit was minus JPY 1.6 billion, ordinary profit was minus JPY 2.6 billion and profit attributable to owners of the parent was minus JPY 3.9 billion.

In the middle of the table, we show CapEx on an acceptance basis for Q3, which was JPY 17.4 billion. 9-month CapEx was JPY 69.3 billion. CapEx peaked in 2023 and has since been declining. Compared to last year this time, CapEx is down a substantial JPY 100 billion year-on-year. Depreciation, on the other hand, which we show in the line below CapEx, has been rising sequentially. Q3 was JPY 30.6 billion, while 9-month depreciation was JPY 80 billion, up JPY 23.9 billion year-on-year.

The ForEx rate was JPY 147 to the dollar. OPM and other metrics are as shown on this table. We show the analysis of changes to operating profit on the next page.

Next slide, please. Starting on the left, in the analysis of sequential changes to quarterly operating profit, Q3 sales fell JPY 3.8 billion to JPY 99.1 billion from Q2's JPY 102.9 billion. As mentioned earlier, there were some larger-than-expected pushouts to shipments and sales at the end of the quarter, leading to timing differences.

Operating profit fell JPY 3 billion. The waterfall chart below shows an increase in depreciation of JPY 2.8 billion, effectively accounting for the majority of the Q-on-Q change to OP. Sales-related variance was a slight negative, reflecting the dip in sales, but this was offset by cost reductions and the positive impact from a slightly weaker yen, hence, the JPY 3.1 billion Q-on-Q decline in Q3 OP.

Turning to the 9-month results. Sales grew JPY 7.8 billion, but OP fell JPY 24.1 billion year-on-year. The yen appreciated JPY 3 versus the U.S. dollar. As shown in the waterfall chart below, the big year-on-year negatives were depreciation and ForEx impact. There was a slight positive in sales-related variance on the back of top line growth, but this was outweighed by the negative impact of depreciation and ForEx with the combination of the 2 accounting for the vast majority of the year-on-year decline.

Next slide, please. On this slide, I will cover the balance sheet and cash flow. Looking at the middle of the balance sheet, total assets as of the end of September were JPY 1,142.5 billion, down JPY 30 billion compared to the end of December 2024. I will discuss the change in cash and deposits in covering cash flow on the right in a moment, but cash and deposits were down JPY 19.6 billion.

In terms of major items, raw materials and supplies were up on factors such as consumption of polysilicon and ForEx. In contrast, tangible and intangible assets, which had been rising to this point are now falling. CapEx acceptance is now lower than depreciation, hence, the decline as of the end of September.

On the liability side, total liabilities were JPY 493.1 billion, down JPY 22.3 billion. In terms of interest-bearing debt, we chose to cover all of our long-term refinancing needs in 2025 in March at the end of Q1, hence, the increase in debt in first half. However, as maturing debt rolls off in second half, we expect the outstanding balance of interest-bearing debt to decline.

Others under liabilities is down a significant JPY 27.8 billion. The major factor is timing differences between CapEx acceptance and actual payments, as you can also see in net others under cash flow for investing activities. CapEx acceptance is now falling with actual payments starting to take place, leading to a decline in unpaid liabilities. Under net assets, I highlight retained earnings.

As a result of the profit decline and dividend payments, there was a slight decline in retained earnings. Based on this, the equity-to-asset ratio was 51% and the D/E ratio on a gross basis was 0.62x as of the end of September. Both are largely unchanged from the levels as of the end of December 2024.

On the right, we show cash flow. Operating cash flow was a positive JPY 78.5 billion. However, as touched upon earlier, the outflow of cash flow for investment activities was JPY 93.7 billion, reflecting both the decline in CapEx acceptance and an increase in actual payments. The resulting free cash flow was a negative JPY 15.2 billion. After factoring in dividends paid, cash and deposits declined JPY 19.6 billion. With regard to our cash obligations this year, we are tapping into cash and deposits.

Next slide, please. Jumping forward to Page 23, I will now discuss our earnings forecast. The projections for Q4 are as shown on the third column from the right. Our ForEx assumption is JPY 148 to the dollar. We project sales of JPY 100 billion and an operating loss of JPY 10 billion. We project an ordinary loss of JPY 13 billion and a net loss attributable to owners of the parent of JPY 16 billion.

In the next column to the right, we show our full year forecast. We forecast sales of around JPY 400 billion, an operating loss of JPY 4 billion, an ordinary loss of JPY 10.9 billion and a net loss of JPY 16.9 billion. Q4 depreciation is projected to increase further to JPY 37 billion. On a full year basis, we project depreciation of JPY 116.8 billion, up around JPY 38 billion year-on-year. Depreciation has been increasing sequentially. OPM and other metrics are shown in the lower part of the table.

Next slide, please. On this next slide, we show the analysis of changes in operating income. On the left, we show the sequential changes. Q4 sales are projected to increase slightly, but operating income to fall by a significant JPY 8.4 billion. The ForEx assumption is expected to be largely unchanged Q-on-Q. Of the JPY 8.4 billion decline in OP, depreciation is expected to be a major contributor at JPY 5.7 billion. On sales-related variance, we project a negative despite an increase in sales related to regular maintenance at one of our main plants in Q4.

Also related to the year-end holiday season, several of our plants plan to pause operations. Relative to sales, production is expected, therefore, to be slightly lower, resulting in the negative sales-related variance. As a result, we expect a sequential decline in OP of JPY 8.4 billion.

On the right, we show the year-on-year change for the full year forecast. We project a roughly JPY 8 billion year-on-year improvement in sales. For OP, we are guiding for a drop from a positive JPY 36.9 billion to a loss of JPY 4.2 billion, a fall of JPY 41.1 billion. We are assuming a JPY 2.6 strengthening of the yen to the dollar on a year-on-year basis.

On a year-on-year basis, we are also expecting a large increase in depreciation and an impact from ForEx. The combination of these 2 elements is a negative of JPY 39 billion, accounting for the vast majority of the year-on-year decline in OP. Sales is expected to increase, but we do not expect much of an improvement in sales-related variance if we look at the product and customer mix.

We are not assuming much contribution from either sales-related variance or costs. So the big year-on-year declines in depreciation and ForEx are the major factors expected to depress profits. We have provided reference material at the end of the presentation. We have received many questions about depreciation and CapEx, so we have added this page. We show the trends for both starting from 2011, along with our forecast for 2025. Depreciation is the bar on the left and CapEx is shown on the right.

If you look at the dark blue bars for CapEx, you can see that after we made the decision to expand capacity in 2021, there was a sharp increase in investments up to the peak in 2023. While CapEx in 2024 fell year-on-year, it was still at an elevated level. For 2025, it is expected to fall further to around the level we show here. For 2026, we expect further significant declines in CapEx.

In contrast, depreciation has been rising consistently on the back of the ramp-up of the new facilities from 2023 onward. Our expectation for 2026 is that it will continue to rise versus 2025. We expect 2026 to be the peak for depreciation.

Next slide, please. Finally, on Page 27, we show historical trends for sales, OP, EBITDA and EBITDA margin on a quarterly basis. Please review these items at your leisure. This completes my section of the presentation.

[Operator]

We will now open the floor to questions. Our first question comes from Mr. Enomoto of Bank of America Securities.

[Takashi Enomoto, analyst]

I would like to ask you about how you are thinking about profitability going forward. Under Sumco vision, you had indicated that you aim to not incur losses even in a market downturn. I think your current vision states that you aim to maintain stable margins. How do you propose to achieve this? Frankly, this year, you are projecting operating losses and operating conditions are tough. What is your strategy for a recovery and stabilizing profits going forward?

[Mayuki Hashimoto, executive]

In undertaking the current round of capacity expansion, we, of course, did simulations out 10 years based on our capacity once the investments were complete, looking at factors such as LTA-related numbers and prices. At that point, we did not expect that we would fall into the red.

The reason why we are loss-making now is because despite having made investments, our customers have fallen short of their expectations for demand growth. Effectively, they became unable to buy the volumes to which they had committed. With regard to this, our customers have agreed to respect the total volume of their agreements, but only by pushing out deliveries. That is the single biggest miscalculation.

The other miscalculation was that we did not expect the trade war between the U.S. and China to escalate to this level. We have seen a decoupling between the 2 countries. With President Trump coming to power, there was a further escalation driving China toward a strategy of developing its own homegrown semiconductor production capability. While we have not experienced this, our subsidiary, FST, had LTAs that fell through. This is the backdrop to the losses.

Some suggest that we should suspend depreciation. Obviously, if we did so, it would put us back in the black. However, it is our policy to proceed with depreciation even if the facilities are not in use and to work through depreciation quickly. We raised funds based on our plan and are now focused on working down depreciation.

2026 is likely to be a very tough year. But given that we use the declining balance method over a 5-year period, the depreciation burden is likely to ease significantly when we get into 2027 and 2028. We expect revenue should recover at around the same time, particularly for leading-edge wafers. This is our current thinking. That said, 200-millimeter has also been much weaker than we expected initially.

We did not read the market correctly, but it's difficult to predict how this will play out. This is because China has shifted to a focus on developing its own semiconductor capabilities, which has hurt our customers' sales. If our customers can't sell, they won't produce, which has led to a decline in wafer demand.

Also, our operation in the U.S. is facing 2 issues. The first is that there were a lot of exports from the U.S. going into China, including semiconductors. The halting of such exports is impacting our U.S. operation. Also, it isn't the case that the U.S. is buying a significant volume of imports. The shift in the U.S. to producing more domestically has not led to an increase in volume, at least thus far, but exports from the U.S. into China are being significantly reduced.

So at this stage, it is difficult to project the net result of the reduction in exports to China on the one hand versus an increase in business as a result of the U.S. producing more within its own borders. With regard to 200-millimeter, I recognize that we need to make some fundamental changes, and we are considering many things to ensure that we have multiple options, but I don't think we're at the execution phase yet.

We still need to do more in terms of monitoring trends and get a better understanding of cause and effect. We don't yet have enough information to make decisions. In the case of 200-millimeter, book value is already low at this stage, so there is no need to rush to make a decision in any case.

On the other hand, with regard to our domestic capacity expansions, we needed to do this or risk missing out on growth going forward. Without these investments, we would not have been able to catch up on AI demand in my view. So I do think that we did need to make these investments.

Our miscalculation relates to our existing plants producing wafers for chips at design rules like 14-nanometer or 28-nanometer. We do need to modernize our facilities to enable these plants for leading edge. This is what is different from our previous thinking. Once we have completed the modernization of our existing plants, depreciation should be significantly lower and any investments we need to do would be small rather than large-scale investments.

Although we have invested several hundreds of billions of yen so far, the modernization investments would probably be half of JPY 100 billion or so. This should be enough to upgrade these plants for leading edge in my view. This is what we're thinking now.

[Operator]

Next is Mr. Watabe of Morgan Stanley MUFG Securities.

[Takato Watabe, analyst]

On Page 15, you show wafer demand. Despite the current buzz, you aren't expecting to see much growth in 2026 and 2027. If we think about profitability in the next year and beyond, it sounds like you are expecting next year to be tough. Although there are differences between your shipments and demand, can you talk about what sort of order flow you are seeing? Can you also comment on how much you expect depreciation to increase next year?

[Mayuki Hashimoto, executive]

In terms of wafer demand, there is the difference between shipments and real demand, but what we are seeing currently is that there is a mismatch between purchases and input volume. Obviously, the reason for the gap is because customers are carrying significant levels of inventory.

In particular, there is a significant overhang of inventory for logic, especially for 28-nanometer, 40-nanometer and 16-nanometer as well. We are not the main supplier for such wafers. It appears that customers have multiple LTAs in place for the mature nodes. There are also suppliers that can only do these mature nodes. This is contributing to the serious inventory overhang. We are expecting demand for us to ramp up relatively sooner. But in terms of a recovery in profitability, it probably does not happen until 2027.

Depreciation will be quite heavy in 2026, so 2026 will likely be tough. I expect the markets to recover in the 2027, 2028 time frame, so I would expect to see a relatively solid performance in these years. We understand that some of our peers have chosen to pause depreciation. In our case, it is our policy to proceed with depreciation. We have already, in any case, paid for our investments, so we will move forward with depreciation. This approach will mean that things get easier in the future. So for now, we simply need to do what we need to do. It does mean that conditions will be tough in the near term.

That is my view. CFO, Kubozoe, do you have anything to add on depreciation?

[Shinichi Kubozoe, executive]

With regard to next year's depreciation, I can't comment on specific numbers at this time, but I expect depreciation will increase by several tens of billions of yen and we will say that it won't be the high end of the range.

[Takato Watabe, analyst]

Does that mean an increase of around JPY 20 billion to JPY 30 billion as an image?

[Shinichi Kubozoe, executive]

What I can say is it won't be a surprisingly large number.

[Mayuki Hashimoto, executive]

Our depreciation method is declining balance over 5 years at 40%, so depreciation falls 60% per year. We have provided our investment amount in the materials, so you can calculate depreciation. But depreciation doesn't always track in line with expectations, which makes it difficult to comment by year, but this should give you a framework for thinking about it. Depreciation drops off quite quickly since it is declining balance.

[Operator]

Next is Mr. Ikeda of Goldman Sachs Securities.

[Atsushi Ikeda, analyst]

Could you talk about your initiatives and strategy related to AI? Recently, you were recognized by SK Hynix as a best supplier. I believe your position in HBM has improved. Can you talk about what it was that SK hynix rated highly? I believe that you also hold a very high share with the Taiwanese foundry, but can you talk about your initiatives here?

Also, demand for NAND for AI data servers appears to be rising, and we have seen the emergence of CBA. With regard to stacking, I believe that planarization for logic is very challenging, but I think this can be a promising area going forward. Please talk about your initiatives in and your exposure to semiconductor wafers for AI as well as expected mix in 2027, '28 to the extent possible.

[Mayuki Hashimoto, executive]

We are particularly strong in leading edge for logic. There are 2 players in HBM, SK Hynix and Samsung. Historically, we had a very strong relationship with Samsung. In contrast, our relationship with SK Hynix had been distant, but in the last 2 to 3 years, we have become very close because of HBM. We now supply significant volumes.

It appears that Samsung has recently entered into contracts with multiple customers for HBM, so their business is ramping up. So we consider HBM to also be an area of focus for us. Effectively, what is growing now for wafers is logic at 5 nanometers and lower and HBM. So we are very focused on both these areas.

If we look at NAND, Kioxia has launched its CBA. We understand that SSDs using CBI are replacing HDDs and SSDs are in great demand for servers. Stacking of NAND is not super challenging. And in fact, YMTC has been doing this from the outset. It is not hugely difficult but still challenging.

Our impression in the past had been that since it was just stacking and all that is being connected is the power source, it is not that difficult. But in fact, it is quite challenging. It does require relatively advanced wafers. So it isn't the kind of thing that can be done easily by the Chinese. For us, very simply, we will devote our efforts to leading-edge wafers.

AI use logic wafers cannot be produced using conventional equipment to date. We have continued to modify and upgrade our facilities. We have significant know-how on how best to use the facilities as well. Unfortunately, existing facilities cannot be used for AI logic, which is why we need to modernize. So what we are seeing now is a wholesale shift to AI.

HBM is driving the growth for DRAM. We have already been certified for HBM, which I think is very important. What is growing is these 2 areas, we are very focused on both. This is why we chose to invest significantly to expand capacity, although it was a very large investment. If we hadn't done this, I believe that we would have seen our share decline sharply. At the moment, depreciation is very heavy, but it was necessary if we think about our future.

[Atsushi Ikeda, analyst]

If possible, can you talk about how much of Q4 300-millimeter revenue is AI related? I recognize that it may depend on how you view servers, but is it around 15% to 20% of revenue? If you have numbers you can share, that would be very helpful.

[Mayuki Hashimoto, executive]

AI-related HBM is probably around 200,000 to 250,000 wafers per month overall. For 5-nanometer and lower, maybe it's around 500,000 wafers per month, roughly speaking. So leading -edge market demand is around 600,000 to 700,000 wafers per month. In terms of Sumco share, certainly, we have a very high share for logic. So it probably accounts for, say, around 20% of our revenue.

[Atsushi Ikeda, analyst]

You mean it is around 20% of your 300-millimeter revenue. Is that correct?

[Mayuki Hashimoto, executive]

Yes, because price points are high for leading edge.

[Operator]

Next is Mr. Nishiyama of Citigroup Securities.

[Yuta Nishiyama, analyst]

I have a question about inventory. The Q3 wafer shipments depicted on Page 8 show a flat Q-on-Q progression. So as input volumes picks up, it suggests directionally that inventory should be heading towards peaking out. But on Page 16, you show wafer purchases increasing. Can you explain the apparent divergence between the data? Also, what is your expectation for purchase volumes and input volumes in the December quarter? What do you think this implies for when we will see a normalization of inventory level?

[Mayuki Hashimoto, executive]

This is quite difficult to explain. As I have been saying, the reason why I expect customers make substantial inventory adjustments in 2026 is because the companies that have committed to LTAs are globally recognized players. Therefore, they are not fly-by-night players or Chinese players that might simply walk away from contracts. Instead, they honor their commitments.

So up to now, they have continued to respect purchase volumes as stipulated in the LTAs. In situations where they can't buy the contracted volumes, they still try to show goodwill by buying a certain level of wafers, but as a result, have ended up with significant inventory. They rented warehouses to accommodate this inventory. We have now reached a point where there is no more room, but to that point, they have been buying wafers they did not need, resulting in a buildup of inventory.

So given this, we will need to see how this plays out. I anticipate that there will be some very tough negotiations. My understanding is that customers will move to address excess inventory in 2026. Wafer inventory for memory is relatively better. So I would expect that purchase volumes will increase in line with input volumes. But for logic, especially 28-nanometer or 40-nanometer, given the weakness of the players, with the exception of one, I think there is a significant overhang in this area.

[Yuta Nishiyama, analyst]

I see, to the extent that you have visibility, how far along are we in terms of the inventory adjustments to logic? How long do you expect normalization to take? Also, you mentioned that leading edge accounted for a significant portion of sales, and I would expect that demand is strong. How do you expect this to be reflected in your earnings?

[Mayuki Hashimoto, executive]

The most advanced within leading edge should continue to increase. But if you look at what is happening with customers, building a fab that handles 150,000 wafers per month requires a very significant investment and the construction lead time is very long. So it's not really possible for the customers to increase their capacity that quickly. That's why NVIDIA has been saying that the shortages will likely continue into 2026. It's because they cannot increase capacity that quickly.

If we think about chip on wafer on silicon, availability of packaging facilities is also constrained. So the customers are in a position where they have not been able to produce as many chips as they expected. I think next year will be the most difficult. In 2027, you will see fabs start to come online, so chip production volumes should pick up. That's why AI chip prices are 10x the price of a conventional chip now.

Given this, it is hard for me to forecast, but I do think we may be in for challenging times over the next 1 to 2 years.

[Operator]

Next is Mr. Yoshida of CLSA Securities.

[Yu Yoshida, analyst]

I would like to ask about LTAs. When do you expect negotiations for the next round to begin? Also, based on what you have said, it sounds like negotiations are likely to be very tough for mature nodes. For the 3 key categories of leading edge logic, mature logic and memory, what will you be aiming for?

[Mayuki Hashimoto, executive]

First of all, with regard to LTAs, as we have said in the past, the end dates have been significantly extended, so they are now expected to run until 2027 or '28. We have also seen extensions as well for memory, but not as far out as logic. My sense is that we will start to see discussions for memory next year. That is probably the case for logic as well, maybe at around this time next year.

[Yu Yoshida, analyst]

From that perspective, although you may be engaging with logic players next year, the LTAs themselves still have more to run. Is that correct? But for memory, depending upon the contracts, do you think there may be some positive factors that could kick in, in price terms?

[Mayuki Hashimoto, executive]

We aren't yet in a situation where there are shortages, so I don't think prices will rise, but I think the likelihood of prices collapsing is low. I think current conditions are likely to be generally maintained, although I expect there may be a slight increase in memory.

[Yu Yoshida, analyst]

You mean prices would reflect inflation?

[Mayuki Hashimoto, executive]

When I said increase, I was talking about volume.

[Operator]

Next is Mr. Omura of UBS Securities.

[Shunta Omura, analyst]

My question is related to the question asked by Mr. Enomoto at the beginning. I am hoping that you can provide a quantitative response. Sumco typically does not provide full year guidance at the beginning of the fiscal year. Hypothetically speaking, if annual sales were to reach JPY 500 billion next year, what level of OP do you think you might aim for? If you have a rough image, that would be very helpful. This fiscal year appears to be shaping up for sales of JPY 400 billion. Taking that into account and assuming that depreciation increases by several tens of billions of yen, I believe that at the top line of JPY 500 billion, you should be able to achieve profits of JPY 50 billion. If you have a framework that you can share, that would be helpful.

[Mayuki Hashimoto, executive]

With regard to next year, I don't want to mislead you. We typically do not announce full year guidance because it is difficult to provide accurate guidance. Ours is an industry that is quite variable, frequently stopping and starting. That said, for next year, given that our customers are still in the midst of building new fabs, I wouldn't expect them to be able to grow their capacity that much. So volume won't rise that much, but depreciation will increase. Added to this, it is our policy to move forward with depreciation even if the facilities are idle.

So I think profitability next year will be very tough. I would prefer not to provide figures given the risk that it might prove misleading. However, I don't have a sense that we can take a rosy view of next year.

Next year will be a year of hunkering down in my view. In the following year, I think we may see more signs of light. But for 2026, wafer consumption by our customers will not increase that much, especially for leading edge. To grow leading -edge capacity, it is necessary to complete the fabs. This is true for HBMs and for sub -5-nanometer.

Also, although Intel is working very hard, there is still a question about what sort of growth they will get with A14 Angstrom or 18 Angstrom. While the fab construction is complete, the technology is very challenging, so it won't be easy. Next year, TSMC is likely to be the only winner in leading-edge logic. We may see more diversification in players in HBM next year, but even so, I don't think volumes will increase dramatically.

[Operator]

Next is Mr. Miyamoto of SMBC Nikko Securities.

[Go Miyamoto, analyst]

I have a question about demand growth related to AI as shown on Pages 11 through 15. At the beginning of last month, OpenAI announced partnerships with Samsung and SK Hynix and talked about 900,000 wafers per month for DRAM. Based on your experience and also the conversations that you are having with memory makers and family players, what is your view of anticipated demand growth? According to what you show on Slide 15, it doesn't appear that you are taking the suggestion of 900,000 wafers per month at face value, but I think there are other customers besides OpenAI, and there is probably demand growth potential apart from DRAM. What is your view of demand growth driven by AI given OpenAI's talk of 900,000 wafers per month over the medium to long term?

[Mayuki Hashimoto, executive]

Samsung is certainly very bullish and appears to be very confident about HBM. I wouldn't dismiss the comments out of hand, but our forecasts are based on our analysis of data from multiple sources. I would be very happy if OpenAI were correct, and we have a very good relationship with Samsung. What I will say is that we certainly cheer them on in their efforts.

I did think it was quite a large number when I heard it.

[Go Miyamoto, analyst]

Last year, you received a best partner award from SK Hynix, so I hope that you will see volume increases through them as well. Is that a fair assumption?

[Mayuki Hashimoto, executive]

Yes, I expect to see some slight increases. My impression is that memory should be more upbeat next year. In logic, on the other hand, one player is dominating the market, so monitoring their progress is a good indicator. Typically, I'm guessing a single fab requires 150,000 wafers per month. The back-end process for CoWoS is still weak, but it appears that they are expanding here. It will depend on the progress here. Probably, their view is that they -- if they produce, there will be buyers. So it is simply a question of when they can start producing.

[Go Miyamoto, analyst]

I have high expectations for an improvement in profitability.

[Mayuki Hashimoto, executive]

I have high expectations, too. Please continue to support us.

[Operator]

Next is Mr. Nakada of JPMorgan Securities.

[Yasuhiro Nakada, analyst]

My question relates to the question asked earlier by Mr. Nishiyama. Can you explain the 300-millimeter wafer shipment figures shown on Page 8 and the chart on Page 16. Page 8 suggests that shipments are matching the historical peak, but customer purchase in light blue on Page 16 for Q3 is below the previous peak levels in 2021, '22. This suggests that purchases are not at peak levels.

Up to now, I had assumed that shipment volumes were equal to customer purchases, but is that the wrong way to read this chart? Is there an intermediary such as a trading company or distributor that is in between you and the customer that is pulling inventory and is not counted towards customer purchases that can explain this gap? Is this what is pushing up inventory as shown in the green bar? Can you explain the relationship between shipments, customer purchases and customer inventory given that there appears to be a gap?

I would like to confirm the flow to make sure I understand this properly. Also, do the increases in the green bar reflect what is happening in China?

[Mayuki Hashimoto, executive]

Actually, the sources of data underlying Pages 8 and 16 and 17 are different. The data for Pages 16 and 17 is the aggregation of what we have heard directly from our customers. The data is only from our main customers. For Page 8, we know the overall picture, but recently, we have found that it sometimes includes China, but sometimes not. So we don't really know how the China data is being handled.

At a minimum, China itself does not disclose statistics of how much and where they have sold 300-millimeter. So my sense is that the data on Page 8 may be skewed. However, the data we show on Pages 16 and 17 is based on what we have heard directly from our customers, so it could be that the data may be conservative, but in any case, the data on Page 8 does not match the data on Pages 16 and 17. These charts are really only indicators of trends. The data for the 2 sets of pages do not match.

[Yasuhiro Nakada, analyst]

I see. So in terms of thinking about Sumco's earnings, we should look at Pages 16 and 17 because it is based on what you are hearing from good customers. Is that correct?

[Mayuki Hashimoto, executive]

Yes. What you see on Pages 16 and 17 are directly reflected in our earnings. The data underlying Page 8 seems to be slightly skewed recently. China is producing 1 million wafers per month, but we don't know what is or isn't included here because those wafers are not being sold in markets outside of China.

There is an estimate that of the 1 million wafers being produced in China, the 200,000 to 300,000 are test wafers, which might be included in this data. This is something our salespeople have heard from customers, but more recently, we have concerns that the data we are collecting may be skewed as well. We actually stopped production of long-term data on wafer demand because the numbers don't match up.

China's presence has grown, but there is no disclosure, so we don't know. However, we have very strong relationships with the top 3 global players, so we are able to collect information throughout these firms at all levels. But even so, the numbers increasingly don't match up. To reiterate, the numbers on Page 8 do not match the numbers on Pages 16 and 17. Please use these charts as an indicator of trends.

[Operator]

Next is Mr. Okazaki of Nomura Securities.

[Shigeki Okazaki, analyst]

To date, you have talked about Chinese wafer makers, but my understanding is that while they are supplying 300-millimeter to local Chinese chip makers, they are not a significant competitor for Sumco.

[Mayuki Hashimoto, executive]

For now, yes.

[Shigeki Okazaki, analyst]

Is that situation still unchanged? So for instance, you explained that the Chinese wafer makers are not reflected in the charts on Pages 16 and 17, but hypothetically, even if they were, this would not impact Sumco's earnings. Is that correct?

[Mayuki Hashimoto, executive]

That's actually a little difficult. We are actually selling furnaces to China. We analyze the data from many different perspectives. So for instance, you have YMTC in China. They are consuming several hundreds of thousands of wafers per month, but they source virtually all of their wafers from Chinese players. But in terms of how much of the wafers produced in China are sold in markets outside of China, when we have conducted interviews, it appears that there is a company that is exporting test wafers, but we estimate that it represents maybe around 20% of total Chinese production, assuming that China is producing 1 million wafers per month in total.

So for now, I believe the impact is not that large. Also, our customers cannot buy Chinese wafers from a risk perspective, and we understand that they do not find Chinese wafers attractive given that you generally get what you pay for.

That said, Chinese chip makers are using China-made wafers by the order of the government, so yields are not an issue for them.

[Shigeki Okazaki, analyst]

So currently, you are not competing directly with Chinese wafers, but we don't know about the future. Is that the correct way to understand the situation?

[Mayuki Hashimoto, executive]

For leading edge, we still have a significant lead. For non-leading edge, I would expect Chinese players to emerge in the medium term. It's also a question of what happens with geopolitics going forward.

[Operator]

Next is Mr. Yamada of Mizuho Securities.

[Mikiya Yamada, analyst]

I actually would like to just confirm 3 numbers with you. We have been talking a lot about leading edge, but when you say there is demand for leading edge of 600,000 wafers per month, I understand this to be 7-nanometer or lower. Is this correct?

Second, with regard to the expansion we have seen to this point, I understand it to have started in the fiscal year ended December 2019. I believe that your total CapEx from 2019 to the present is around JPY 880 billion. Has all of this been used either to invest in leading edge or for the modernization of existing facilities?

Third, next fiscal year, CapEx will fall further. Given that 9-month free cash flow for this fiscal year is already in the black, can we assume that free cash flow will grow from here and that this growth will be possible with the facilities in which you have already invested JPY 880 billion.

[Mayuki Hashimoto, executive]

For the first question, you are correct. When we say leading edge, we are referring to 7-nanometer or lower. On CapEx, almost all was invested in new facilities, but it isn't the case that the remainder was invested in upgrading existing facilities. There were some investments where we were building exact copies of existing facilities. This applies to our foreign joint venture. The JV has strong relationships in China. There isn't that much demand for leading edge, so the investments there were basically replicating existing facilities to expand capacity. This accounts for around JPY 100 billion of the total.

In hindsight, I do feel that we need to modernize even though the facilities are new. For the other investments, I am pleased with what we have achieved. Third, on free cash flow for next fiscal year, at a minimum, CapEx will fall significantly.

I will hand over to CFO, Kubozoe.

[Shinichi Kubozoe, executive]

On free cash flow, Mr. Yamada referred to Q3, but on a quarterly basis, it can be lumpy, but free cash flow should improve from second half. For next year and beyond, we should see further improvements. The level of modernization investments we are contemplating can be covered by the free cash flow we expect to generate, and we will be disciplined in controlling investments into the medium term as well.

[Mayuki Hashimoto, executive]

Even if we do all of the proposed modernization investments, it won't require JPY 100 billion over the next 3 to 4 years. On an annual basis, therefore, CapEx would consist of maintenance CapEx and modernization investments. We have completed our greenfield investments and aren't expecting to do more. Also, at this time, it is not the environment for expanding capacity further. We started out by accumulating cash, then raising equity from the market and expected that we would not incur further cash needs.

We did come up slightly short, so our borrowings increased slightly, but we don't anticipate big outlays going forward, given that we have largely completed our payments. So I would expect to be free cash flow positive next year.

[Mikiya Yamada, analyst]

The modernization investments of JPY 100 billion that you referred to is the cumulative total for the 3- to 4-year period. Is that correct?

[Mayuki Hashimoto, executive]

We won't need JPY 100 billion. It will probably be about half that amount. And yes, it is cumulative.

[Mikiya Yamada, analyst]

Congratulations on achieving positive free cash flow for the 9 months of 2025.

[Operator]

We will end the meeting here. Thank you to everyone for joining the Q3 fiscal 2025 results briefing. We are grateful for your participation today.