Phison Electronics Corp. (8299) Earnings Call Transcript & Summary

Good evening, everyone. My name is Jeff Ohlweiler. I'm Head of Research at Macquarie. Very happy to be hosting Phison's second quarter conference call. I'm just an amateur NAND enthusiast, so I'll turn it over to a professional, K.S. Pua, CEO. K.S. will do a presentation, and then I'll lead the Q&A after that. K.S., over to you.

Yes. Jeff, thanks to you and thanks to everyone to participate in our CQ2 '24 earnings call. Okay. This is a breakdown of our CQ2 revenue. From this chart, you can see consumer wise. Actually, in the May CQ1 earning call, I already emphasized the retail is very weak. We have no any interest to follow the price, total loss in the retail. So the portion of retail will decline a lot because I have no intention to go there. Controller wise. The reason is that last year's CQ4 and Q1, our NAND company, as the customers, they're putting a lot of controller to get the major share in the PC OEM and the smartphone. So CQ2 is a bit slowdown. So that's why ratio also going down. But from CQ3, the demand is bouncing back. Gaming one also deployed, the consumer market is weak. The key word today is enterprise. From CQ1 5% we increased to 14%. This is what -- I keep saying that enterprise is a trend of the storage. Without enterprise, I think the business is gone. So this is good to us, okay? So these to share with you. And these 17% others, this is a geopolitical impact. China and India, they keep asking to produce the storage by themself. So this we ship as a CKD, okay? We shipped the flash, plus controller, plus PCB and plus manufacturing software to them. They are produced in the local. This is only way to gain share in China and India, okay? So overall, retail is very weak. Without the new business such as enterprise SSD and our Phison AI on-prem and embedded ODM, OEM to their PC and the mobile, I don't see the hope in future. The reason is, in retail SSD, the user has not any incentive to buy, because if today you go to buy either a PC or notebook, you already have a high density of SSD, so you don't need to upgrade. The upgrade market already passed, okay? So who rely on the retail, who will be troubled. But Phison's good is, since 4 years ago, we keep investing to enterprise. Now we can see our fruitful return to the company. So revenue-wise it's $15 billion. Gross margin is $5.5 billion. Y-o-Y we have a 70.5% improvement. So this is our revenue, I think we announced to the market, okay. Gross margin rate, 34.86%. Actually, we grow than in CQ1. The reason is we still got many design in business carry from CQ4 and CQ1 to CQ2. And we also have inventory write-down by 0.5%. If this deflects to the gross margin, gross margin was 35.3%, which is in the high ranking. Earning wise. Earning per share is $11.97, EPS. But the amount of profit increased. The reason EPS is slightly down because we have a treasury share issue in CQ1. So the total share number increased in the CQ2, okay. So overall, the first half, we have a $23.99 earnings per share, which, I think, is an all-time high in our history. Even though the market is still making noise, but we still showed the very good result to our shareholders. And this is our income statement, okay? The OpEx we start to control. That I'll explain to you. And the total -- the financial ratio, which is gross margin, OP margin and net profit margin, they all improved Q-o-Q, okay? So I'm happy to see this result from our hardworking staff. In balance sheet, you can see inventory -- later I'll eventually put it down to tell you what is our inventory ratio. But overall, we are still doing okay. And I need to emphasize that by flash petabyte actually we are declining, we are declining. But why the amount increased? Because the flash itself, the net value increased 3x in the year. Since last year, August, until now the flash price bounced back 3x. So petabyte we are controlling well, but net value increased. We have to maintain enough petabyte for our future business. Okay. In Phison Non-TIFRS this quarter not any big difference. CQ1 a bit different because we issued the treasury share. We need to cover the write-off of the profit. But CQ2 no any treasury shares, so almost -- slightly different, just a small difference. Okay. Business highlights. In this Computex, Phison -- we first attend and we announced our aiDAPTIV, which we -- very covet. A lot of visitors come to Phison to ask what is aiDAPTIV. We attracted a lot of new customers to this business. We also launched our PCIe 5.0 solution to the market, and this product is going to market by September. Okay. We also win the award from the media, and this product is mainly for the image processing products. Last week in the San Jose, the Flash Memory Summit, we launched our 64 terabyte PASCARI SSD, which is to be the AI ecosystem. We met 5 of the storage box maker. They asked Phison to customize the 64 terabyte. Actually, they are also asking 128 terabyte. In October, the OCP in San Jose, Open Compute Platform, Phison is going to announce our 128 terabytes to the market. So if you go to this industry, enterprise, Phison was behind. We're just a follower, okay? But today, we are -- deliver the 64 and 128 terabyte same as the key player like Samsung, Solidigm. So we have become -- from follower, now we are able to achieve our position. So this is a big breakthrough of Phison, our technology investment and also our future business. We also launched and demo our aiDAPTIV Pro Suite, okay? We got at least 2 Tier 1 server makers show interest and the POC done. Now discuss is a go-to-market policy and strategy. And Retimer, we also attract 2 CSP to come to us to talk about using Phison's Gen5 and we commit to help them to make a good cost down to them. We also get an award from FMS, is Most Innovative AI Application. According to the analysts, when we talk to them, the cloud -- the Gen AI still in the cloud. The Gen AI definitely need to go to the on-prem. It looks like aiDAPTIV is the only solution to go to on-prem because of cost performance. So later I will show more to you, okay. We also announced our Phison PASCARI in the FMS, which is X-Series up to the 30 terabyte. This is for performance. Most important is the D-Series for the AI ecosystem, up to 128 terabyte. This is helping Phison to achieve our position in the enterprise industrial. We also bring our AI-Series to the market. Okay. Enterprise. We are proud that from -- CQ1 was 5%. CQ2 improved to 14%. We can see the trend. Next year, when Phison -- our D-Series is going to the market, we can see the percentage of the revenue will improve a lot because this is all high density. To give you a number. USB memory card is a single-digit dollars. SSD in retail OEM is a double-digit dollars. But go to the enterprise AI, it's 4 digit dollars. It's a few thousand dollars per drive. So this will help Phison to gain our share, our revenue. So our goal is to achieve enterprise storage 3% to 5%. This is our goal. And in the second IC, we start to ship, start to ramp up. We gain the CSP business, we also gain the server company business. Okay. Again, let me brief what is generative AI. Okay. Now everyone is crazy about Gen AI and we believe everyone will need Gen AI. But how? To give an example. 25 years ago when you started to use an e-mail in U.S., you use AOL and -- the AOL and Yahoo! In Taiwan, you use hotmail and somebody use Yahoo. But today, any single company, you have your own mail server. What's the difference? 25 years ago server from CISCO too expensive. And this is too new. In the market, only a few people know how to install the new system. But today, the OS is there, the package is there, the servers are so cheap. So use the same theory. Now the cloud-based GenAI service, you can -- GenAI can only be cloud because too expensive, hard to use. So future, is this going to on-prem? Based on the mail server system we believe the GenAI will go to your server room in your company. But how? The reason the cloud-based today is people using because you got no choice, no other candidate, no other solution. But cloud still got the 2 issues which is hard to solve: first, you are not going to share your sensitive data to cloud; second, once you subscribe the cloud service, your fee will be un-ceiling, you have to pay your whole life, okay? So go to Phison aiDAPTIV, you build a server in your server room, your total costs already there. So you know your payment in the future, "Just that," okay? And you can put your data in your server room, we can train by ourselves. And Phison already start to deploy a lot in our sever room to let my people to use. So up to today, we have more than 150-plus POC case in the market. A lot of the AI ones start to ship. And also, we have a lot of projects ongoing to develop the software for our customers. Recently, we are working with one of Taiwan local hospital to help them deploy our aiDAPTIV with our proprietary software to install into their server room to service the whole hospital. If this success, we'll take this as a model start to sell the software to any other hospital to improve our gross margin. We are glad to share this news with you. NVIDIA is going to event at Shenzhen and Chengdu to promote the aiDAPTIV, jointly promote aiDAPTIV. Why? Because aiDAPTIV finally been recognized as the best and only solution to go to on-prem, okay? So with the GPU card supplier to promote this, I think my life will get much easier. I don't need to -- no need to keep talking to anyone because a big guy already there. So we are going to send our people to Shenzhen. This happens by August 21 and August 23. We are going to make live demo to the users there to help them to promote this to the market. If this going is more popular -- and we believe the worldwide will see this going to happen. And you also agree, gaming GPU starting from U.S. but popular in China. Mining starts from U.S., but popular in China. So we believe that Edge AI will get popular in China. At Phison we are able to have this achievement just because of heavy investment, okay? We keep investing in R&D and we continue to keep invest, okay? So back to today's summary. We all agree and we observe for the desktop and notebook, which shipped to the market already preinstalled the high-density SSD. The users got no any incentive to buy the upgrade. So the low-end retail SSD is going to decline very fast. In retail, the market will be there only like high-density, high-performance, high-quality and gaming. This is good to Phison because at Phison we have all the in-house solutions. And these were bad to the module house, the pure module house. If they got no any new technology, new products, they will suffer the declining of the share and revenue. And with our AI, we are good to have a GPU company to co-promote in the market, okay? This will be helping Phison to make these popular very, very soon. And enterprise, we been follower, but now we achieved a good position because we're going to high-density products we launched same time with the big guy. In automotive, Phison keep investing a lot and we gained a lot of share in the EV, mainly in China, U.S., in the Germany. In Japan, we also gain our share from the hybrid car business. By the way, we also get a lot of design wins with the PC OEM. The client SSD without PC OEM gone, retail is gone. But at Phison we have a controller to the NAND company. For PC OEM, we are a key player. We also have a module, a direct design into the PC OEM. Okay. Then we'll finish a few Q&A, then we go to the online discussion. The dividend will be $13.19, which is 55% of our financial book, EPS $20.99. I need to highlight one thing. Treasury share, we ran off around -- I believe it's $1.2 billion of the overhead. Right now, we are not considering this as a dividend for 2 reasons. We have a CB1, is going to be expired by end of this December. The converting rate is still low. So potentially, we may need to buy back the CB with cost. I believe this may not going to happen, but we request by our Governor to keep some financial backup in case we need to buy back. But I don't think this is going to happen. Second, we gained a lot of our design into the enterprise with high-density drive. We need to start to prepare our inventory for our coming CQ1, CQ2 business. So more members decided to keep the high level of cash flow instead of pay dividend off that. But the treasury share happened by 2024. We are going to pay back the dividend to the shareholders by second half next year, okay? We keep the commitment. But we just need to keep our high level of cash flow for our coming enterprise business. And a lot of market analysts showed their worry about the so-called inventory risk. So I breakdown here. Controller is a very safe, 70%. Industrial, mainly is a legacy. And EOL flash, special for automotive. And the cost we accumulated -- this happened by 2022 and '23. So the cost is very, very low. Enterprise. We need to increase our enterprise ratio because the business next year mainly coming from here. Embedded ODM is a design win to the PC and the smartphone and some systems, which is follow the customer forecast. Of course, CQ2 smartphone is very, very slow. But now we got a lot of inquiry for smartphone bounce back for the CQ4 Christmas season. PC OEM was very good in CQ1. CQ2 very slow. CQ3 we need to watch the CQ4 demand. Retail is around 20% of revenue. So Phison -- next, the buying policy, buying more in enterprise, buying less in the retail or almost not going to buy in retail, okay? So the risk exposure in the inventory, I think, is quite low. A lot of shareholders also complained, when Phison business is doing good, earning good, overhead increase because head count increased," okay? But look to the '23, '24, even though our project from the NAND company, from enterprise, from AI increased a lot, but head count increase is limited, less than 5%. Why? First, I care about the OpEx because I already saying that we are not going to increase the OpEx anymore, okay? Second, we start to deploy our AI aiDAPTIV in our internal use from April. This improves our cycling time and improves our efficiency, okay? So I'll give you an example here. We developed a lot of controller for automotive customers and the NAND company. When engineer finishes a C-code design, they need to typing, writing article for certification. So they finish the code one day, they need to write article for 3 days. But now we use our aiDAPTIV, automatically convert the C-code to the Word -- to the document. Then you can see for the one function in the past, we need 20 engineers working for 7 weeks. But now with the GenAI support, 4 engineers with 2 weeks. This increase lots of efficiency and also reduce a lot of head count. So we are going to deploy more GenAI into engineering, not only engineering, in the whole Phison activities. We are not going to increase head count, okay, but we are going to improve our cycle time with our aiDAPTIV to prove this is really helping us. At the end, the earnings definitely will improve. So ESG, I think our score from '22, you can see from ranking 3 now go to 4.1. The full point is 5. So we keep improve our ESG, okay? So I finish my presentation. You can go with the online Q&A.

Great, K.S. Thank you very much for that presentation.

I guess the first question -- it was great to break down the inventory. That's very helpful. If you look at the cost of inventory versus current NAND prices, is there a pretty big gap between your inventory cost and NAND prices now? Or is it pretty similar?

Industrial wise, it's much lower than current price, okay? Enterprise -- actually, we also accumulated some inventory since last year's '23 CQ1 and CQ2, which is still there because we think we -- at that time we prepared this inventory for the designing business. But from now on, going to buy the new flash for future use will be market current price. But with the enterprise today, our cost is still really competitive. For retail-wise, not because of my cost high, just the end market price is low. China module house are dumping, everyone is dumping. So that's why in this part we are -- I have become more slow selling. So I'm also trying to convert some [ flash ] here, go to embedded and go to enterprise. But overall, our inventory is still very healthy by average cost, still very healthy than current market. And with today's NAND market price, we are much -- far lower than market price.

And just a reminder to people calling in, I'll ask a few questions first here, but feel free to raise your hand. You can ask questions next. And also if you have a question, you can put in the chat box as well. So also, K.S., I guess just in terms of overall demand going into the second half, how are you seeing demand half-on-half? Obviously, a very strong year-on-year growth. What does second half look like?

Okay. Second half -- I talk about CQ3. You can get from the market, right, consumer is very were slow, okay? This, I think, is normal issue. But we also can see by CQ4 the Christmas season coming, smartphone demand start to ask for new order. PC OEM not yet, not yet. Gaming really start to prepare some -- asking for CQ4 shipment gaming. This is all about the season, okay? But enterprise very strong, very strong, okay? We don't have enough parts to ship to get the project for CQ1. That's why we need to start to keep cash to buy the new flash for CQ1 demand. So overall, CQ3 weak in everything other than AI and enterprise. We can foresee CQ4 retail will bounce back. But how strong, I don't know. Depends on interest rate, okay? But CQ4 AI will be getting much stronger, enterprise getting much stronger.

And since you have this chart up, can you talk a little bit about the difference in the margins if you go from enterprise, industrial, embedded and retail? Any -- can you talk about major kind of gross margin differences there?

Of course, the highest is the controllers, the highest, okay? And the lowest is retail, which now is very bloody. Industrial is really stable, very healthy because the price -- cost is there, price is there, so very healthy. Embedded OEM recently very challenging because the demand is low in the PC OEM, but in the smartphone start to bounce back. Enterprise depends. If we go to the mainstream -- mainstream, which is NAND company there, the price is very, very challenging, okay? So that's why we go to most likely a lot of customized, customers asking for special function, customizing. And this is one we can keep some better margin.

And does this chart give a good sense of maybe revenue breakdown in a few quarters from now?

I think for sure by CQ1 next year you can see enterprise portion will be much bigger because this depicts our future trend of enterprise. Retail was shrinking a lot, shrinking a lot, because -- I don't see retail can be healthy after CQ4. Industrial, I think, will be stable, okay? Embedded OEM, the reason good because we get several new design. This will be okay. So in the coming few quarters for sure controller will be there because we got NAND company, industrial flat, enterprise increase a lot, embedded most likely flat.

And 3 months ago when you reported first quarter, you talked about your NAND suppliers were kind of almost telling you they're not going to be able to give you as much as you want this year. Has that changed at all over the last few months? I mean, do you feel comfortable with what the NAND makers are offering you now? Or is that too much what it was 3 months ago?

Okay. The NAND supplier, they also supply the flash for industry, supply flash for enterprise, and supply the flash to the retail. Overall, for retail wafer, they keep asking Phison to buy. But I have not any intention to buy, because retail is bloody, okay? But for enterprise, we talk to our suppliers, they are shortage, they are shortage. So this is the other challenge to me. I need to negotiate and to ask LTA for next year.

So I guess -- before I thought -- NAND used to be more of a commodity where you could use the NAND for any kind of end market. So now you're saying the NAND you buy is really specified for each end market.

No, no, no. Any NAND can go anywhere, but you need controller, right?

Yes.

And enterprise control development, the cycle time is 14 months to 18 months. So when I decide to pick up any flash to go enterprise, I need to lock this supply, okay? I can use a retail -- okay, today, if I have a company, retail flash, which is available, I cannot go to enterprise tomorrow because I need 18 months from design to go to market, okay? This is all about the design activity. So when we decide to go enterprise with our NAND supplier, we need to have a contract with them for price and long-term supply. Yes.

And maybe for the enterprise end market, can you talk a little more about the customer base and how that's evolving as you get more and more sales there?

Yes. Okay. Basically, the NAND supplier, they prefer to go to the big CSP, one order a few hundred million dollars, okay, with the one category. But Phison, we are not going to do that business. We go to a lot of system maker. They try to differentiate their systems and mainly for AI. So we customize this design to them, okay? So Phison business enterprise is mainly in customization, not standard part.

And in terms of pricing outlook in the second half, do you see any major moves? Obviously, we had a very good movement in the NAND price over the last few quarters outside -- you mentioned retail maybe a little low. But any kind of ASP changes you expect in the second half?

NAND company regular business, they still have no intention to lower the price by quarter end, especially in the retail part. They've got a lot of second [ win ] in retail wafer. They may offer some more aggressive price. But for enterprise, they have no any room to negotiate the price. So it means the price in enterprise is still strong.

And any comment on your market share in the PC client SSD market. That obviously has been an area that you've just entered recently. And how is your market share progressing?

Okay, this is hard to say because China lots of downgrade -- in retail we got no number, right? But for PC OEM, we are increasing because we get the Tier 1 direct designing business. Okay? And in retail, because it's -- a lot of China is under cost dumping, I have no intention to follow. So we give up the market.

I think we have a hand raised from Jason.

I just wonder how do we look at your gross margin probably in second half or in coming quarters? Because it seems like your product mix keeps improving. But you mentioned that probably pricing for retail is not so good. So how do we look at this kind of improving product mix to sustain the gross margin in coming quarters?

Okay. Retail wise, okay, we believe supplier may come to us to offer a surprise by quarter end. We believe, okay? It's not happened yet. So in this case, it may help in our gross margin. But it's hard to say. It depends on their policy, okay? For enterprise wise, I think for designing, customized it's a better margin, okay? But we also suffer the net price increase from the suppliers, okay? So the goal is to keep the gross margin flat, our goal. But the goal is margin flat, but revenue improve next year. Then we are going to gain our net profit improved.

And K.S., I guess your growth this year year-to-date is very high. I think like 60% plus year-to-date for your sales growth. Can you talk -- a little bit of breakdown between that big growth and ASP hike?

I think we prepared this breakdown. Let me see.

I guess the follow-up question on that would be, how does next year -- when you talk about revenue growth next year, is that coming from ASP or is that coming from big shipments?

No, next growth definitely enterprise, enterprise and embedded ODM and controllers. Consumer again -- like USB market, right, USB was helping us go to the $1 billion. But recently, USB only like single digit -- low single digit of our revenue. Why? Because legacy. The SSD to retail definitely next will be legacy, okay. So in order to grow Phison, enterprise is our first choice, embedded ODM is one -- is important. Then to improve our gross margin will be aiDAPTIV. This is the high gross margin products.

And can you maybe talk a little bit about -- more of the aiDAPTIV+ gross margin and why that can be so high?

Because this is niche Phison. We invent, we got patent, we have no competitors. And with our aiDAPTIV, the system can improve their value in the systems, okay? And Phison will not only provide just a single hard drive, but -- single SSD, but we also deliver our software with license. And with software gross margin is 100%. So to make it simple. Next year, the revenue growth will be for enterprise and PC OEM. Of course, controller, we're getting many projects from the NAND guy. But gross margin can only come from aiDAPTIV, because this is a really high gross margin rate. Okay?

And on the controllers, can you talk a little bit about the foundry pricing and just general controller ASP momentum?

ASP -- since 2022, '23, NAND guy losing big money. We also wanted to support them, lower our ASP. But recently, they also support us to give better gross margin rate, okay? So in gross margin rate, we are going to improve. In the NAND -- in the wafer foundry, basically, since Phison, we show we have a much better synergy, we got some rebate system. Means if we're buying wafer, the rebate will come back to us after quarters. So overall, the cost may decrease, but the gross margin will slightly increase.

And I got a question on the chat line here. So just talking about first half and second half, since second half, I guess, is going to be pretty much in line with first half revenues. Is that for retail or for all businesses?

I think retail [Audio Gap] we'll have some big business deployment to the market in CQ4. aiDAPTIV, if it happens in CQ4, will be very helpful to Phison.

And you've also already had -- already a very strong first half, especially on the margin front. And obviously, the OPM in the second quarter was very strong as well. For the next, say, maybe 6 months to a year, what's your biggest concern?

My biggest concern is the NAND price of my enterprise business. If NAND still keep increase that much, then I'm in trouble, okay? Okay. But we gained a lot of business there. So I think I have bargain power to talk to my NAND suppliers. Second, retail -- again, I keep mentioning retail, retail -- don't expect that much, only season business, okay? But PC OEM, we keep working very hard. Next year, we are going to gain the Gen5 design win business. This is helping our margin. Last, next year, if you're talking about gross margin, aiDAPTIV will help us. So we need to push aiDAPTIV, go to popular as soon as we can. But again, with the GPU company adoptions, this will help us a lot.

I see we had a question here. Hand raised from Donald Jung.

So I have 2 questions on your Retimer business. So you mentioned you have like 2 CSP customers. So I just want to clarify. Are these like general servers or like NVIDIA H100 AI servers?

Not yet AI server, not yet. It's a CSP general-purpose server.

And like what is your strength in this market? And do you have like a target market share because we know there are some other competitors in this market as well?

Okay. Phison, we are latecomer, okay? We are latecomer. But our first -- we would go to MP and we got many design. Of course, now we are convincing the CSP and GPU makers that we are pin-to-pin, BOM-to-BOM, nothing changed, and we can give much better cost structure, okay? So a lot of activity ongoing. But anyway, we are latecomer. Latecomer -- of course, I hope our first version can gain a share. Yes, we are gaining the share. But with the high wish of a major share, we need to work very, very hard. But we got the open window in Gen5. Next will be Gen6, okay? And in Taiwan a lot of server makers, right? Of course, not the mainstream. But they want cost down. We already got many projects from them and start to ship for cost down purpose. Okay?

Are you going to like enter into the AI server market?

Of course, we try very, very hard. We try very hard. But you know this is not easy, right?

Yes, K.S, maybe a follow-up question on the Retimer. So why do you have a better cost basis than some of the players who have been in this market for a while?

Not better cost. I think the costs almost similar, just overhead. The leader, you see that last year they are only supplier, they're still minus in the net profit, right? Even the -- and just 70%, means overhead. At Phison we are good in overhead.

And on OpEx, I think if you look at first quarter, second quarter, you had about a $700 million drop. You talked about the treasury shares and also employee bonus. Is that going to -- is the second quarter level, is that going to be pretty flattish for the next several quarters in a row? Or is there any other kind of one-off on an annual basis that should come up in like next year and the year after?

Okay. CQ1, CQ2 -- CQ1 has treasury share, yes, there. And CQ2, we made a good profit. So we also try to write-off more bonus as a result for our future use. CQ3, CQ4, I think the OpEx was slightly lower than CQ2.

And looking at short-term monthly sales, you had a bit of a drop July month on month. When do you think you're going to start to see that month-on-month momentum start to go back up again?

Okay. This is just because of retail business, okay, and the PC OEM slowdown. We hope by September, October, when the market for season that retail is coming back, PC OEM is slightly back. And the price, of course, we gain lots of opportunity. And of course, when we improve that business, right, gross margin will be a challenge. You know we're buying high price in NAND, okay, and need to go to the market with the market price, okay? The margin is slightly low. So we expect aiDAPTIV -- we need to start -- actually, we start to ship, but not that many. With the GPU company adoptions to help promote to the market, we need to ship more aiDAPTIV to the market to improve our gross margin rate.

And we have a hand raised here, Simon. Do you want to unmute Simon. And Simon, please go ahead. Simon, you want to go ahead with your question?

I think again technical issue.

Well, maybe I continue. And Simon, feel free to jump in when you're ready. So I guess for your China subsidiaries, I guess you had some positive/negative quarters, especially last year. Any kind of major gains or any kind of potential losses from those subsidiaries this year? Or what do you think next year?

So sorry. I didn't catch the point, sorry.

Like for instance, HOSIN Global had a big impact on a couple of quarters last year, both negative and positive. So any major kind of subsidiary gains or losses that you expect?

I think coming few quarters, okay, we suffered the loss '22 and '23 first half. Then they turn to very profitable '23 and '24 first half. I think it's all about inventory, okay, in the China market. I think in the coming few quarters, we are able still to get the profit distribution from them, but the number will be quite constrained because the China market is there, like you can see China market.

And I know in the past you talked about a lot of your enterprise team was in Colorado, U.S. Can you give a little bit of kind of overview of your R&D team globally and how, one, it breaks down geographically; and 2, how everyone works together?

Okay. The whole R&D headquarters still in Taiwan. Colorado, actually, they did a good job. When we gained around 5 storage box maker in the U.S., the reason they come to us not because of Phison is attending. We have support in U.S., okay? And those companies, they also have a R&D team in the Colorado, which is our neighbor. So the Colorado team helping us to get a lot of opportunity. Other than U.S., right, we also just established a team in Malaysia. The reason is a push by U.S. customer about [indiscernible], okay? Malaysia team mainly is going to enterprise, automotive and aiDAPTIV. At the same time, China and India are also asking for local design, local built, national products. We have no choice. So we already have a China Xiamen branch, mainly focused to the enterprise development for China market. India, we also start to find out any partner to make the joint venture in India to support in the market. So this is overall our maps in the world.

I see we have a question coming from [ Terence ].

Just a quick one for aiDAPTIV, right? Are you also constraint NAND or you can use any NAND and just use your own controllers?

Okay. Basically, when NAND company, they produce the wafer, the NAND wafer, there's only one way for nothing to do -- nothing to define as a client or enterprise, only one wafer. Just after package, go through the different testing, different trim, then they define, oh, this is a client, this is enterprise, but the material is similar, okay? Phison, we are capable to take the wafer, to pack the wafer -- to pack the wafer to the package, go into client and go to enterprise. Because we've got all the NAND know-how, we've got the controller, we have algorithm, we have a packaging team -- we have a packing team. So this is Phison's value. So to Phison, wafer is only one wafer.

So aiDAPTIV is not constrained by enterprise NAND?

No. Phison can survive just because we get a one wafer. We can distribute to low end customer and super high end to the aerospace with a one wafer.

So earlier when you say that your enterprise NAND is constrained by the suppliers just because you don't have that specific controller…

No. Let me clarify again, okay? This is nothing to do the -- okay. Every NAND supplier has a different flash technology. But the flash we need to use controller to build this to enterprise drive. But controller developed one single NAND will take 14 months to 18 months to mature. So this is a heavy investment. So when we decide to go one single flash to enterprise, we cannot change to other flash tomorrow. We'll need 14 to 18 months. So the concern is when I pick up the flash, I need to make sure the supply and the price is consistent. Okay?

We can try Simon again.

Number 1, yes, great pie chart regarding the second quarter inventories. So number 1 question is, over the past several quarters, your inventory is always $30 billion according to your balance sheet. So any reason why you maintain such a high amount of the inventory consistently meanwhile, your revenue has been growing? That means you've been buying the NAND chip as much as you're selling your NAND product every quarter?

Okay. We increased our inventory since 2022 because we see Phison made a lot of design wins. Today, when you're talking about the amount is $30 billion versus 2 years or $20 billion, but the pickup by itself, now we are declining. The reason is the dollars might increase just price improved 3x within a year. So our pickup and component unit actually is lower than years ago, okay? And the reason is we keep collect the same amount of pickup by just because we got the forecast from the customers. We have a demand from customers by petabyte. But ASP increased 3x in the year. So that's why the amount increased. Okay?

So your volume down, but the price up. That's why your inventory value is $30 billion?

Yes.

In that case -- sorry, for the accounting question. You have to recognize inventory valuation gains a lot [indiscernible]?

No. In the past, right, we keep buying, we keep buying. By quarter end, some NAND vendor asking to lower the price because they ask Phison to take. So by quarter end, we need to write down our inventory because we buy so chip. Okay?

So going forward, do you think that you can get more NAND chip at lower price these days because NAND makers are increasing the utilization ratio?

Answer is yes and no. No, just because enterprise demand is very strong. I think you agree, right? Yes. But by quarter end, consumer is so bad. They may need to dump some wafer to consumer customers. In this case, I'm able to buy chip in quarter end.

And then very quickly, the July revenue declined, you said because of the retail business, consumer demand is not good. But why sudden in the July data points weakened versus Q1, Q2 broadly great? What happened to July? And why this is a 1 month specific short-term issue?

Okay. Day 1, this month, if I call it the order, the revenue were differed only after 2 months. You agree, right? The order we get shipment is behind 6 weeks to 8 weeks. So in May, my earnings call, I already saying that retail is broadly I have no interest. So for May, June, July, I asked myself, don't take any retail if minus margin, okay? So that's why delaying the revenue from May, June to July, August. On the other hand, enterprise still flat, increasing. Okay? And second thing is, for some reason, after June, July, the market suddenly frozen. I don't know why. I think you agree, right? Other than AI, smartphone, PC, this, that, start to go slowdown. Only AI getting strong. So our enterprise still doing good, aiDAPTIV improving but the PC OEM and the OEM and retail are very, very weak.

So you think it's a matter of the overall the macro demand issue because 6 months so ago, memory makers has said maybe some demand that can recover better situation, et cetera. But these days, yes, many people saying only the AI relate to the product is strong. I mean the AI server is strong, others not good. So…

No, we at least now that CQ4 seasonal coming, the PC OEM, the retail, some gaming start to ask some inquiry, okay? This is seasonal. Yes, AI is very strong. All the box maker, they asked about high density of storage.

Lastly -- so overall, any rough idea what percentage of your revenue for the aiDAPTIV solutions and what percentage.

Well, it's more because it's just a start. This is something like NVIDIA start to talking about AI computing with GPU, right? Nothing happened before 2021 after ChatGPT. But the good thing is, at least, we're already doing way hard to deploy our products to the market. And the key point is our GPU partner, they are going to promote this to the market. We have co-activity start from China, Shenzhen and Chengdu.

You mean this company?

Yes. By the way this is not we use. They already have an invitation letter to the public.

So maybe 1%, 2% of the total?

No yet. Next year will be significant.

But even late this year, very small?

You know what, this is totally new design.

New design.

New design. Okay? But I believe after this event, CQ4 can significantly increase. And next year I think I can enjoy the business.

So today -- so your enterprise solution is mostly typical conventional storable storage, maybe 10 terrabyte.

No, no -- yes and no. Some legacy yes, but we are dual -- but most of our business is coming from designing and customizations.

So any example how to understand your enterprise SSD product? What I know is mostly TLC-based, maybe, what, 10 terabyte, 12 terabyte level. Any rough idea of your enterprise solution specs?

Okay. Today, our shipment is here, is TLC, Gen4. Gen5 is coming by CQ4, okay? This small, very small, okay, still like 32 terabyte, but we are going to go to 128 terabyte by CES start to ship for AI shipment. Of course, our [ starter ] is legacy, right? It's -- major is 8 terabyte, 16 terabyte.

And so based on your controller, yes, PCIe...

No, based on what the market needs. And this year is all QLC.

Okay, this one is for QLC?

Yes.

But the PCIe Gen4 controller is yours or...

Phison only use in-house controller. This is Phison policy. All the design from Silicon IP integration software, firmware, service. Only ARM -- we license from ARM. The rest is in-house.

So here the PCIe for up to 4 terabyte or even up to 900, 600 gigabyte is your own controller?

100% in-house controller.

And then, again, fabricated by UMC these days or TSMC?

All TSMC.

TSMC. 12-inch [ node ] these days or…

12 and 7.

12 and 7?

Yes.

We have another question here from [ Gordon ]. Okay. I guess a couple more questions from me, K.S. Number one, inventory. Is there a set number of like inventory days or an amount where you want to be? Or is it really a moving target given the changing end market and the products?

Jeff, if I'm going to control this, right, then Phison will be constraining the business. This is difficult, okay? First of all, flash is not produced by ourselves, right? NAND guy willing to offer the low price to us when there's no spike, right? And Phison, we can gain business just because in some critical time we got inventory. So I have not any much concern about what the shipping day or this or that. But this is only way to fight make Phison stronger with good enough inventory and a lot of energy activity in the controllers. We are able to differentiate our value, then we can grow the business. So Phison is like Walmart, where you have a $0.99 product and you have $2,000 products.

And my last question, I guess, we'll see if anyone else has any questions to end this off. But when you market aiDAPTIV+, obviously, you're marketing together with the GPU supplier. But given the costs that you're offering versus their high-end sales for millions of U.S. dollars, do they not like your product because it could take some more share from their higher end stuff or no? Is it just a different market?

No, no, it's totally different, totally different. High end is multiple billion dollars, right? This is normal business. We're just helping the GenAI go to the on-prem, like, okay, this is shut. Okay? Without aiDAPTIV, I don't believe GenAI can go to on-prem, okay? Then you're talking about the on-prem versus the cloud GPU amount. GPU on cloud a few million piece, right? On-prem can be a few hundred million set of a server. The amount of GPU is a lot, much more than the cloud. But of course, the ASP is a bit different, okay? Cloud already have a good enough GPU for high end, but on-prem is nothing today. But we believe on-prem is a trend and definitely will happen, like a new server today.

I think we're almost at an hour here. I don't see any hands raised. So K.S., thank you very much. Back over to you for any kind of final statements.

Okay. Again, I think Phison just want to show all the analysts, investors that Phison the value is not asking the NAND price. Our value is we are able to create a lot of new products that can fulfill what the customer need. In our future trend, enterprise is a big growth, no doubt. And most likely, Phison is only third party other than Huawei. Huawei is a third party can serve enterprise for the market because these you need engineers, the technology and finance ecosystem. Other than this, aiDAPTIV can help improve. aiDAPTIV is a trend. All these solutions go to on-prem. No doubt, no argument. This can help Phison improve our gross margin next year. Okay.

Thank you, everyone. Thank you for your time today.

Thank you.

Bye.

Bye.