【精選】美國「絕不允許科技泡沫」不給中國崛起機會！沈萬鈞揭AI永動機「賺錢時機點」明年應用出來開始獲利-沈萬鈞 劉寶傑 《寶傑怎麼說》

0:00 Who will reap the benefits of this AI process? China will reap the benefits. Oh. Because China has been restricted from the beginning, they have been forced to be self-reliant; so if this AI bubble in the US bursts, China will just rise on its own. If you look at it from this perspective, you'll realize how could the US possibly let AI just blow up like dot-coms? Currently, the incentive for AI is hollow, because right now it's just me making chips and selling them to those who want to build, and then they take out loans to pay me back ; there's no actual application involved, so you might think it's a scam, but the real AI applications will start to emerge next year. Once there are applications, there will be profits. Now you see the SOX index, you see Japan, you see South Korea, and the stock markets are experiencing

0:45 volatility. Taiwan was doing well, but once it crossed 46,000, you noticed that the selling pressure became extremely, extremely heavy; after that pressure became so intense, everyone realized one thing: there seems to be a massive shift in global capital, especially at the epicenter, the United States. As mentioned earlier, the US is under inflationary pressure, there's the Middle East crisis, rising oil prices, and rising bond yields, all of which are sucking capital in. With all that money being sucked away, wouldn't the levels in the stock market drop? And if the stock market levels drop, won't there be insufficient momentum going forward? If there isn't enough momentum, won't it become very difficult for the stock market to return to its previous highs? Actually, besides the volatility, there's something very peculiar lately: trading volumes across global stock markets have shrunk, yes. It's just as you said, Bao; everyone is observing, and while the long-term outlook is optimistic, AI isn't a

1:30 demand bubble, it's just too rushed. Too rushed. Too fast, and all the resources are too concentrated. Actually, if you think about it from another perspective, if all the companies were willing to... Not to rush, even with 350 billion in orders, I don't need to deliver so urgently, and I don't need to be so rushed to build data centers. If they release things slowly, they won't need to spend so much money, they won't have negative cash flow, and the capital expenditure market won't have to worry about free cash flow becoming negative. Right, but the thing is, first, they have issues with competition. And second, the potential demand for future AI applications is too enticing, so being too rushed and too fast will inevitably lead to these problems. So, for us institutional investors and those in the industry, when we look toward next year and the year after,

2:16 we’re optimistic, but the short term is full of doubt, which is why trading volume is trending downward. As you mentioned, government bond yields impact stock market valuations. Take the last time, when the Russia-Ukraine war triggered an interest rate hike cycle; US bond yields rose from over 2% to 4%, which relatively speaking, meant stock prices were essentially cut in half. That had a huge impact. Even though TSMC’s performance was great, its stock price kept falling. This time, people are less worried because this inflation seems more like a one-off issue, much like that war. Crude oil prices hit over $ 100 in the second quarter of this year, but looking at it

3:02 from next year’s perspective, even if it stays at $ 80 or $ 90, the year-on-year growth will be negative. So inflation should ease next year. The stock market is currently waiting to get through this transition period. That’s why we mentioned earlier that things took a sharp turn starting in July of the second half. Right. Then after a rebound, everyone started taking a wait-and-see approach again. The main reasons are, first, the inflation and interest rate hike issues , which will be most severe in the second half of this year. Next is AI— the concern is that everything is moving too fast. The demand is still there, but the capital can't keep up. Uh-huh. Then there’s the issue of mismatched supply and demand. As we

3:48 just mentioned, advanced packaging is in short supply, and TSMC has even said , "We’re actually envious of the memory chip shortages." The implication is that they are in even shorter supply than memory, but because they didn’t raise prices significantly, it didn't grab headlines. They didn't hike prices. Right, they didn't raise prices across the board. In reality, they are saying, "I'm in even tighter supply than memory." From our perspective, what they said is true ; advanced packaging is indeed the most bottlenecked link right now. HBM is also scarce. But think about it: with these things in such short supply, won't there be supply-demand mismatches in other areas that aren't as critical? Once you can't get advanced packaging or HBM, orders for other components might get pushed back. Like with Apple's iPhone, if you can't get those parts, you ship fewer units.

4:35 If you ship fewer iPhones, wouldn't that impact the rest of the supply chain? So, all these supply imbalances, excessive resource concentration, and insufficient capital, plus the valuations being back-loaded to the second half of the year, especially in ...That's a huge reason why the stock market correction is happening right now, exactly. Looking at the big picture, whether it's regarding chips or the return of capital flow to end-user applications next year, it's highly likely to be resolved by the end of this year. Why is there a bottleneck in the third quarter? Does passing the third quarter solve my funding issues? Does it resolve my interest rate problems? Can it solve my oil price issues? That's a great aspect, but if you look at their trajectory, take Anthropic for example, they are now

5:21 starting to turn a profit. Yes, they're making money. Their ARR is 14 times higher year-over-year compared to last year. And you see various companies, like Google, Facebook, or Amazon, starting to find business models for AI. Not to mention that the applications haven't fully emerged yet. For example, Appier is going public, and Tesla is preparing for mass production of Optimus. These are the real AI applications that will start to appear next year—perhaps mid-year or in the second half. Once these applications exist, there will be actual, tangible profits coming in. Right, like how LG told NVIDIA that they will definitely start shipping robots next year. Once all these things start shipping

6:06 and you begin to have positive free cash flow coming in, that is what we call the AI momentum. Right now, AI feels empty because it's just me manufacturing chips and selling them to someone else, who then takes out a loan to buy more. It doesn't involve an actual application or anyone making money yet, which is why it might seem like a scam. But if you consider that this perpetual motion machine will soon include people buying iPhones, robots, and electric vehicles, Then the logic holds up. So the main reason we feel next year will be better—and I don't mean immediately in the first quarter, but at least by mid-to-late next year—is that the concerns surrounding this perpetual motion machine will likely ease. After the last dot-com bubble

6:52 burst, it took seven or eight years for real demand to emerge. Take the iPhone, for example; it requires the internet to function. Without the internet, you can only make calls or play Snake. How else would you have all these apps? So, these developments paved the way for the birth of smartphones and apps. You have to pay attention to this. The dot-com bubble was a bubble created by the U.S. itself. Mmhmm. And in the end, it was the U.S. that reaped the final results of that process. If this happens with AI, who will reap the rewards of this process? Everything related to it—it’s a pity for Taiwan. If the U.S. bubble bursts—and we just mentioned OpenAI bursting— Who will step in and reap those profits ? China will.

7:39 Oh. Because from the very beginning, China’s engines, its infrastructure, and all its resources have been constrained by you. I’ve been doing it all myself, researching it myself. So if the U.S. AI bubble bursts, China will simply rise up on its own. Therefore, in terms of great power competition, last time with the dot-com bubble, the U.S. didn't really participate; it was just you guys bubbling, being too passionate, so there wasn't that flavor of national competition. The U.S. was like, "Well, you guys love playing this game!""You go ahead and build all these fiber networks and equipment when the applications haven't even appeared yet, and you're so eager, driving stocks so high. If it's a bubble, it's your own bubble." Then, after a while, others couldn't even figure out what kind of game the U.S. was playing. This time is different. With AI, you don't even need to talk

8:25 about China; AI is exploding everywhere like mushrooms after rain. If American companies like OpenAI were to collapse, others would take advantage of the situation, rise up, and snatch away all those applications. But when you look back, was there really that much money lost? Not really. From a national perspective, the money Nvidia has made more than covers it. So, if you look at it from this angle, you’ll realize how could the U.S. possibly let AI burst just like the dot-com bubble? Because if you collapse this time, others will directly harvest the market . In the future, all the engines will be Chinese, all the robots will be Chinese, and all the automation will be Chinese. If that happens, the U.S. is finished, right. So So you’d be doing someone else a favor. Exactly. All the things you worked so hard to build, you’d just be handing them over to someone else.

9:12 Of course, the overall speed of AI might slow down a bit, but the problem is the disparity in status would be absolutely massive. So from this perspective, why would the U.S. ever let AI wipe them out? First of all, it’s not like they’re just blowing smoke or building castles in the sky; the key applications are right around the corner. At most, in a year or two, they’ll catch up immediately. It’s just this one hurdle to get over, so why would it be so hard? This hurdle is just 2026 to 2027. Right, or 2028 at the latest. So I feel that, at our end, given our work, we mostly collect information, but you travel all across Taiwan. You’re saying AI isn’t just for TSMC or the companies we mentioned; you’re seeing many industries in Taiwan adopting AI—is the speed,

9:58 depth, and breadth of this adoption beyond our imagination? Does it mean that if you go into any industry now, everyone is working on AI? Actually, even back when I was at TSMC, my department was in yield analysis. I remember back then, our engineering team—we all studied electrical engineering—would analyze and find about 20 different problems, classify the issues, and run experiments. Now, AI can directly classify over 10,000 patterns and can even take a problematic machine offline in real-time. Right. It puts it on hold and then calls the person who needs to wake up. To take care of it. So haven't you noticed that in the past, whenever TSMC

10:44 switched to a new process, the yield would be lower, and gross margins would drop? But now, it seems so natural. Logically, the more advanced and difficult it gets, the harder it should be to produce. Now they have back-side power delivery and all these new processes, yet when they introduce these technologies, TSMC’s yield doesn’t drop. This is the biggest impact of AI. And recently, this year, TSMC has been building its own data centers. Oh, didn't they have them before? They likely had some of their own before but used third-party services for part of it. This brings us back to the core point: when you truly use these AI tools in an enterprise, sorry, but everyone needs their own. Oh. How could I possibly give my data to someone else? Exactly. If we wouldn't even give away

11:29 our own bank account details, then it’s even more critical for TSMC. So before, I might have trusted giving things to the cloud, but no more. Right , even if I’m TSMC, even if I put all my production details in the cloud, and even with encryption and all that, how do you know someone can't crack it? If I’m not running AI on my own servers, the level of risk is very high. So look at TSMC as an example; TSMC needs to build its own large data center to run its own AI. The same goes for all the world-class companies. Jensen Huang said, "I buy things from you, but in return, you have to buy things from me. " I heard an expert say, "How much do I sell you a chip for? But I buy an NVIDIA chip from you, and it costs 200,000 dollars." Because it creates a feedback loop.

12:16 So everyone is essentially asking, "How can you charge ahead so fast?""I'll help you make it, I'll buy your stuff; not only will I help you make it, I will buy your stuff." Everything in this loop means that if you are not in this race, you will be eliminated. Actually, for the tech industry, if you haven't started yet, you are really likely to be phased out. That’s why I’ve been giving speeches at schools lately; I think it has a huge impact on students. We have already finished our schooling and have experience; if we know how to use AI, we can quickly boost our productivity or technical skills, but it's different for students now. What they are learning might be completely useless in the future. You know, when software companies are

13:02 hiring people now, they still test if you know Python or other software, but these things are actually things you won't even use in the future. Right, AI will just do it all for you. But why do they still test you on it? Because those inside the companies don't even know what else to test. Exactly, they don't know. Should I test your IQ instead? Right? So you can see the impact is huge. We are wondering what big differences there will be for future kids. For instance, when I look at ancient civilizations, people back then had an average life expectancy of only 35 years. Right. But they were able to create things like the Maya civilization or Bronze Age cultures. Why? Because they didn't need to go through middle school or high school; they might have started working at the mental age of 12. The future is very likely to be like that.

13:48 All the things we learn can be shortened by AI to a certain extent, so what should our students focus on cultivating? Once your mental capacity is mature and you can use AI with your intelligence, that’s when you can learn professional skills. So, what do I learn in school then? Doesn't that mean 99%of what's taught in school is becoming obsolete? Socializing, for one. Your mental age needs to be mature enough, otherwise you're easily deceived or manipulated. So school is really just for social functions and making friends. It's very likely that mental maturity training will start from a young age. Take TSMC right now; you know, Taiwan and all these advanced industries aren't just short on power or cleanrooms—they're short on engineers . Oh. TSMC is now working with former public vocational colleges, building programs similar to the old cooperative education systems where you'd learn

14:35 things like hairdressing. Right. Or baking, but now the cooperative education is with TSMC. Working for TSMC. You might graduate from middle school and start learning how to use semiconductor tools and analysis as soon as you enter high school. So as soon as I get into vocational high school, I start learning it. Exactly, exactly. Germany has actually been using this system all along. Their cooperative education is now entirely based on this approach. Because the US is doing the same thing in Arizona now. Arizona even has a specific university department dedicated to TSMC, because otherwise, TSMC would have no one to hire in Arizona. And there's something else I find quite interesting: Taiwan's small size has actually become a major advantage for us. Like you said, it’s not just TSMC

15:20 ; there are many supporting vendors. I once saw data where Morris Chang said, "Within 200 kilometers, I can solve all my problems." Later, I saw a report saying you don't even need 200 kilometers; within 50 kilometers of any TSMC plant, all the necessary supporting vendors are fully available. If anything happens, I can handle it immediately. This is why South Korea, China, and other countries can't compete; only Taiwan can. In fact, TSMC is playing a very long game. This is something Jensen Huang didn't mention, but I can add: while Huang says TSMC's advantage is trust rather than technology, its greatest advantage is actually its partnerships. All these so-called little friends of TSMC are actually the key to its

16:06 victory. We’ve seen so much, we’ve visited so many things that affect TSMC's yield, so when this EMAS person said he wanted to build a FAB, he was eating a burger inside a clean room. We aren't doubting him, but you know, he might be thinking, "I want to beat TSMC, I want to partner with Intel ." But he forgot that what TSMC possesses is the entire corresponding ecosystem. For instance, when we look at it, even when he went to the central-southern region, he had a yield problem he couldn't solve, only to discover what was it? It turned out to be dust from chicken farms a few kilometers away. Because the dust was floating in the air, just a tiny bit, and when the air came in—you wouldn't have this in Hsinchu—your filtration

16:52 system wouldn't filter for this. This caused the plant to be strange; I clearly made it well in Hsinchu, but I shipped it to this central-southern location and the yield was slightly lower. Right. For TSMC, even 5 or 6%is unacceptable, and it was consistently low, so they found out it was the dust. You can't tell them not to raise chickens, right? So it’s simple, we just add an extra layer to the filtration process. Right, so you'll find that from this single example, you can see TSMC's excellence; TSMC's advantage is built by all its partners. Because the air filtration you just mentioned couldn't possibly be done by TSMC itself, so it’s the air filtration vendor who has to find a way to help you, so that even if very fine dust particles drift in, they must be kept outside. Yes, and when we look at

17:37 more, this is just a minute example, there are many manufacturing examples. For instance, chips used to be single layers, now they are stacked higher and higher. Each chip has some bumps or components, and stacking them is like a cake. If we make a cake, making one layer is simple, but making many causes it to warp. This is called warpage; there’s a stress issue, meaning things will shrink on the edges. Ah, this is very hard to solve. What to do? No one can solve it. One company says, "I have a material, I’ll stick it on, like a piece of tape." For example, if it warps here, I stick it on, and when I bake it again, it can be pulled back. Mm. Oh, it’s solved. Good, so he quickly finds this partner to collaborate. These examples are endless. Yes.

18:25 If you encounter these issues in the U.S. or elsewhere, just solving them might take you a year or two without success, because back you can't find the right people. You might go back to MIT or Harvard and ask them to help you solve it, but their perspective might be how to fix it so it's not uneven. Solving the unevenness is too difficult , it’s a physical limit, so you might eventually just say forget it, I won't do this. But Taiwan has a way of saying, "I have this, do you want to give it a try?" So you can see these kinds of examples popping up one after another. Even for the suppliers we interviewed, just the equipment for cleaning alone. Right. Their capital expenditure is over a billion. Cleaning specialists, cleaning specialists. Yeah, why? Because when they clean, they ensure there are absolutely no particles. Right.

19:10 Like us, we wash a car for a few hundred or a few thousand, maybe 2,000 for a supercar. Washing semiconductor-grade equipment with no particles costs tens of thousands per time. Oh. So you realize that in the end, what they do is ensure that everything I need to transport to the cleanroom is cleaned until there are no particles. So when you look back and think about Musk, he uses his brilliant mind to think, "If I just seal everything up, I can smoke or eat a burger outside the enclosure," but you know what TSMC does is level the entire factory floor. Right. Let's talk about an older example, our former stock market king, Largan Precision. Do you know why they are more capable? While doing plastic lens injection molding, to stack the plastic lenses, what level of precision do they need to

19:57 maintain? They drill into the ground. Hmm. Three times deeper than a typical factory. The foundation is three times deeper. When building their factory, the construction workers asked, "Why build such a deep foundation?" Building it three times deeper makes it more stable , reducing those very subtle vibrations . You know that twenty or thirty years ago, Taiwan's manufacturers—the boss of Largan was a doctor. Right. His precision-focused factory thought of this, and Report to you, did you know that when Largan was building this factory, they didn't have the money? We all know that the person who helped Largan build the factory back then, when he went to collect the construction fees from Largan, He said, "I'm sorry, can I take stock

20:43 instead of payment?" He even used his own stock as collateral; everyone who Googles it knows that when his stock was at the bottom, he sold it all at 150, because he said, "Fine, I didn't lose money." He later regretted it, saying he could have reported five or six thousand. Right, you know, when he was building the plant, he had no money, but he thought, "No, I have to build it deep." That’s why Taiwan has so many hidden champions; we are unique. Usually, people say, "I'll make money first, then improve," but no, we make it great from the first step. That’s Taiwan’s strength. When I looked at Largan, I noticed that besides laying the foundation three times, the distance between each pillar was shorter than others. And for the cement, while others would pour again in a week, he waited two or three weeks to ensure it was absolutely stable

21:29 before building up. That’s the true spirit of a Taiwanese craftsman. And what’s more, this was before anyone had even received an order. Before even getting the orders. People just thought, "Wow, this guy is too insane." Too insane. Anyway, as we are recording this today, there happens to be this touch event. People used to think Taiwan's panel industry was doomed—a "two-trillion-dollar disaster"—and thought panels had no future, no prospects, and were just miserable. But surprisingly, what's scary about Taiwan is that under the AI wave and TSMC's massive ecosystem, as the saying goes, "The only way to be invincible is to be fast." Things I never imagined, like these panels, have actually made a comeback in Taiwan. Taiwanese companies

22:16 'weakness is a lack of branding, but we are great at mass production. We excel at surviving through cyclical downturns by doing niche things, staying longer than others, and when the tide turns, we become the most profitable. It’s the same with every wave— passive components before, or memory chips recently. Like panels before; because mainland China produced way too much, the panel business itself isn't actually profitable. But they are all working very hard to transform; many foreign competitors just went bankrupt and gave up. In Taiwan, we are working hard to transform, and there are two things where cleanrooms are in short supply. So, you saw Innolux sell their cleanrooms to TSMC, although TSMC later complained that they were terrible,

23:02 since semiconductors and panels are vastly different. But nowadays, whoever has a cleanroom is king. They are still buying them up; you can see many idle factories being snapped up. This is their first transformation business opportunity. The second one is what we call FOWLP, the Fan-Out Wafer-Level Packaging. In the past, as everyone knows, wafers were circular. And it's quite simple: are our chips circular? Our chips aren't circular; I mean, they are individual square pieces. If I have squares inside a circle, there is bound to be some waste. Yes. No matter how you arrange them, there will always be waste, which is a pity. Furthermore, there is another key point about circular wafers that most people

23:47 don't know: they aren't actually flat. They have a "dishing" effect, like a donut—thick in the middle, lower on both sides, and thick at the edges if you look under a microscope. So, regarding the chips, the yield and performance are best in that ring area. That's why we often used to talk about buying graphics cards back in the day. When building PCs, we at TSMC would even go online to check which zone a chip came from. The performance difference is huge, as people use graphics cards for gaming. You could even choose which chip yours was. Because they have serial numbers. We realized it's just like produce; the best parts are picked by the premium brands. Why were computers from Sony or

24:34 other high-end brands more expensive? It’s because their chips came from that golden zone and were picked first. Even though they were from the same wafer. But different areas of the same wafer, like the outermost ring, were of lower quality. They might have lower yield and worse performance, and even the very center might be worse, so those would be sold at Guanghua Digital Plaza. Scattered, exactly. So, going square would solve this problem. Why didn't we start doing it sooner? Because it's too difficult to make square ones—such a large piece—and the technology wasn't complete, meaning you'd have to change everything. In the past, the carriers used for wafers were circular. It’s quite troublesome to fit a large square piece in, and the stress dynamics are different, so everything needs to be redesigned. Why the need for a change?

25:20 It’s because we’ve discovered that as advanced process nodes continue to evolve, yeah. Perhaps switching to a square format could increase our yield. Right, because there would be no waste. Exactly, no waste, and it's more uniform. Yes, and my performance might improve as well. As we keep pushing technology forward, as we’ve been discussing, any improvement that can be made becomes a subject of discussion, turning it into a real trend. But why does this shift involve panel makers like Innolux? If we’re talking about changing from circular to square, there are many ways to do that, so why do panel manufacturers have a competitive advantage? This goes back to the basics, because typically our wafer substrate is silicon. It’s glass—glass relates to what you’re used to, it's panel-related.

26:07 Since panel makers have been handling square formats all along, they’ve been developing in this direction and hold various patents and early-stage designs. Of course, this isn't fully ready yet; it’s still in the CoG stage. That’s why you see these companies starting to collaborate with them, and you’ll hear about many getting ready to move forward, but there’s still a way to go before true mass production. Oh, I see. Since we’ve been accustomed to the old ways for so long, and as you mentioned, all the infrastructure is set up for circular, the concept you’ve brought us today is truly a new one. Yes, what we have on the left is the original wafer design, and what you see on the right is FOPLP, which is the square panel-based approach. You can see the difference here: if I use a square format, it’s almost

26:52 entirely uniform, and I have virtually no waste, whereas it's a different story on the left. Yes, and in terms of performance, it’s more consistent. So , in the past, if my process wasn't that advanced and my chip complexity wasn't that high, the wafer size didn't matter as much. But as we reach more advanced nodes, this becomes an issue, which is why everyone is looking back at FOPLP. Looking ahead, we'll see this being applied in areas like automotive, AI, and mid-to-high-end consumer electronics. Another factor is output, which is the core issue—TSMC just can't keep up. Major players worldwide like Google, Apple, and NVIDIA are all

27:38 telling TSMC that their expansion isn't enough, even with TSMC reporting record-high capital expenditures for next year, potentially reaching 60 billion. This year's 50 billion is already staggering, but even so, everyone still feels it's not enough. I remember when we first started discussing TSMC, a 30 billion annual capital expenditure was already breaking news. That was already huge; it used to be around 10 billion. So, to demand 60 billion now, it's essentially doubling that again. Exactly. Moreover, even for older manufacturing processes, aside from 28nm and some purely consumer-grade chips, they are actually all at full capacity. Everyone is fixated on 2nm, but it's actually the 3nm line that is in the shortest supply, so... The shortages aren't limited to just

28:23 the most advanced tech. Every generation has its own iterative products—cars, and future robots haven't even used that much yet—and they're already at capacity. This leads everyone to wonder: since TSMC is fully booked and most others are too, where are all these robots supposed to be made? There's no capacity. Where are these cars going to be made? There's a massive amount of demand waiting in the future. But like you mentioned with Elon Musk, they talk about going to Intel or Samsung— didn't Musk try collaborating with Samsung before? Did those attempts fail ? In the end, were they unable to handle the pressure? Does this mean that, in the end, whether it's 2nm, 3nm , or other advanced nodes, TSMC is still the only option? I can say they won't necessarily fail. Oh?

29:09 But it won't impact TSMC's competitive advantage either. The reason is simple: manufacturing isn't just about the most advanced processes. You can see that if future demand for physical AI scales up significantly, it won't just require the most advanced chips; it will need a wide variety of different chips. For instance, low-orbit satellites and optical communications—the hottest topics right now—don't actually require the most advanced chips. Even UMC produces a lot of optical communication chips, simply because they don't require the most advanced processes. Just like when there was a severe shortage of automotive chips, the best one wasn't TSMC, because it has high-voltage processes. So, if you say using mask to make a fab, as long as they don't do anything reckless and have Intel help them make the chips they need, I think it's okay.

29:55 So the demand is just too high. Right, but TSMC takes the lead. They follow companies like Nvidia and AMD, constantly pushing forward and paving the way for the most advanced AI. This means Taiwan has captured the biggest niche. And as just mentioned, the hottest topic now is silicon photonics. There's a connection between panels and silicon photonics, where Taiwan excels. I think of a book called "The Infinite Game." Yes. What I mean is, victory or defeat isn't about one or two battles; if you can survive, you are the winner. With current panels, we in Taiwan have survived and held on. And now, there are many technologies, including those related to silicon photonics. As long as you have the technology, you can survive, and we can all find our place in this battle. In fact, what we’ve

30:40 always seen as Taiwan's greatest strength, as Baojie just said, is that they survive first, and when the time is right, they rise. Of course, stock prices sometimes rise faster because everyone is looking for new themes, but just as we saw in the past with things like passive components or substrates, Back when we looked at substrate manufacturers like Kinsus at thirty or forty dollars, people said it would get better. The numbers hadn't improved yet , but the smart money already saw the future shift in supply and demand for the industry. I think the panel industry is in that early stage. It's just beginning, so people watching the sector are looking at whether their old facilities can be revitalized to secure new funding for expanding into new technologies. How this evolves, I think , will take about one to two years.

31:27 Alright, Guan-jun, you often say your investment approach is to identify trends, then find industries, and finally pick individual stocks. You just mentioned the trend is photonics. So, going back to the individual stocks , including equipment manufacturers, where are the trends now? Hmm. Actually, because technology is constantly evolving, the next area related to equipment that everyone cares about most right now is SOIC. S So-called SOIC is like how memory evolved from DRAM to HBM, where they stack it up; instead of one, it becomes a stack of three. SIC also involves stacking three chips. It means that in a future chiplet package, specifically an SIC package, there will be three HBMs paired with

32:12 three ICs to form a single unit, so it's not just the HBM being stacked, but the ICs as well. And finally, silicon photonics is used for transmission because the data volume is too large. Actually, the simplest reason for using light is that there’s too much traffic. TSMC and Nvidia often say, why must we use photonics? If Neihu is already congested, and suddenly I have three times the number of cars, Right. It would be gridlocked, and no one would be able to move. No matter how good my car's performance is, if the roads are terrible, I still can't get out. So if you ask TSMC employees, they’ll say that if this CPO technology doesn't succeed, TSMC is in trouble. They even say, what’s the point if I make it but it’s useless? Exactly, I can’t drive my car out. Yes, it’s useless, so no matter how

32:58 difficult, they must complete it; that’s the key. As for SLIC, it will be used for all future equipment, as well as optical testing and verification equipment. Everything you see here represents the latest technology in our 2027 to 2028 ICs. All equipment must be bought in advance; if you buy what's available now, it's already obsolete. That's why you can see all equipment manufacturers moving in this direction, and their stock prices react the fastest. But I’m a total amateur when it comes to stocks, so I’m confused—what kind of equipment manufacturers are you talking about? For example, we previously mentioned Gudeng makes FOUPs, we know Kinik does grinding, and we know about cleanroom services or inspection firms. But in this new system you mentioned, which ones are the key equipment

33:45 manufacturers, and what do they actually do? It’s better now because the foreign investors we mentioned aren’t just looking at current EPS anymore; they’ve started looking at the evolution of technology. Like I just mentioned, when SY was discovered, wet processes were highly needed, so foreign investors are actually issuing reports on these Taiwanese wet process companies like Hungv, Scin, and others. And you'll notice, how come the EPS valuation is projected so far into the future? What is the main point? It tells you the direction lies here. When we talk with foreign investors, they break it down—future optical inspections have stages like Inspiration 1, 2, 3, and 4. They identify which companies are being integrated at each stage and which have business opportunities. Many Taiwanese companies already have products

34:31 undergoing verification, and foreign firms are actually issuing reports on this. That’s why you see companies like P; foreign reports for each such company might target 10,000, and their price-to-earnings ratios are extremely high. Why? Because they see them securing a spot in the future technologies we just mentioned. In the past, whether you were a fund manager or a retail investor, you would have had no way of knowing this. Fortunately, foreign investors now publish all of this; it's all written in their reports and in the news. So, I need to pay attention to these things from foreign investors. I want to report to everyone that not every foreign report reflects these things, but if they are discussing technological evolution, you need to pay special, extra attention at this point in time. Instead, don't worry too much about

35:17 their EPS estimates. Someone might say this firm pegs TSMC at 96 yuan for this year, while another pegs it at over 100 —no, nobody really cares. Right. What everyone truly cares about is TSMC's technology, TSMC's competitive dynamics; what people really care about are the future trends of these manufacturing processes, the CPUs, and the technologies used in optical communications. Conversely, if your technology gets phased out, it's over; the stock will truly plummet. Whether you can secure a position in it is very important. Right, so this brings us back to your personal experience. I find it very interesting: you studied electrical engineering at Tsinghua University, worked at TSMC for three years, then went abroad to study finance, and have been an analyst for 20 years, winning many major awards. You realize that now, financial operations, business logic, or EPS are

36:03 not the most important things. Look at many of the current managers and investors; they basically need an engineering background. Why an engineering background? Because I must have an engineering background to understand, and if that’s the case, I need to know if the person managing my money has an engineering background. Because this requires background research, I can then determine what the next steps are in future industries, what technologies are needed for those next steps, which Taiwanese manufacturers align with that, and whether these companies can secure that position. That becomes what you mentioned as the trend, and that is your investment logic now. Managers today actually have it quite hard. I don't mean that you must do homework to do this, but from my perspective, the knowledge that managers need to learn today is at least ten times more than

36:49 before. In the past, there might have been one or two major themes and five or six sub-themes a year. Now, there are hundreds of new technologies every quarter, and each technology significantly impacts stock prices. A company might double or triple in value once it breaks into that position. And these aren't small companies either . We used to look at the "elephants"— TSMC, Delta, ASE—and see them grow steadily, rising 10%to 20%a year. That's not the case anymore. They grow by several times over. Yes, several times over. Delta Electronics has risen by several times. So, whether you are an analyst or an engineer—no matter your background, if you are an analyst or a manager— you really have to study industrial and technological changes very deeply. Fortunately, there are many excellent analysts and AI tools you can use now.

37:36 But if you don't use them, and just rely on the past, thinking, "I'll just look at some technical analysis, check the chip positioning, sit at home and think about it, and decide that's the logic," then in today's market, you'll really need a lot more luck. My own experience was when I was taking a break last year, I was bored and did some homework at home. After doing some research, I realized the AI era was coming. Once AI arrived, I felt that electricity was going to be critical— electricity is crucial. I used AI to ask which companies in Taiwan are important for power, and it told me Delta Electronics. It told me that power management is extremely important , and I was very curious at the time: should I buy Delta? I remember Delta was under 300 yuan back then, maybe around two or three hundred. I'm not saying I was investing; I'm saying that based on my own feeling, when I saw this, if I had recognized this trend back then and knew that it was telling me electricity would be very important in the future—electricity is critical . There’s a TSMC in Taiwan; if I had invested in it back then, I’d have a

38:23 tenfold return by now. At that time, when I saw people starting to talk about silicon photonics, I didn't quite get it. But as you just said, aside from needing to run fast, I need a high-quality path for transmission in between. I used to think that investing meant looking at charts or checking financial statements, but not anymore. If you're looking at financial reports, you're already too late; you have to look at the future viability of the technology. You have to check if the company's technology is substantial enough. I actually feel that the investment logic is completely different from before. This has really been the case over the last three years. Yes. We have gone through many cycles, and each one is different, but honestly, this cycle is truly long. It has reached a point where all managers are forced to have this trait. If you can’t discern new technologies or identify industry trends now, you’ll easily be eliminated;

39:09 there’s no way around it. Alright, I know you often go to schools to talk about investing. Many students ask you, "What should I do? How do I build my own logic?" Can you give us a simple answer? I think this is very important. I think it’s very difficult to rely solely on your salary to get ahead these days. Unless you rely on, you rely on, you rely on technical skills—that would make you truly exceptional. Otherwise, generally speaking, I think investment has become a key factor for a person to get ahead, or at least not be left behind. Yes. So, whether you can invest or read trends isn't just about making a few extra bucks; it’s an essential skill to avoid being eliminated from society. Exactly. When I go to schools, the main thing I always share is that investing is crucial. It’s said that investment is the greatest invention in the world.

39:56 The key is that by investing, you can stand on the shoulders of giants. TSMC represents the collective wisdom of so many people and a massive amount of capital; with your money, you can invest in that. Some people say things like, "I'm single, let's end my' Singles 'Day,'" and buy one share of TSMC every day. Looking back at Taiwan, our GDP now surpasses Japan and South Korea. But have our salaries surpassed theirs? Exactly, so you need to rely on investing to stand on the shoulders of these giants and move forward. When I share with students, I say if your investment only focuses on price gains or "hot stocks," then that investment is meaningless to you. The core key to investing is me, standing with the giants and letting the giants make money for me.

40:41 So when we look at things, we look from an industry perspective; naturally, we think TSMC is great, ASE is great. There are many companies in the world in optical communications; I might not understand Taiwan's, so I can buy the best ones in the U.S., like NVIDIA— why wouldn't I invest in that? Exactly, I can invest in world-class enterprises and have them help me capture the profits of this trend; this is the best way for ordinary people to change their lives. Otherwise, look at it now; honestly, prices are rising far faster than our salaries, so if you can't use investing as your greatest weapon, the headwinds are quite strong right now. So we see that in this era, Taiwan happens to be riding a wave, but we certainly won't push you to buy any specific stock; however, as mentioned earlier, we invited Wanjun here today

41:26 because I admire his investment logic. You need to find the right trend, understand the industry, and then think about individual stocks, and when you go to understand the trend, you must do your homework; it's impossible for a stock to just fall from the sky for you to catch. So how do you catch the trend ? Now we have AI and various auxiliary tools, so we certainly hope to provide you with enough information. At least in this current phase, in this K-shaped society, the future changes in society will become increasingly drastic; can you use investing to ensure that, I won't even say you'll change your status, but at least you can maintain your current livelihood and security. Thank you so much, Jun. Thank you, thank you, thank you. If you enjoyed today's content, feel free to donate to support us—a cup of coffee or a cup of good tea is a gesture that helps the team produce better, more exciting content, and we

42:11 will read every one of your Super Chats . Yeah.