Jeff Ohlweiler: Then we will go into Q&A. A lot going on in AI, in tech, in NAND, in memory. Look forward to hearing what Cass has to say. Thank you, Cass, for letting me join with you. Pass it over to you.
K.S. Pua: Yeah, thanks, Jeff. Good afternoon, good morning, ladies and gentlemen, to join Phison our Q3 earnings call. Let's restart. Okay. This is our breakdown of our Q3 revenue. In this chart, you can see by our definition, our so-called the AI ecosystem module, over 57% of our Q3 total revenue, and the growth QoQ 88%. This will show the AI-related ecosystem, SSD, boot drive, aiDAPTIV, the demand is still getting very strong. The reason we show our embedded ODM industrial, the QoQ here, just want to. Some analysts, they may write. They say, "Hey, the consumer market, the embedded market, PC, smartphone, IoT." Yes, this market now, they have a headwind, and the challenge is the cost increase that much, the consumer may not able to take it.
K.S. Pua: Embedded ODM, our Q3 with minus 7%, and the industrial module with minus 30% in the QoQ. If Phison, we still stuck in the consumer or embedded market, then we may suffer. You can see some other module house Q3 revenue dropping. I think they are mainly stuck in this industrial. But again, we have our solution in the enterprise AI and data business. The controller wise, even though this year the smartphone declining, PC declining, retail SSD frozen, but our total controller shipment by unit will increase, by revenue it will increase. Means what? Means Phison, we win the market share in our controllers business. The retail is, since early this year, we already highlight, retail is not our focus at this moment, so it is only 2% of our business.
K.S. Pua: But again, after the other 2, 3 quarters, this declining in embedded industrial definitely will bounce back. Why? Because a lot of device equipment, they still have to renew, they have to upgrade. So we keep our R&D development in these 2 field to make sure after 2, 3, 4 quarters, we will back to this industrial to secure our share and our gross margin. Let me have a Mandarin to explain since more Mandarin speaker here.
K.S. Pua: 第三季呢，我们的营收是有显著的成长，特别呢在AI ecosystem相关的，我们的QoQ成长88%，总营收呢占了我们的50%以上，这个符合我们当初的预估。那我们今天要特别强调呢，在embedded ODM跟工业应用这块可以看到吼，这个是手机、PC，其实整个销量是下来的，我们也策略性地选择呢，低价的生意不做。那工业用呢，是因为所有成本缺料跟这个太高，这个单价太高，所以呢造成整个需求瞬间的这个减少，那对我们呢看到有30%的一个衰退。消费类呢是更严重，40%。那我们要强调一个controller呢，虽然今年的消费性市场几乎是冷冻，那PC销退，然后手机下降，但是我们的出货量还在成长，营收也在成长。这表示什么？表示呢我们过去的努力，技术努力，我们的这个market share是在成长的。整个市场是衰退，我们出货量在成长，表示我们的成长率挺高的。那这个点我要特别再强调一次，群联电子呢，如果过去没有努力在AI的ecosystem、enterprise、adaptive这边呢去着力的话呢，我们如果只在这个生意，那我们就符合某些分析师讲的衰退。那我们同业里面也出现这样的情况了吼。那我们从年初已经开始有特别的强调，我们呢研发资源会往AI落地企业级应用来这个加码。那也因为有我们的这个过去的努力，所以第三季呢我们的营收还是有很显著的一个成长。这边我觉得有必要呢做一个比较详细的报告，免得说这个鱼目混珠吼，觉得这个评估来评估去，没有看到一个重点啦。30 minutes ago, we just released our earnings.
K.S. Pua: Our gross margin in Q3 is 51 million. Gross profit QoQ 15%, and gross profit up 700% YoY. Revenue wise, Q3 86 million is the all-time high, follow our expectations. This part, we like to show you gross margin wise. Since Q2, we already highlight to maintain the mid-60% gross margin is not a reality because the market, the cost increasing. But we still able to close to 60% gross margin. As a Q1, I also keep highlighting that the gross margin rate, we decline, but the amount, total amount of gross margin, actually we are increasing. This is to show our business configuration. We move more into enterprise, even though embedded ODM, consumer, industrial declining, but our gross margin still able to increase.
K.S. Pua: 这边再强调一次，当然过去讲说低价库存，但这个我想——讲的在某个领域是对的，在消费类等等是对的。但是，虽然我一直认为说要维持一个高于60%的一个毛利率，以我们来讲是一个很困难的工作，我们努力会维持我们的相对好的毛利率，但是我们认为我们毛利会增加，这个就是我们今天证明。我想往后毛利的成长，是我们期待的，也是我们努力的，而且我认为是可以达到的。 Earnings per share 138 is our all-time high. We have around 20 EPS growth from Q2. Back to the income statement, I want to share with you that at our Q3, the whole quarter, we are running with zero debt. The whole market keep highlighting the AI ecosystem, the player, they are short of cash flow. But Phison, we are running in the zero debt. We didn't take the loan from bank. We still able to keep the free cash around 20 something billion. I think we have a 20 something billion cash on hand. 这边，咱财务报表各位可以看，整个产业从第3季开始都面临资金的不足，我们还是有能力维持在一个零负债的情况，来经营账上还维持200多亿的一个现金，库存水位也明显地有增加。 This is to show you our balance sheet. I think you are concerned about our inventory. Inventory is around 111 billion. We are increased by 20 billion compared to Q2. Our cash flow, we have around 27 billion cash. Overall in finance, we are still doing very healthy. 在整个市场一直讲到说资金短缺对这个产业的影响，到目前为止，我觉得我们还是维持一个相对健康，库存还是持续地增加。为什么？因为我们要准备企业级的AI的应用，还有开机碟的应用，那个需求是非常强劲的。我们还是持续努力，维持一个零负债的经营，让公司的一个财务情况在最健康的情况底下。 This is to disclose our NAND Flash result. Here no much different, only a small gap. Just reason we have ECB and we have issued the stock option a few months ago, so we have to every month depreciate the cost, but the gap is very small. This is mainly for the employee benefit. 那个NAND Flash的差别不大了，主要是因为我们现在只有几千张的一个选择权给了员工，所以每个月必须要摊提。 Revenue wise, by October is 31 billion, QoQ is 2.9%. Demand in the enterprise still very strong. This page, we want to show you what is our next. From first decade, we'll be in the consumer controller, consumer module. Second decade, we move to the embedded smartphone, PC. We keep being highlight, we are in the cyclic memory business. So we work very hard to enterprise, now we move to the AI ecosystems. According to NVIDIA disclosed the 5-layer AI stack, Phison, we play the three roles. In the chips, we develop every kind of chips, eMMC, SSD, UFS, Retimer. In the infrastructure, we are providing our whole AI server with aiDAPTIV AI PC, enterprise SSD, boot drive and customized solutions. Application wise, we are developing a big group of our software team to provide our AI Data Platform, software and aiDAPTIV solutions. 我们一直努力地在做一个转型，这个转型的成果可以期待，很快就可以期待。怎么说？ We have investing huge amount of new software engineers here, mainly in the AI software developing and the AI software business development. We are now start to promote our solution to the banking, hospital, government office, use education. Later I will show you our some achievement. This is to help Phison to migrate, to move to the next system business, which can ensure we can enjoy the much better gross margin rate. 我们持续地努力在这个AI的生态圈的开发，最主要目的就是希望在未来可以维持一个相对好的毛利率。当然毛利，我认为我们会很健康，但是投资人希望看到毛利率。毛利率用模组来看不现实的，我们希望有系统的生意，这就是我们过去 3 年的努力，很快就要开花结果。 We are keep mentioning our next Phison third decade. We from controller, now we're doing the chip design, our in-house IP, our modules. Now we're providing our AI Data Platform as a system which able to gain the more gross profit, gross margin rate from this system business. Q3, we'd like to have some business highlight. In this year, until end of September, the NAND company, three NAND companies, they release their TLC and QLC Gen 5 client SSD controller project. Phison, we won 100%.
K.S. Pua: 100% means the NAND flash program, we took NAND flash program.
K.S. Pua: Means in the coming soon, the consumer SSD for PC, Phison will become the key player, very key player, okay? With this, our NAND flash partner, we also helping them to convert the future client SSD solution to the aiDAPTIV compatible solutions. AI PC, within 3 years, every new PC, the new CPU, they are able to run the AI inference, but memory is a constraint. With the Phison aiDAPTIV, can help every PC convert to AI PC. With this 6 design win with the NAND controller means 100% hit rate, can help Phison by 2028, we will become very big player in this PC market. 今年 3 家 NAND 的原厂放出了 6 个案子，Gen 5 的下一代的 TLC 跟 QLC 的 controller 的专案，群联电子我们 6 个案子 100% 命中都拿下来了。同时我们协助他们在未来的 PC 领域里面，除了传统 PC 储存，我们会把方案做成相容于 aiDAPTIV 的方案。意思说在未来 AI PC 普及之后，群联电子就是这个领域的一个非常大的提供者。 Smartphone even though declining, but Android phone, Phison we gain our share in the unit shipment, especially in the UFS 4.1. We are shipping big amount of controller to almost every Android phone makers in the world with the TLC solution, but we are the second company MP the QLC solutions. First was the Samsung, they killed the program, but Phison now is a key player in the UFS QLC solution supplier, mainly to the mid-high end Android phone. We also win the several UFS 5.0 controller project and get the full design win in the China Android players. 手机在下滑，但是我们的总出货量还在成长，表示我们有一个很大的市场率的成长。在 UFS 4.1，我们也协助了伙伴把 QLC 导入量产，这个是仅次于 Samsung。Samsung 案子做了一段时间把它停掉，因为速度不好，我们这个案子还持续地在放量当中。这显示了我们在技术的投资得到很好的回报。 Enterprise SSD is a trend, mainly in the AI ecosystem. Here we have two part, one is the boot drive. Our boot drive growth by unit, by revenue, YoY is almost 20 times per year, okay? The demand is strong, so we keep get the new design win. You can read here, we have many design win from different tier of players. Mainstream wise, mainstream will consume more flash capacity. We win a lot of new design win in Q2, 3, more coming in Q4 and Q1. Why? Because Phison we prove we are capable to deliver from controllers to modules and to system and service. 企业级 AI 应用，不管怎么说，未来一年半里面绝对是这个产业成长最快的，消费类是很辛苦的。我们在这里面攻城略地，每一个季度都得到很多新的 design win，而且这个成长的幅度越来越快，表示我们的技术能力、供货能力、服务能力已经被整个市场所肯定。 Enterprise SSD, we also disclose our Gen 6. We are going to have samples by end of November, and we are going to have a co-announcement with the CPU players in the Gen 6, mainly for the inference solution. Phison will become very earlier supplier in the Gen 6 solution by next year. UFS controller IC is a small kind of a business, but Phison, our Gen 6 controller already in the market start to have a shipment, okay? Even though this is niche market, but to show Phison capability in design. aiDAPTIV, we keep talking for almost 2 years. This was happened in the early August FMS, Intel, we have a co-announcement, Phison engage with Intel to promote aiDAPTIV go to the AI PC. Very soon, CES, you can see what is going to happen in this industry. We are become the very dominant key player here. Just recently, we win the Micron two very historical awards. First, this is our third award in the supplier day in the semiconductor part, and we won also the quality award. In the Micron last all the event, Phison is the only company to win two awards in the one supplier day. This is to show Phison our capability in design and also in quality. 这点还是要特别强调，美光是一个非常严苛要求，或者讲直接点是非常 demanding 的公司，要求是很严苛的。我们是第 3 次得到它的最佳供应商，同时也得到它的最佳品质奖。从历史上他们的活动里面，我们是唯一一家在一个晚上得到两个奖的，这个是对我们一个很大的肯定。 Okay, this one to show you about with so-called the Phison AI Data Platform.
K.S. Pua: I do not want to spend more time here. I just want to show you, this is a hardware layer. It is a GPU server, cache server with a Phison aiDAPTIV and our storage. On top of this hardware, this is a different layer of a software application, and this is all developed 100% by Phison. We try to build a software barrier in the on-premise AI solution. If so, we can make it, we are able to get more benefit from this business. We keep investing, keep hiring people, and more business are happening, and we are going to show more our output in the market very soon. This is to show you how we convince SME to use a Phison solution. Every SME boss, they want to go to the AI on-premise, the consultant may ask $1 million, $2 million, $3 million.
K.S. Pua: With Phison solutions, with a small amount, 100K, 200K, to do a POC. After the POC good, they put online, then they are facing the compute shortage problem. At the end, they need to keep buying the server. They have to spend the same amount of the day one what they have been asked. With our solutions, gradually, they buy the server, they can achieve to improve the efficiency, and they are willing to buy more servers. This can help the industrial. This is our accumulated data. We already have 17 banking customers. We ship our AI server with our service. We have 85 plus SME in Taiwan, 50 in Malaysia. We have 10 plus hospital in Taiwan to use our AI service. We have more than 18 school, high education university to use our solution for AI training.
K.S. Pua: We have some government offices to use our solution to do their workload in the office. It is on-prem. This to show you how Phison use our AI on-prem. We develop the solution for ourself. What we gained? By October, we have 84% employee, means close to 3,000 plus employee using our AI on-prem service. We save 333 headcount of R&D, or we did not hire this headcount, but we have this headcount output. In here, we are saving close to $3 million a month in the salary. Means we improve our cycle time and also efficiency. The server we use here, I want to highlight, once the server get to our server room online, the server cost depreciate to zero within four to six weeks. Weeks, not months. Four to six weeks, we gain the headcount, then the server value can direct move to the zero. Means what?
K.S. Pua: Means second month, the server we use is all free.
K.S. Pua: 中文讲解一下，我们努力用我们的AI方案自己建置，然后自己内部解决掉很多问题。10月份使用率，全公司大于84%，产生300多位的这个工程师的output，等值呢将近900万美，900万，300万美金9,000万台币的一个薪资。同时呢我们的server上线之后呢，4到6周就可以直接归零了那个成本，剩下的第二个月只有电费。这是我们目前以这个成绩呢来告诉所有的中小企业，我们的方案是可以协助他们得到这个效益的。And we also made the one simulation.
K.S. Pua: By our October, the token we use on-prem, if we convert to the cloud, the most cheaper cloud, the cost will be $2.1 million. We did not pay. We pay only our electricity cost. This is to show on-prem is coming. You are the investor, I think you know more than me, a lot of industrial giant, they start to limit the AI service to the employee because the cost is incredibly expensive. We are now providing our AI on-prem with our AI Data Platform to help every company to save the cost in their server room, and we prove. I see the very good result in improving our design efficiency and also our headcount saving. We keep buying server, more servers coming, put to our server room. Now our server room run out all the power supply.
K.S. Pua: Now we move to our outsource to the AI data center. We are going to save, my target is a 700 headcount R&D by end of this month. At the end, we also improve, strengthen our design capability in the very short cycle time. Our R&D expense is still high, very high. 20% of our revenue we keep flat in the last few years. We are keep crazily invest to the R&D. Why? Because we can see few years ago, we invest to enterprise AI, now we get a good return. Coming years, I wish Phison we still growth our profit. This means what? I need to burn out my earning to engineering to increase more product line. This is to show you, Phison's our hierarchy. How can we make the good gross margin heavy R&D investment? This is happening.
K.S. Pua: Up to first 9 months, we already spent TWD 39 billion of R&D cost. We keep training more product manager, more BD for our AI business. We still need to hire up to 250 engineers in the coming 6 months. Why? Because we win every controller client project. Every. Coming months, we are going to get more customized controller project for AI networking, for AI systems, for AI PC. We need to hire extra 250. We now are going to appoint headhunter to help us to gain more R&Ds. Last summary. We work hard to increase our value, our profit return to the shareholders. We want to make sure Phison, we are able to influence in this industrial, not in storage, but also in the on-prem AI solutions. We keep investing in the market.
K.S. Pua: We also try to build a more joint venture with a different kind of partner. In Malaysia, we already have a joint venture with the core banking developers. Now, we are moving to the banking business with the core banking plus AI solutions. There is a question here I want to share here, then we can go to the open Q&A. This morning board member meeting, we just approved we are going to have a treasury share buyback, 2 million share. This is going to be encouraged to our employee. Recently, the stock market has some noise, but we want to show to the market we still have a good confidence. We got a question from an investor. Why Phison keep have a higher R&D expense? I think few reason here. We win almost every controller project, so we need more people. We need the more equipment.
K.S. Pua: We need more tools. Gen 5, Gen 6, now we are in the leading. We need to have more samples, enable more NAND flash. Fortunately, we win again many AI data center, not hyperscale, OEM, AI ecosystem, whatever. We need to have more people to in charge of those kind of business. Again, we need to have a headcount. On-prem AI is coming. We are developing more software application. We hire more software engineers. We are hiring more, invest more. Why? Because we can see the big return very soon. Also, there are some question about you keep saying about aiDAPTIVCache, what this can help after few years? This is number from our client. We provide our AI server to the customer with aiDAPTIV. Here is a simple few numbers. Without Phison's solution, concurrent user can be 16.
K.S. Pua: With our aiDAPTIVCache, can improve to 100 concurrent users. Time to first token with a different concurrent user, you can see the improved times here. The throughput, I mean, the SME more concerned about token per second. Token per second related to performance and also power consumption. With aiDAPTIV, we can improve three to six times of a server's token per second output. This is what we can gain to help the AI on-premise. They buy limited server with aiDAPTIV, they can create more efficiencies with the token per second output. This is also some concept. With our Phison solution, AI on-premise KV cache can help the efficiency. Actually, when the users rate the efficiency is good, they are willing to buy more servers.
K.S. Pua: 概念是这样子的，他们用起来之后呢，本来以为费用便宜一点，结果好用，好用更多的用，更多人用就要买更多的AI server跟SSD，因为它得到了成果。就像群联电子用几台伺服器就可以省下300个人，那我当然买几百台伺服器来用啦，大概就是这个概念。Okay. Our inventory breakdown still over half is our enterprise-related component, okay? And many ODM, even the market declining, but we have our strong partnership with the OEM. We still have secured a big amount of orders we need to ship.
K.S. Pua: We just walk away from the low-end business. Because low-end business now is very bloody. But with the high-quality products, we still able to get our gross margin rate. We go to the open questions.
Jeff Ohlweiler: Yeah, great. Thank you, KS. So reminder, you can either raise your hand or put your question in the Q&A box. Let me ask a few, I guess, to start here. KS, can you talk about the growth, the sequential growth in Q3, rough breakdown between ASP, hire, and bit improvement?
K.S. Pua: Actually, it's a unit shipment increase. Hello, Jeff, can you hear me?
Jeff Ohlweiler: Yes.
K.S. Pua: Okay.
Jeff Ohlweiler: Mostly units.
K.S. Pua: Unit shipment price increase, actually Q3 is quite flat. Unit shipment and some other new products, customization products, means we can improve the ASP with a different function.
Jeff Ohlweiler: What's your expectation for ASP over next couple quarters for both straight up NAND and also module price? Is there any reason why they'd move separately?
K.S. Pua: I believe CQ4 to Q1, you can read a lot of noise in the market, right? Just because the consumer is so ugly. We believe the price potentially flat or little bit increase, little bit. My worry is after Lunar New Year, the demand of AI will boom again, especially from China side. There are some reason, I do not want to disclose what happened here, but just a highlight. After Lunar New Year, China AI demand will boom again. Big boom.
Jeff Ohlweiler: In terms of sourcing NAND, how much are you able to get versus what you need or want over the next several quarters?
K.S. Pua: I am talking about next year. I have a target for next year. Until now, achievement of my target is around 70% plus. 70% plus. So, in November, I need to visit every different supplier in every different area to ask to fulfill what we need in 2027. I have a confidence I can make it. Why? Because we can bring more value to the flash they ship to us.
Jeff Ohlweiler: Okay. We have a question from Charlie. Charlie, do you want to unmute yourself?
Charlie Chen: Oh, yes. Thank you so much. I wanted to ask a little bit about your gross margin profile for your AI part of business and non-AI part of business, because you seem to suggest that the kind of help from low-cost inventory can help your non-AI side. It seems like AI part of the business, there seems to be higher cost or lower markup. Can you let us know the margin profile difference for your AI versus non-AI business?
K.S. Pua: Yeah. Thanks, Charlie. First of all, I just want to say I beat your report in February, right? February, your forecast at Q3, we are going to suffer the low-cost inventory, margin, blah, blah. Now we show to you, we are here. Okay? Actually, you are right. You are totally right. If Phison, we are only in this business. Okay? If Phison still stuck in the consumer, embedded, industrial, yes. The so-called low-cost inventory will start. I think some module house is going to disclose their Q3. I read some gross margin decline 2 digit. China module house decline more. Okay? Again, you can see our portion. 50 plus percent of our revenue are enterprise. This demand is very, very strong. This business asking a lot of customizations. Our October breakdown, this AI ecosystem is 65% of my revenue.
K.S. Pua: I believe every month, the portion of AI solution is increasing until maybe end of next year. Now this NAND industry has 2 portion. One is AI, one is a non-AI. Non-AI is really, really suffering. Okay? Luckily, Phison, in February, I keep saying we are heavy invest to our AI solution. Now we are here. Why you keep asking the low cost? Low cost eventually gone. Okay? We are still able to have some low cost, but we can't rely on low cost forever. The value is coming from R&D investment to improve our gross margin, such as AI ecosystem and our AI Data Platform. Right now, after the US midterm election, the noise saying that the data center may be under limitation. This can help to push AI on-premise going to popular.
K.S. Pua: Right now, we are providing not only a hardware, but we're also providing the total solution, including software. If this happened, we are able to gain more income through the software business, which is 100% gross margin. Means what? Means I don't count too much on low-cost inventory.
Charlie Chen: Okay. Hey, Jeff, may I ask a second question?
Jeff Ohlweiler: Yeah, sure. Go ahead.
Charlie Chen: Thank you. Thanks for hosting this call, by the way. Yeah, I think another thing that people kind of worry a little bit is about the PC demand. You see that Intel's earnings is pretty poor, right? Do not even look at Intel, just look at those ODM shipments, YoY, decline like 20% to 30%. I know you have a very, very exposure to that end market, meaning your AI PC. But can you share with us whether that part of business you are seeing kind of weakening demand? To combine what you just answered for the first part of the question, this is second question, what should be the new norm for Phison's gross margin in 2027, 2028?
K.S. Pua: Okay. PC market, since Q2, we already foresee the demand declining. No doubt, no question. Everybody knows. If you read books, you read news, you know it is going to happen. Again, show to this chart. We knew already, but at the end, we still grow the revenue, we still grow our gross profit. We agree regular consumer PC in coming 18 months, they are going to suffer. But look to the market again, NVIDIA, AMD, and Intel, they are strong pushing the AI PC. So every CPU from their new generation, they have AI inference capability, but unfortunately, they are asking for LPDDR5, 128 GB, and the market price is $8,000 to $9,000 per device. This is not going to be popular. So Phison aiDAPTIV, now we are doing very, very hard to put SSD aiDAPTIVCache into the PC.
K.S. Pua: We have a few secret programs ongoing. Let's wait, you can see in CES. I believe AI PC will go to the market, will go to the commercial market, it is going to happen by 2027 H2. If so, it means the PC industrial can recover from their so-called traditional PC with the AI functions. Second, you ask me where the Phison gross margin is coming from. Next year, we are still good in the AI ecosystem modules. Here you can see Phison with chip design win in many, many, many projects here. Every quarter we have a three win, one win, anyway. Second, this, I mean, use smart tool, right. This is a hardware business. I buy net expensive, then you still calling other low-cost business. I agree, but again, this is our so-called AI Data Platform.
K.S. Pua: If AI PC goes on-premise AI solution, Phison, we prove, we use our AI solution, we can improve this kind of efficiency. The AI server I purchase can depreciate to the zero cost within four to six weeks. We believe a lot of SME owners, they need this. We are going to take this business as a service. We are able to get the annuity income with 100% gross margin. So overall, I make it short. AI ecosystem will be my gross margin cash cow next year because we keep design win many customer projects. Second, AI PC, we are pushing this to happen. Third, AI Data Platform, we are going to make service. We have many programs ongoing, but I keep deliver a lot of message. People only can believe when they see the result. So let's wait to see the result coming soon.
Charlie Chen: Okay, great. Thanks, Kaz. Thanks, Jeff.
Jeff Ohlweiler: Yeah, thanks, Charlie. Kaz, just to kind of follow up on that, I would say pre this recent upcycle, your normalized margin was like low 30%. Because of product mix change, do you think over time that gross margin will be significantly higher throughout the whole cycle? Not just after the inventory is kind of not impacted anymore.
K.S. Pua: Okay, first of all, in my whole life, I never imagine we can hit the 60%, and I do not think this is a reasonable one. Okay? I do not think so. But I think coming quarters, if Phison can keep maintain above 50%, we are perfect. We are perfect. But overall, I keep saying that I want to improve gross profit, not percentage. Means revenue keep growing, dollar amount profit increasing, the percentage may decline, but this force us need to increase more software business to help to maintain the number of the percentage.
Jeff Ohlweiler: Okay, great. A question online here says, "In Q3, your revenue grew 27% from Q2, while Longsys grew only 3%. Your gross margin fell about six points, but theirs fell 16.5 points. Is this because years of heavy R&D gave you a different product mix with AI products now 57% of revenue, or because you had more low-cost inventory?
K.S. Pua: No. I mean, Charlie already asked, right? Charlie, February, when he said something, actually he bingo. He bingo is a typical module house, but no Phison. Okay? Phison, because we have our growth, our portion, a big pie in the enterprise AI ecosystems, which the module house at the last decade, they enjoy good, they did not invest much, but we are really working hard. Okay? We are good just because we are now in the AI ecosystem business. Without this business, you can see here, okay, minus 7% and by the way, minus 30%, then Phison will same as Longsys. Charlie bingo, what is the report? But now, I keep saying, listen to Phison. We are investing, we are smart, we are hardworking. Now we enjoy our last seven years investing in the enterprise SSD AI.
K.S. Pua: Coming years, I hope we are going to enjoy the AI Data Platform bring the much higher value to Phison. This is going to show to you by 2027.
Jeff Ohlweiler: Okay, great. Here's another question here. Has operating cash flow kept pace with reporting profits or declining?
K.S. Pua: Jeff, I want to answer there on the chat room, Mandarin, okay?
Jeff Ohlweiler: Okay, sure. Go ahead.
K.S. Pua: Talk about Kioxia release, their market is on the peak. This is something strange, okay? Very, very strange, okay? Kioxia released that their growth only 1%, 1%, 2%. It's kind of a joke. Let me ask a question to everyone, okay? If you have a clean room fully used with the current generation of a product, you don't have a new clean room, but you need to migrate your new design to the next generation. Are you going to stop some portion of production line to upgrade your equipment? Then after that, you can increase the new output. You catch my point? This is my personal study. The Kioxia Q4, most likely they are going to shut down few production lines to upgrade the equipment to the next generation, which has a higher output. Means they lost some capacity, but they maintain their revenue. Means what?
K.S. Pua: Means they are going to increase the price. Anyway, this is too bad to me. I wish they are not going to increase the price, but this is the fact which is my perspective, my view.
Jeff Ohlweiler: Great. Thanks. Since you have this slide up, a quick question on industrial module. Advantech just reported a couple of days ago, and they had pretty strong Q3 momentum, pretty strong Q4 momentum, as an industrial PC player. So curious, they are seeing very strong industrial PC-related momentum, but you are showing pretty strong industrial module decline. Anything worth mentioning there?
K.S. Pua: They are also my customers. My portion with them also increasing, but Phison industrial is a very big portfolio, which including automotive, industrial PC, equipment, and a lot of traditional infrastructure. Those portion due to the material shortage. Many material shortage, they cannot put in online. Second, some player, they already accumulate some inventory in their H1. Since the inventory high, they are going to reduce. They want to reduce the inventory risk, so they are not going to pull in. Overall, at the end, industrial infra is a slowdown. Only few key player growth. The rest are slowdown.
Jeff Ohlweiler: Okay. Let me read some of these other questions here. Number one, has operating cash flow kept pace with reported profits or declining?
K.S. Pua: Operating cash flow kept pace with report. I did not get to study this one, but so far, we are still running with the 0 debt, so I think I am good.
Jeff Ohlweiler: Has inventory turnover days decreased?
K.S. Pua: I think flat. Why? Revenue increase, but I keep accumulating more inventory. Why? Next year, again, gentlemen and ladies, you cannot believe. You do not know how tight supply in the flash after Lunar New Year, especially in the AI, because every cache is going to use every SSD. So I need to accumulate what I can buy.
Jeff Ohlweiler: Okay, next one here. When or have aiDAPTIV+ and Pascari produced commercial results that can be quantified?
K.S. Pua: Okay. Pascari is enterprise brand. Pascari aiDAPTIV is AI. We doing this design promotion development for 2 years. Next year, let me show you the number of aiDAPTIV. Next year. We do a lot of POC. We are waiting POC mature, the market buy-in, the use cases happening. Next year, Phison will have a special number to show you what is our aiDAPTIV business.
Jeff Ohlweiler: Okay, next one here. Kioxia seems to be making progress on a 100 million IOPS SSD for AI GPU servers. Where is Phison on this? Do you have a controller or product for it, and when could you ship samples?
K.S. Pua: Okay. First of all, this is very niche product, need to have a special NAND flash. The NAND company, they are not going to sell this NAND to third party. They only for their own use. To me, like HBF, if I go HBF, to me it is a nonsense. This 100 million IOPS, the controller need to go to 4 nanometer investment will be like $300 million. I no want to spend this money. Nonsense. Phison, we develop our own controllers with traditional flash. We can able to go to 50 to 70 million IOPS. We are not going to hit 100 million, but we use a low-cost NAND. Our current controller, without R&D expense, we go to 50 to 70 million IOPS, but not hit to 100, because that is not necessary.
Jeff Ohlweiler: Okay, next one here. For the CMX context memory storage tier, will the final design use TLC, TSLC, or a mix of both? What is Phison's role in it?
K.S. Pua: This all about Gen 6 TLC. Phison, we are now play the role, this so-called the cache offload SSD. Yes, we are there.
Jeff Ohlweiler: Maybe more of a question on your thoughts on demand trends going forward for new AI data center builds, given the recent US elections. Anything that you are hearing or seeing or expectations change over the last week?
K.S. Pua: Okay. We are not going to give a comment on the macro political things. But again, I believe no matter government change or not, they need GDP. They need to get the leading in the AI industrial. So I don't think they are going to shut down or slow down the investment. I don't think so. But again, second, no matter they are going to slow down or not, AI on-premise is going to happen. If the new government ready to slow down the AI data center deployment, this will be help to me to push more go to the AI on-premise. So this is our business.
Jeff Ohlweiler: Okay. We have another Chinese question here. Do you want to read that one, the aiDAPTIV+ one at the bottom?
K.S. Pua: aiDAPTIV 适合朝向消费性 PC. Yes. Actually, aiDAPTIV, we aim at two markets. One is AI server, go to the SME. The volume-wise is kind of small, but if it goes to AI PC, it can be a few hundred million sets per year. We win every NAND client controller outsource, we win. This 100% will convert to the aiDAPTIV. So 2027, please wait and see.
Jeff Ohlweiler: Okay. You obviously have a great improvement in product mix since your R&D investment over the last several years. A very strong NAND market, at least, is where your product mix is. What is your base concern now going into 2027?
K.S. Pua: I need to hire an extra 250 headcount to fulfill. I mean, my executive blamed me, "Hey, you go to ask so many projects, you win so many projects, now they have to work hard." So, okay, I say, "Okay, I give you 250 headcount." We need to hire more talent. Phison also announced next year we are going to pay a high amount of bonus, a very high amount bonus. So I believe this can attract. Of course, I make our current, our staff happy, and we can attract some more talent to join Phison.
Jeff Ohlweiler: How difficult is it to find good engineers, and is that mostly Taiwan or US or a combination of all over?
K.S. Pua: We still prefer in Taiwan because all our system here, right? How good depends on how much you are going to pay.
Jeff Ohlweiler: Yeah.
K.S. Pua: Now we are so profitable, so I don't think it's a problem to me.
Jeff Ohlweiler: Yeah. All right, another question online here. Many PC users now put MoE experts and KV cache on regular consumer SSDs. Is a normal consumer SSD durable enough for that, or do you really need an aiDAPTIV SSD? What is the difference in endurance?
K.S. Pua: Okay, you can buy the consumer SSD, then very soon the drive is going to crash. Trust me, okay? It's going to crash, okay? In the past, right, there are some automotive makers, they use a consumer eMMC design into the car. Then the car has to return after 6 months, and you know how much damage is going to happen, okay? You put any SSD, it's working, it's working, but at the end, you will suffer with data loss.
Jeff Ohlweiler: Okay, here's another question on inventory. How is the inventory situation in different types of end markets, and what do they see their customers doing? What do you think is the potential impact from YMTC growing capacity?
K.S. Pua: First of all, inventory in the PC retail downgrade, quite big amount stuck in China. Those OEM is back. Price is good, but you go there, the army coming back after 6 months. In India, industrial part, I think okay. Enterprise is far then okay because still very tight. So most inventory stuck in the consumer SSD. By the way, you can't believe consumer SSD market decline 90%. So when you just now ask me about Longsys, yes, they are suffering. Phison, we okay. Inventory to me, I move those component all go to enterprise and the pushback, so I'm okay. What's your second question? Jeff, just now you got question, right? Inventory and?
Jeff Ohlweiler: Yeah, the second question was, what do you think is the potential impact from YMTC growing capacity?
K.S. Pua: Okay. My personal thought is- Hello? 雪莹的, 雪莹的声音。拜托。哎呀。 No matter what in coming years, the flash will be very tight, very, very tight, including TLC, QLC. We wish, we like to see YMTC improve their capacity, they grow the output. At the end, every players, they have to invest. This will make the AI ecosystem much healthy. If not, the supply limited, the AI cannot move. At the end, the user have to pay much, much, much higher price to get the AI service. So, YMTC output actually is helping to make the AI systems, the smartphone, the PC go to much healthy supply chain.
Jeff Ohlweiler: Okay, Charlie, your hand is still raised. Do you have another question, or is that from before?
K.S. Pua: Maybe this one right here.
Charlie Chen: That is from before.
Jeff Ohlweiler: Okay.
Charlie Chen: All my questions. Thank you. I am sorry. Thanks.
Jeff Ohlweiler: Yeah, Kev, could you talk about your overall China exposure and how Hosin Global is doing given a tough consumer market right now?
K.S. Pua: Well, Hosin is going to IPO public in the Hong Kong market by next Thursday, 12 November. So congrats to them. They are doing quite okay, still profitable. Next year, of course, the consumer market is very challenging. They also start to move some to the enterprise business. Other than this, we are going to help them to move more into the enterprise hyperscale and the AI solutions. China, I believe, consumer PCs, smartphone, 2027 is challenging, unless we are going to push aiDAPTIV into AI PC. If this is a new growth, it can help the industry.
Jeff Ohlweiler: Okay, another question online here. A recently popular view suggests that power grid capacity bottlenecks are effectively stranding hardware, including memory chips, resulting in a relative glut. How do you assess this dynamic?
K.S. Pua: Okay. Power, I agree. Power supply is one of concern, but we can do nothing. I believe a lot of ecosystem in power, they try to solve a problem. Memory, yes. That is why we need the supplier to heavy investment, not only in NAND, but also in DRAM. If not, users have to pay more in the future. I just get one more message. PCB is going to shortage from coming month. Big shortage, material shortage. So everything shortage. Now the question is, if everything shortage, how Phison can secure every component, what we need can deliver on time? This we need to tell every supplier, Phison has a value in this business. We have a better position. How?
K.S. Pua: We need to keep delivering valuable products and services to the market, such as aiDAPTIV going to AI PC, aiDAPTIV going to AI on-premise, and 2027, please wait and see how this is going to be popular in the market.
Jeff Ohlweiler: Okay, great, Cass. I do not see any more questions on there. I think, maybe pass it back to you for final comments.
K.S. Pua: Okay. Last, since the last six months, we have been challenged, okay, low-cost inventory. Please do not come to us to mention anything about low-cost inventory, because low-cost inventory in the consumer market, yes, but not much related to Phison. They are mainly to the module house. They are not consistent investors into R&D. We are good because of heavy investment for many years, and the price we prove. When I start to talk about AI Data Platform to the market, they also ask, "Hey, who are you? Why do you have the confidence you can make it?" My answer is, "Sorry, I do not know if I can make it or not." But enterprise SSD last seven years, we proved it. In the AI Data Platform business, we invested three years already. We hired people, we keep investing, we engage.
K.S. Pua: I believe with the Phison heavy R&D investment, our commitment, our smart working hard, next year, please wait to see how we deliver our results to you, not only AI ecosystem, but also AI Data Platform. So, please trust us, not to blame us too much, not too much negative, please, because we are working hard. Today, we proved something saying six months ago, we will prove we are good. So, give us more encouragement. We will deliver more back to you. Thank you.
Jeff Ohlweiler: Great. Thank you, Cass. Thanks for including me, and thanks everyone for joining today. Have a good night.
K.S. Pua: Yeah, thank you.
Jeff Ohlweiler: Bye.
