Good afternoon, everyone. I am Jeff O. Weiler, Head of Research for Macquarie, Taiwan. Very happy to be co-hosting Phison's third quarter 2025 earnings call, with a lot of interesting stuff going on in memory, especially NAND these days. Always very happy to hear from Phison's CEO and founder, K.S. Pua. I think we'll start off with K.S., who will do a presentation and answer some of the previously asked questions, then we'll go into Q&A from the audience. With that, K.S., I'll turn it over to you. Thank you.
Thank you, Jeff. Thank you, everyone. Good afternoon, good morning, ladies and gentlemen. Thank you for joining Phison 2025 third quarter's earnings call. Okay, we start. Okay, the revenue breakdown of Q3. Controller-wise, we improved from the ratio of 25% up to 30% just because we have made a lot of design win in the client SSD for PC OEM, NAND company, and also the mobile company's controller business. We hit a very strong gain. We gained a share in both client SSD and also mobiles. In industry, industry is a start to debounce, and Phison for many years we made many design wins, and we are now enjoying the return from the last few years' investment. In the embedded ODM mainly is for the mobile, IoT, and also PC application. This also we have a strong improve in this business.
Enterprise-wise, CQ3 keeps fresh just because of, they still have some inventory in the market, but recently all the NAND companies start to, increase their selling price almost up to 50%, 75%. The demand is turned to very strong in the CQ4. Revenue-wise, CQ3 is, NT 18 billion, all-time high. Gross margin is, NT 5.8 billion, also is all-time high. This is our revenue, so accumulates across the NT 50 billion in the first three quarters. Gross margin-wise, 32.4%, which is a turn to better than this year's CQ1 or CQ2 just because the market turned to stable, and also we have some inventory accumulated from the last year, CQ4 and CQ1. The demand is turned to very strong after, mid-August when the cloud start to, pull in the, demand and also rush orders.
Right now we're also facing a lot of strong demand, but also we need to settle down the supply. The earnings per share is NT 10.75, which we have some inventory reversal and also the foreign exchange. I think we all know the CQ2 this year we suffer actually quite a lot on the foreign exchange, but slightly, this is coming back, but not 100% back yet, but slightly is coming back. This helps to our earnings. Income statement, gross margin is all-time high, and the R&D expense actually we also increased just because CQ3 we have a tap out our new 6 nm controllers, which is for the PCIe Gen 6 and also PCIe Gen 5 controllers. This R&D expense increased.
Also, due to the profit increase, we also write off some certain amount of a bonus as a compensation for the staff. Overall, the earning 10.75 per share is the best in this quarter. Of course, I think the best is coming. CQ4 definitely is better than CQ3. This is our performance in our CQ3. In the balance sheet, everyone may be curious about inventory. We still keep a high inventory, NT 31.5 billion in the end of CQ3. We are starting to manage and control the shipment and also to control not to take that many new orders. The market actually has turned to crazy in the supply side. We have to manage really, really carefully and also make sure all the design-in customers are happy to get what they need.
The supply side is really cutting the supply to not only the module house, but to everyone, even to the cloud company. NAND TIF. Phison, we issued the stock option by, I think it's by end August. It's a slightly, slightly different. NAND TIF, we gain around, it's a little bit higher than the TIF, but from CQ4 every month we need to write off the options compensation. Here's why we need to give a good incentive to our employees. October revenue is NT 7 billion. Again, it's an all-time high by monthly and accumulates NT 57 billion.
October we start since actually since August we start to manage not to take that many orders and the customers are no matter from retail, from enterprise, from PC OEM, from design-in mobile, everyone start to ask for big shipment, but we have to manage our inventory and also manage our supply side. We are very optimistic in the coming months. The demand is still strong and this will deflect to our revenue numbers and also the profit. Okay, this is some highlight here. Client SSD you can see Phison's, our controller shipment actually, growth, 72% QoQ and the 380% YoY. Means in the last few years all the Phison, our effort, R&D investment, design-in start to get the return. And revenue-wise also has a big growth, 66% QoQ and 350% YoY.
We gain almost every major NAND maker and the shipment actually by October we also facing some controller side supply. Next year TSMC 6 nm supply also tight. We have to secure what we need for next year's shipment. By all year modules we also growth, by year-over-year is 2.5x and then 2. just because of strong design win and the strong demand and we gain the trust from mainly the PC OEM no matter Tier 1, Tier 2, Tier 3. We start to ship our Gen 5 modules to the Tier 1, Tier 2, and Tier 3. Unfortunately recently price is tight. We have to manage the shipment. In mobile-wise controllers, year-over-year growth 65%, quarter-over-quarter 9. Revenue also is 120% just because we gain the more UFS project from the NAND company and also our NAND partners.
This is helping us to grow our revenue instead of unit growth 65%, but revenue 120%. We believe Phison now is the major, the biggest, share in the phone's controller in the storage. Module-wise also is a big growth, 290% YOY and the shipment 290% and revenue also 85%. You can read here almost all the key player in mobile, in IoT, in industrial we got all design win. Now our challenge is how to make sure to fulfill their demand. Enterprise-wise, we work very hard. We gain a lot of design win every year. We also in the three of the top five AI optimized storage software company and start to ship from August. We also secure good amount of SR by NAND supply in the 2026 to fulfill this design win.
We believe next year Phison Enterprise SSD revenue will be very significantly a good number in Phison's total combination. We keep working very hard not only in the ESSD but also work closely with the AI-related company. Most of our ESSD go to the AI optimized storage. We just tap out a new controller, PCIe Gen 6, which we believe we are able to have silicon by January next year. We are going to have a CS sample by CQ3 next year for the key customers who are asking for PCIe Gen 6 interface. We also believe, in enterprise SSD, Phison started to invest from 2018, over seven years. We are from behind two and a half years now. We are going to be ahead in the market.
This also to show with the heavy R&D investment, Phison can be the key players in the coming futures. By the way, Gen 6 controller we already have, already have one project ongoing with the GPU makers to customize this Gen 6 interface for the storage systems. Many investor asking in enterprise what kind of a customization Phison able to deliver. We give example. This is one project we work with the server company, storage server company. They're asking for mixed mode of the storage applications, which we engage this project from August. We are going to deliver the samples by end of December. This to show you how Phison doing customization. We are not only provide the controller because let's say if a customer asking for 70,000 piece a year, 70,000 piece a year controller, $50 is nothing.
Phison is providing the modules. If 70,000 pieces of 32 TB modules, then this is a good amount of revenue and also the profit. This is the one we show to everyone. We are almost ready to go to do customization for the enterprise SSD, no matter for the server supplier, CSP, and also the AI system suppliers. In this quarter we also have some very good significant design in Phison's first worldwide first Gen 5 derender solution. We build design win into the world leading robotic company. This is already in the system and we get very good response from our design win customers. In our Gen 4 4 TB, we are already in the smart street light systems. Also, our very strong SATA embedded already in the medical ecosystem. In automotive, again, you keep hearing invest huge amount.
Excuse me, I think, please mute if you are talking, okay? Automotive, we invest there and the market is there. Phison also gain a lot of share. Okay, sorry for interrupting. Phison also gain a lot of design win in the EV system. We just have a picture here to show. This is the automotive application and this is Phison's product line. You can see we are almost in every kind of application. Phison is a, we can say Phison is a company. We have a comprehensive storage system for any kind of automotive systems. We are not only providing controllers. We provide controllers to the NAND company. We also provide design service, the modules, and also the full turnkey to every single automotive, and also Tier 1, Tier 2, Tier 3 manufacturers.
This is our strength in the automotive industry. Portable SSD, we want several of the NAND company and also start to ship. In the memory card, we also have a good in the surveillance and also have lots of in the gaming. Unfortunately, right now it's a shortage, but it's why is. We also have a new controller for high bandwidth, high capacity. This is mainly for the YouTuber and also for video recording. Go to the retirement and re driver. Re driver we believe Phison is a bigger supply in the world. Of course, the TEM is small, but now we have a very big share, over 60%. Going to Gen 5 retirement. Recently we made our design into the B300 platform. We also got a good amount of order. Unfortunately, recently the supply is tight.
We are working very hard with TSMC to pull in the wafer to fulfill the demand. We can see the AI inference is coming. Every cloud starts to build a data center for AI inference application. In Phison retirement, we get several design wins into these systems. In aiDAPTIV+, again, we worked for many years. We finally design win. We got the order from banking systems, medical systems. We have more than 200+ POC ongoing and we got around 100+ POC orders. This is a first order and we believe repeating orders are coming. We also strongly promote the AI training PC to the education systems. In Taiwan, they have over 20 universities start to deploy our AI TPC in the class, try to let their students be able to practice their AI skill.
We made the AI TPC down to $4,000 that every student can afford to buy. If they are interested, they can buy a set in their home to practice their skill. This can train a lot of new talent to service in the market very, very soon. We also work close with Intel and AMD with their iGPU and the APU. We are going to convert every PC, which has an iGPU or APU, to the AI PC without any hardware change. By coming CES, we have many PC Tier 1 OEM and ODM going to show Phison solution aiDAPTIV+ into the Intel and AMD platform. This will help the AI PC become popular with the cost, very competitive because no hardware change is needed.
I also been invited to the GITEX, the biggest AI show in the world, to give a talk about how to deploy the AI on premise. We got very good response from the Middle East potential partner. Also, we've been asked to have a joint development and joint venture to develop special storage system and AI system for Middle East market. This is ongoing. R&D expense. First half, the profit is not that exciting. We manage our R&D expense, but from CQ3 the profit turned to good. Then we start to improve the R&D expense to make sure Phison are able to leading more. Imagine three, four years ago, I keep saying Phison has a heavy investment. Of course, some shareholder complaining spent too much money on R&D.
Now you see we get the major share in the PC OEM, major share in the mobile. Coming soon in enterprise, in AI, Phison will play the very, very unique role. This is all coming from the R&D investment. Market turned to good, Phison turned to profitable, very good profit. We are going to again highlight we have to invest more in R&D to make sure Phison will be the winner in the AI ecosystems and the enterprise business and also in every customized service in the storage. Summary here, I think no need to mention the NAND is turned to very, very tight. I believe a CQ2 earning call I keep highlight, but of course some of investor do not believe, but it is what it is right now from a media you can read. I do not need to mention that that much.
In the R&D expense, we believe the NAND company this year may not announce the expansion, but I believe after next CQ1 earning call they have a very profitable business. They have to start to expand and build a new fab to improve the supply. One more thing is, no matter what, DRAM now is a big shortage, next year also shortage. We worry the shipment of a smartphone, PC, all the end device may not able to maintain because DRAM shortage. We believe PC company, they have to reduce the every PC DRAM capacity. They have no choice because DRAM tight. When you decrease a DRAM capacity, you definitely get a complaint from the end users. Phison provide the aiDAPTIV+ solution with the limited DRAM.
With Phison solution, the AI inference performance can be improved and we prove that. Next year's aiDAPTIV+ with the Intel AMD will be there. NAND controller shortage. We get very strong support from TSMC, ASE, SPIL, and also KYEC since they treat Phison now as a very strategic customer and partner in the storage. Thanks to them helping us to make everything shipping smooth. Even though we got the NAND company upside, in two, three months, we still able to fulfill. Again, thanks to TSMC, KYEC, ASE, and SPIL. In the enterprise SSD, we are gaining more design win. By the way, since market turned to very tight, a lot of server makers and the system integrator come to Phison to ask for supply and to go to the ABL. We are going to get more ABL by CQ4 and CQ1 next year.
Mobile wise, now we are the biggest controller supplier to the mobiles, and we believe we can maintain this position for many years. In automotive, we win the business, and we are going to gain the most share. aiDAPTIV+ education wise, I think we are start to have a good, training to students. Also in the PC base, aiDAPTIV+ with the iGPU, HAPU is going to be popular after next CES. In the server application, we have a more POC ongoing. This is today's brief and few questions. Let me finish, then we go to live. This flash will sustain for how long? I had an interview from [Tianxia Zhazi], and there was a mentoring. I do not expect they go to translate to English, but they did.
Making a lot of, question from the market, from supplier, why I'm saying 10 years. Ten years is just a description, okay? This time the NAND just caused by the AI inference. AI inference, the CSP getting that return go through, through the AI inference. Inference, the output is a data, data need to store. I make it simple summarize. CSP, AI business, their revenue income linear to the capacity of a storage. They want to have more revenue, more income. They need to have more storage. Hardware deplacement is happening, okay? We believe NAND company, they're not willing to expand the fab until now because in the last five years from COVID until now, the total profit in the NAND still even. They're not making any good profits, but from now on they start to get profit.
Next year they start to announce the new fab, but the output can happen only by end 2027. We believe the cloud AI, the edge AI going to need more storage and hard drive deplacement is happening. NAND will be supply tight for many, many years. Inventory wise, actually consumer business to Phison are reducing a lot. I give example today, we have a white industrial customer come to me to ask for 75 pieces, 75 piece, okay, piece of the storage for the system. They can't ship anything without this 75 piece of a storage. Market is going to be very challenging. Phison has to cut the supply to retail, get those components, then go to the industrial design in PC, mobile, and also enterprise.
We are focusing more in the enterprise, industrial, embedded, ODM, and of course controller wise, we are going to gain a share. ESG wise, I got the new report because we keep being asked, what is Phison's performance in ESG? This is a report from some analysts. Phison versus the key IC design Taiwan, MediaTek, Realtek, and Novatek. Looks like Phison still doing good, better than our industrial in the ESG. Above is today's report. We can go to live.
Great. If you have any questions, you want to raise your hand, we'll call on you. I guess maybe I can start with a few questions on inventory. Can you talk a little bit more about the cost basis for the inventory? I know you had some write downs and some reversals. Where is the current, you know, rough cost of the inventory versus like where market prices are right now?
Oh, this is, may difficult to answer, but, give you an example, okay? One of a NAND supplier, the NAND, the one Tb TLC price was in July was around $4.80. Yesterday they asked $10.70 and no choice replace order. Just only three months. Talking about low density of the MLC, which Samsung is going to EOL, we still have a few 10 million pieces on hand. The cost is around $150, but the market price go to $380. I'm hard to answer what is, inventory cost versus market price, but we need to take this to make a good service to our customers.
Okay, great. I guess in the past calls when you talked about NAND prices going up and supply demand being tight, your biggest concern was your ability to supply enough NAND for your customers. You also commented early on in your talk today that you actually are pretty comfortable about next year. Can you talk about how much NAND you secured for next year?
Okay. Actually, we, by slow season, May to July, we have some discussion with our NAND supplier. We already secure NAND supply in the next year. Of course, no matter how many petabyte, exabyte we secure, still not enough, right? We secure some good number of the NAND for next year. At this moment, we still go to talk to our NAND supplier. We need more.
How we convince the NAND supplier to allocate more to Phison, I just showed them what is the value of Phison. Phison helping them to design win, design in every application in the market and to make sure the demand will be there. We can help them to gain market share. This enterprise, we gain the good supply from the three NAND suppliers. They believe Phison Enterprise PASCARI will become one of the, I'm not saying key player. We are third party. Can be a minor player, but in the third party, we can be a very key player. They support us good enough of a NAND flash because we prove we can bring that component, go to every AI system, go to the server system, go to some customized systems. Summarize is we have secured NAND supply next year, but not enough.
We still need to fight for next year.
Okay, great. I think Donnie from Nomura has a question. Donnie, you want to unmute yourself and ask?
Thank you, Jeff. Can you hear me?
Yes, please go ahead. Please.
Yeah, thank you. Hi K.S. Thank you for taking my question. My first question is regarding, yeah, probably a very stupid question, but I'm just curious whether, I mean, the NAND makers, they can convert, for example, their TLC production to QLC production in case that the enterprise SSD demand is very strong and whether they can quickly react to this kind of sudden surging demand from the server and also, potentially, you know, the PCs or smartphones demand next year could be weakening if the memory price continuously trending up.
I'm just curious about whether they can change this demand supply dynamic in a very short time. The second question is regarding your CB. Previously you have announced the plan to issue the new CB. I wonder if you can give us some detailed updates and also what's the, you know, potential pricing. Yeah, any idea would be appreciated. Thank you.
Yeah, thanks, Donnie. I think right now every generation NAND, every generation the 3D stacking, right? When they move to the next generation, they need to buy the new equipment because of a structure. Okay. When they move to new equipment, they need to have a CapEx. Convert from TLC to QLC, I may not be the right guy because every NAND, six, five NAND manufacturers, they have their own unique technology. Again, from TLC to QLC, they need equipment first.
Second, if they can convert no matter what, every NAND manufacturer told us 2026 sold out. All the capacity sold out. No matter they can convert or not, it is nothing help because capacity sold out. They try to improve. What they can do right now is improve the cycle time, improve the EO rate to gain a few percent of output. What we need is a 2x,3x of output. This is not going to happen 2026 because this need new fab. Okay?
Yeah. Okay.
Talking about CB3, we got the CB3 approval from authority. The time market price was around NT 400, NT 500 something. Recently the stock price deflate while the market, you know, tight. One thing is, Phison, we invest a lot in the adaptive. I believe Phison's value still not be recognized by aiDAPTIV+ because it is not turned to revenue yet.
The challenge of a management team is we do not know how to issue the CB3 with the price. If we issue the price today, after a few more months, if the market gives us a good valuation, I believe the current shareholders will make, you know, really negative comments. We decide to extend CB3 to December, February. We have to monitor what happens in the market. When market goes to stable, then we, no matter what CB3, we still have to issue. Phison needs to gain, to get more cash flow to prepare for our next growth. Hopefully CB3 we can decide by mid-December to February.
Okay. Yeah. Understood. Will this delay cause some of the, you know, our cash flow insufficient in case you want to accumulate more inventory?
Very good question. Of course, after mid-October, everyone wants to buy inventory, but NAND supplier keeps saying sold out. Okay. We are not that rushed at this moment, but we still need to accumulate the cash flow for our next growth in the business.
Okay. Because one of the questions being asked by some of my clients is like whether Phison can procure more NAND than other competitors. My answer has been always yes, for sure. Phison has, you know, always had a good relationship with the memory makers like Kioxia, for example. Just, yeah, just wondering how confident that you can secure the NAND supply in the coming quarters.
Okay. Talking about NAND supply, I give example, okay? There are six NAND manufacturers in the NAND supply in the world. When they are facing some market difficulty, they call Phison, they ask Phison, Phison we support. We have a mandatory record there, okay? Second, of course, right now memory is a tie, but NAND supply will, when they allocate the flash either to CSP, PC OEM, phone makers, module house, they will think if I am supply you more 2026, how can I make sure 2027 you are going to buy more? This to show the value, the supply chain, the product lines to them, right? I give example, we have three NAND company, they are willing to support enterprise QLC to Phison just because they believe, by the way, we prove that we made the design, win design in, and we are helping to gain more share to them in 2026, 2027, 2028. How to gain the supply more? First of all, we need to make them trust Phison we are capable to gain more business for them. And we prove.
Okay. Understood. Thank you, Kes.
Yeah. Welcome.
All right, Kes, we have an online question here. What's your target sales percent for enterprise SSD next year?
Definitely will be go to the two digit, 20%,30% depends on the flash memory supply for other client business. For sure we gain good enough of a supply for enterprise SSD next year's. Okay.
Okay, great. Also, does the company currently have plans to develop the HBF controller? What is the current progress?
This is actually making a lot of confusion. What is a HBF definition? Definition HBF is putting the flash memory solution into the HBM bus. We talk to GPU makers, we talk to system makers, we also talk to the application customers. Flash memory has two constraints. First, DRAM do not have a life cycle, but flash memory has a life cycle.
If you put the flash memory as a corpus to the GPU, when flash memory runs out the life cycle, you have to throw the GPU board out. And the GPU is not single dollars, it's a few $10,000 . This is a question. Flash memory into the system, the best way is plug and play. You need to swap. When the flash runs out the life, you can replace the new one. Go through the corpus with the GPU, I don't think this is workable. We also consult many experts in the GPU company and the flash company. I'm really confused, why so confident to put the HBF to the HBM bus. Second, DRAM operating temperature is 125°C . Flash memory is 80°C- 85°C. The GPU board, the thermal is very high. This is a challenge.
Again, we believe all the AI systems in the past, they use DRAM, no matter HBM, low power or anything. DRAM capacity, small DRAM supply constraint, DRAM is too expensive. Flash memory definitely will be in the AI systems, no doubt. What kind of an interface form factor, somebody saying HBF, Phison promoting aiDAPTIV+, we believe with the constraint of flash memory life cycle, thermal, PCIe bus is the best solution or can be only solution because PCIe bus is mature in the hardware and also in the service stack. We believe aiDAPTIV+ will be one or, or maybe unique, a print system in the AI system. It's hard to mention this time moving and time will tell. We believe aiDAPTIV+ is the best in the AI systems.
Okay, great. Another online question. Do you have any forecast for NAND price hikes in 2026?
Oh, this time is difficult, very difficult. Okay. Give you an example. If flash memory price going 2x, system makers such as mobile, PC makers may cut down 50% the capacity. For example, a smartphone using 256 Gb, price go 2x, they cut down to 128 Gb. The demand suddenly drop, the price going down. I mentioned that for the AI inference systems, CSPs income linear to the storage capacity. They are not going to cut capacity. If they are not going to cut capacity, they have to keep invest because AI inference generate only data, net need a storage. Will be really difficult to set what is a price level.
We hope, we wish, we ask, we request if the NAND company enjoying the good gross margin, 50%,60% good, do not try to go to 80%. This definitely will kill a lot of industry. This is not help to industry. Of course, I am, we Phison always keep saying that we need to help the NAND company to have a better gross margin because they need to have a heavy R&D and the Capex investment for the next expansion. Without next expansion, we cannot get a good enough of an amount of supply and also the good price, in the investment. We need fresh company to make good profit, make the new investment, make the more supply. If gross margin go to the HBM level, this may kill the industry.
Okay, great. I think we have a question from Simon Wu. Simon, do you want to unmute yourself and ask your question?
O kay. Yeah. Great. Thanks. Hi, K.S. The number one question, you have a NT 31 billion worth inventory. This is just the book value or you revalued your inventories for Q3?
This is our book value.
Oh, so if the NAND price rises, you do not have to adjust the revaluation?
No, no, no. We not play this game, okay? We still keep as a cost.
Yeah, great. And secondly, the NAND demand these days is more for enterprise solution, right? ESSD rather than Client SSD?
Answer yes and no. Actually, the demand in mobile and the PC is still very strong. Just suddenly the AI system need more storage for AI inference. You feel all looks like the demand coming from AI, true, but not true yet because the PC also has a really good amount consumption on the NAND.
Yeah, that's good. In that case, let's say here, the NAND makers, Japan, Korea, U.S., if they can do better job with the OEMs, so they try to sell more their own NAND solutions than Client SSD, ESSD, maybe using, you know, their own controller or otherwise controller. My concern is over the next few quarters, if the NAND demand getting better, supply tight, then not sure whether you can easily get the NAND chips from the NAND chip makers because they want to sell their own product.
Good question. This is a wish of every NAND company. They wish they can supply everything. If you want, you wish to supply everything, you need to hire many engineers, then return to your operating expense, right? Mm-hmm. Every company, even Samsung, Hynix, they're good profit, but they still have a resource constraint. They need to put their limited resource to have a much high value return. The market still needs a lot of customized system in the NAND. You need controllers, right?
Yeah.
That is why Phison can survive until today. I give you an example. Other than the five NAND companies, China YMTC started to build their own controllers, but they said they are not ready yet. Mm-hmm. The five NAND companies have many in-house controllers, but they still outsource controllers to third party, right?
Other than the NAND company, third party who are capable, I mean, I repeat, capable to design from concept to a product, from concept define the spec, silicon, ASIC, PCB board, firmware, software, application, validation, the full systems in the world today, only two company. They are Huawei and Phison. Mm-hmm. I repeat, from nothing to a product, not saying, I get the controller, I license Synopsys, I'm doing this, doing that, I outsource. No, all in-house development from nothing to product, Huawei and Phison. This is our value. NAND company that need either Huawei or Phison to service a lot of small, medium size NAND applications. This is Phison's value.
Yeah, yeah. A no Kingston then?
No, from nothing to products, you need to have an IP designer, file designer, NNO designer. Kingston is buying controller assembly.
Yeah, yeah. Very quickly. If the ex-China NAND chip makers' NAND production tight, NAND price rises, how about you get the more NAND chip from China, China memory chip makers?
Phison, we have a business, the supply from the six NAND manufacturer including China. Phison also played a certain role in China. We have a JV in China. This company now doing great job in the mobile business, also in their some retail business in China. We also have enterprise solution in China, mainly service to the server and the some CSP in China. China asking to use China. So we have our value also. Again, I keep repeating, third party are able from nothing to systems only to Huawei and Phison. So they need to.
You think a Chinese NAND chip supply will grow further and then you can use made in China NAND chip for your business that will grow further and further?
Before this tight supply, only China manufacturer expand their factory. They are getting the share. China has a demand. We have been asked to use China flash, so we have to.
Maybe lastly, sorry, did you say DRAM shortage or which can continue? What is your view? You are saying DRAM shortage and then the current spot market price is very, very high. You think, like $20, $30 DRAM spot price is really realistic?
No, no, it is possible. I have no position to say DRAM shortage. I am not in the DRAM. From my perspective, my insight, DRAM is shortage. DRAM is a really shortage.
I don't need to mention, right? DRAM shortage, what impact to the NAND flash? We found a good formula. PC next year, if you want to keep same amount capacity of a DRAM, you cannot ship equivalent unit next year. No way. You have to reduce the DRAM capacity in every single PC. If you reduce, PC customer definitely complain, right? Phison promoting our aiDAPTIV+ solution improve the AI inference in the systems. If you reduce the DRAM capacity, you save the cost to get Phison aiDAPTIV+ as an SSD and AI inference, you can sell better PC with performance, but the cost is less than the past.
Even for the general PC, like $500 PC.
Yes. Phison are able to convert the PC if they have the so-called iGPU or AG APU, we can convert them to AI PC immediately without any hardware change.
Okay, great. Thank you, sir. Appreciate your great explanation. Thanks.
Okay. I got a few online questions here. Would BT substrate supply be a production bottleneck? Are we able to pass through the substrate price hikes, if any?
Yes. Substrate now is also one of big heads other than NAND. We have some customers, they rush order. We have a flash, we have controllers, but we do not have a substrate. Since July, we work really close with our substrate supplier to secure our allocations and also some upside flexibility. We believe with Phison's position in the market, we are able to convince substrate company to give better support to Phison.
It's a challenge, but I don't think it's a big problem to Phison.
Another question here. Costs are rising throughout the supply chain. Is TSMC raising prices and does TSMC have plenty of 2026 capacity for Phison's needs?
Again, he cannot mention what is a TSMC price, but we give a hint. The TSMC advanced node is very tight in supply, very tight. If we move to the most advanced node, the supply is not able to fulfill and the price is going to raise every year. This is what I get from my suppliers. We also get very clear comment from a TSMC side.
Phison become very key strategy partner in the storage system because we gain the share in the PC and the, mobile and also we create the AI solution, aiDAPTIV+ to prove, to prove this can help to ship more products in the market. Tight supply, yes, but Phison we have a commitment we can solve this problem. One thing is we also heard the rumor, no rumor, I mean you can check, our competitors moving their design to Samsung. I think this also may impact the TSMC supply to them. If this is the case, right, definitely we can secure more supply from TSMC.
Okay, great. Any other cost hikes or shortages you're concerned about that we haven't talked about?
Again, we need a more NAND supply. We're working very hard. We are flying China, Korea, U.S., Japan every week. We try to convince our supplier we are better and of course they agreed we are better, but they say, "Sorry, K.S., we are sold out." I say, "Hey, no matter what anymore." If we ask, they are still able to allocate more to us, but still this is far than enough. It is what it is. We work very hard. Second, substrate is a problem, but I think we settle down. Packaging side, I think we have a very good partner in packaging, OSE, ChipMOS, ASE, SPIL. The trade Phison is race strategy partners. We are good. Other than this, I think we all good.
Okay. I think we have a question from Warren online. Warren, do you want to unmute yourself and ask the question? Maybe if not, we will go to the next question online here. Do you have any guidance for OpEx for 2026?
Good question. Phison, we keep, I believe, I remember OpEx was flat 2024, 2025 because market, we not able to expand on the expense. Phison now plays a very key role in the controllers, PCIe Gen 5, enterprise Gen 6, mobile. Most important thing is the edge AI aiDAPTIV+ is proof. We are waiting CES, the tier one, tier two, tier three going to announce Phison aiDAPTIV+. We need to have more R&D in AI applications. I already opened some, heck out to, my team members to recruit some good talent in the AI software development. We need to make sure we have enough main resource to doing a lot of POC to prove the POC turn to revenue and the repeat order. Answer is yes.
The R&D expense definitely will increase, but we also still need to follow what our net profit is going to happen. Twenty twenty-six, we believe really optimistic because the NAND is tight and we win a good position design win and we believe NAND company recognize our value. I do not have a number, but R&D expense is going to increase.
Okay. Another question here. What is your procured capacity in 2026 in bits?
I want to say this, with me, ask to keep this confidential. We secure six NAND company, the flash supply is a LTA commitment, but at this moment we are hard to disclose this because everyone asking for more.
Okay. Here is a question about Hosin Global. Any updates about your China subsidiary Hosin, like earnings outlook or IPO plan?
Again, I can't represent Hosin's position to announce there. Phison also in the quiet period. We believe they have a plan. No matter what, their earning is turned to unbelievable good because of market tire, they have an inventory. So I believe they also have their own IPO plan, but we are not going to get involved into that.
Okay. Simon, do you have a follow-up question?
Yes, yes. Could you show the Q3 sales breakdown instead of here the inventory value breakdown? Because I have a couple of questions. The next time, K.S., would you share your presentation material just before this earnings call starts? Because still we cannot find your presentation material. It will help. Yeah.
Okay. They may talk to my next earnings call.
Yeah. Yeah. Yeah. Yeah. Here the question is, for example, October revenue good, right? 8% up. Sorry for the short-term question, but the October revenue growth month on month, which application has led the growth?
The main strong. The main very strong. By the way, from August, I already gave warning to my COO not to take that many orders, not to take, so we took down many orders. You know what? Because Simon, what my afraid today is Phison a month hit the NT 10 billion revenue. Then I'm in really pinpoint, you know? I need to secure my supply. So we need to manage our shipment.
Maybe another way, quote unquote revenue growth, is not 10%, 20%, but, you've been saying NAND price has been very, very good. Let's revisit the Q3 result. Any rough idea overall your ASP increase in Q3?
No. The ASP increase only started by end August. The market started to have some belief only starting from July. July when the market price raised, customers still complaining. They have a two months behind. We take orders shipment two months. The price will only defer from October, November, December. You mean the price increase from October? Mid-September. Mid-September, late September, then starting from October.
Oh, I see. A little bit up September and then more meaningful price increase October, November.
I asked my guy need to manage well, not to take that many orders at this moment.
I think it was March early, around early this year, you mentioned around the double digit was 20% ASP increase. Now the sequential price recovery even stronger than the,
already 100%. Already 100%.
You mean the 100%?
From March, from March to end September to today, already 100%.
Oh, that's why you mentioned the $4.80 price became $10.
$10.70, not $10. $10.70. Unbelievable.
$10.70.
Yes.
Wow. More than 100% increase.
Yeah.
Wow. Then your gross margin should be very high, right?
Again, we are managing our shipment.
Oh, okay. Oh, sorry. One question from some investors. Would you recap your plan for the convertible bond issuance? Sorry, we missed your previous comment or quarter on quarter . What's the current status and then what's the plan?
We decide the CB3, we get approval by around, I remember August, the market price was around 400-500. We are not decided yet. I push out the price decision to mid-December to February. We need to wait until the stock price stable. To be fair to the current investors, I don't want to investor feel the CB3 buyers take the benefit from the current shareholders.
Roughly the new CB will be issued late this year. No, no, decision late this year. Actual issuance sometime next year then?
No, when price decide, then issue immediately.
Oh, I see. Rough idea how much? $10 billion, $20 billion or?
No, no, no. It is NT 6 billion. The amounts are fixed. Just the price, comparable price is not decided yet.
Oh, only total amount, only NT 6 billion.
$200 million.
Yeah. Yeah. Yeah. Yeah. Very small.
Yeah.
NT 6 billion. It is nothing. Nothing for you.
Like LUT with the current share percentage is very small. Yeah. Price is going up.
Overall, you sound very bullish today and you think the memory cycle gray. Do you want to compare current cycle better than even 2018 with the cloud boom?
Never seen. Never seen. Never seen.
Never seen in your life?
Yes. Maybe only once in my whole life.
One of the key contributor definitely is AI.
AI influence.
AI influence. I see. Oh, one last question. You mentioned the six NAND maker, but if I count two Korean maker, one Japan, one U.S.
Two U.S. Two U.S.
Oh, you want to count the JB guy with the Japanese?
They own the factory. They own the factory.
Yeah, but there is, okay, you mean the Kioxia and West Sandisk is two, two, two players.
Y eah, two. They are not agreed. They are one. They are not agreed.
Oh, okay. Okay. You count individual Kioxia versus the Sandisk separate entity for you. Okay. Oh, lastly, YMTC portion is very big out of your total NAND sourcing. I am feeling less than 10% out of your total NAND sourcing.
This is mainly for China market.
mainly for the China market. Yeah. All right, K.S. Very, very helpful. Thank you so much, sir.
K.S., do you think your inventory level on an absolute dollar basis will go down in the next few quarters?
We still get supply. We still get supply. And the supply ASP increasing. I think, yeah, incoming quarters, inventory definitely going down, but I do not think it is significant.
Okay. Another online question here. Is there any cooperation with AI GPU chip maker?
Yeah. Yeah. aiDAPTIV+ is ongoing with several GPU makers. That is what until CES, you can see the demo in the market and the launch in the market.
Okay. Okay. I do not, I do not see any other questions here. I do not know if anyone else has any last questions.
One question. They are asking about percentage of sales to export to sport market. Phison is not going to sell to sport market. We do not have a component business to sport market. We only to retail the brand customer ODM, but we are reducing a lot this model. Mainly it is go to the design in, project, industrial, PC, mobile, server, and deleted. Yeah. We do not have a sport business. Okay.
Yep. Okay. Yeah. Yeah. I think that is, I think that is all the questions. K.S., I want to send it back to you for final comments.
Simon, do you have a question, Simon?
Yes. Just one last question. Sorry. Keeping asking, what is the percentage of your China sales out of the total? Maybe 10%, 20%.
Sorry. It is difficult because China sales to have our Hunsin subsidiary there. Phison sales to China is slightly small because more business go to the Hunsin .
Hunsin means what? Hong Kong, Singapore or?
No, Hunsin Shenzhen, my JV.
Oh, your JV. So once you fully consolidate all the business and any, what, 20%.
I need to check. Yeah. Yeah.
Oh, okay. Okay. That's fair enough. Thank you so much. It's a great, great, you know, today comment. Thank you so much, sir.
Thank you.
Okay. Any last questions from anyone? Okay. All right, K.S., turn it over to you.
Okay. Thank you everyone. CQ3, of course, the first half this year was really challenging business, but CQ3 turned to good. CQ4 turned to much better. Again, Phison, we're working very hard to win more projects from the NAND company, from system company, from KICSP. We believe Phison's after last many years, heavy earning investment, now is the time to see the return.
Again, I keep repeating in the third party, Phison has already uniquely positioned, able to deliver the storage system from a concept to a product. We believe our value can turn to a much better return to the shareholders. Thank you for supporting. Thanks.
Thank you.