TIVP024: TSMC (TSM): The Most Important Business in the World? w/ Daniel Mahncke & Shawn O’Malley

28:07That being what has made them such special companies to own over the last few decades. And if not, will all of this capex bidding reduce the quality of these businesses over time? And yet with TSMC, as we've seen here, it is possible to still, if you have wide enough moats, you can still have a very high quality business, even if the business is highly intensive in a way that Google or Meta aren't. And if most of their capex is going toward growth, allowing them to produce even more cutting edge products that create even more value for the companies that use these chips, they can charge higher and higher premiums for those high end chips while also still manufacturing the lower end chips for less cutting edge purposes.

28:55And I'm thinking about like the chips that might power a smart refrigerator that's connected to Wi-Fi or something silly like that. But yeah, I'm not sure if there's any like a cannibalizing element there. where you are always unlocking the benefits of the next wave of leading edge chips with growth capex, but you're also disrupting yourself to some extent because you're previously cutting edge chips are now no longer at the forefront and then by default less valuable. So it's probably a little bit subjective how you define maintenance capex versus growth capex in this industry, since you almost have to invest in growth to maintain your competitive positioning, which is normally what we would think of maintenance capex but it's just such a fast changing industry regardless that the company is investing heavily in protecting and to differing extent still expanding its modes which is what ultimately matters that mode is what will enable the company to generate excess returns on capital for long-term investors and even if we want to stay in the framing of growth and maintenance capex.

30:02Since TSMC has a much higher yield than its competitors like Samsung or Intel, they're also below the industry average for maintenance capex, which leaves even more room for investments in growth, building out scale, and then once again manifesting their position. This brings us to the next point, which is the actual manufacturing process. And it might be the most sensitive manufacturing process in the world. The air has to be cleaner than in a hospital and vibrations from nearby roads can destroy an entire waver. You're literally building transistors smaller than viruses and billions of them and then layering them onto silicon wavers with nanometer precision.

30:46And because the process is so sensitive, the outcome can vary a lot for different factories and even more so for different companies. That's where TSMC really differentiates itself from the rest. TSMC has achieved a yield of over 90 % for some of the highest end knots, while Samsung has only reached 60%. And the difference in the newer generations, the two nanometer knots, is even larger than that. All right. So we've talked about how TSMC changed the game with the Foundry model and why chip manufacturing is one of the hardest, most capital-intensive businesses to be in. but let's dig deeper shall we because if we're going to say that tsmc might be the most important company in the world we really need to unpack what makes it so dominant what are the other layers of their moat what makes it so hard for anyone else even someone with lots of money to replicate tsmc's position and i imagine a company like apple or amazon would have the money to get into this business if they really wanted to.

31:54In our Amazon episode, we discussed that they will spend more than$100 billion in CapEx this year. So yes, money is not necessarily the problem. But as you know, sometimes you should outsource a part of your business when it's more profitable to do so. That has been the case for a long time. But of course, that's not the only reason. Companies like Apple or Amazon would create these chips in-house if they could. That would save them a lot of dependencies that they would likely want to avoid, especially in times of increasing geopolitical uncertainty. So let's break down TSMC's mode in more detail and find out why Manga would have probably called it a Lola Palooza mode, which is many competitive advantages stacked on one another.

32:40At the core, you have process technology leadership. TSMC has consistently been the first or at least the fastest to ramp up each new knot. They led the transition to 7 nanometers, then 5, then 3, and now they're even working on the 2 nanometers and it's coming pretty soon. And in an industry where scale is king, the first move advantage is crucial. TSMC's 3 nanometer knots are already accounting for almost 20 % of its entire revenue and growing from just 6 % last year. So there's huge growth rates in these higher and high-end Noughts. Combined with the other high-tech Noughts, they come for over 70 % of revenues.

33:22That's important since very few other players even compete in that segment. So that's where you lock in customers and that's where you can charge higher prices. But the older Noughts are also important for TSMC. So about this cannibalizing argument, there's a case to make for that. but you need those older knots since they tend to drive more profits than the revenue share would suggest. That's relatively intuitive since most of the R &D spend is targeted at these newer generations. GCNC doesn't report the actual profitability of its knots but just making some rough calculations based on how they account for depreciation and where the investments are going you can see that the operating profits of the older knots are higher than the revenue you share would suggest.

34:10And it's the opposite, obviously, for the newer nodes. So without mature node revenue, you're not only shouldering enormous upfront costs to develop the next generation of chips, but you're also potentially waiting years before those nodes become profitable. And that's also why yield is so important, right? If Samsung only gets a 20 % yield on the newest generation, most of its wafers will produce unusable chips, which makes the economics of producing them less attractive. And TSMC, by contrast, runs this flywheel where they have the profits from older, higher yield nodes, which fund the development of the newer ones.

34:54And because of its experience scale and ecosystem and everything we've talked about, TSMC will then basically typically ramp up new nodes with better yield sooner than anyone else. And that gives them this recurring first mover advantage, I think, that allows them to attract more customers and then that boosts scale even further, which again improves yield and so on. And eventually those new nodes become profit drivers themselves. And when the next node comes, the cycle repeats. But could you just give us some insights on who is buying the older ones? And once this technology is no longer the cutting edge, I'm just wondering what the end clients look like.

35:33You would think that people only want the most cutting edge stuff but also you probably don't need the most cutting edge stuff for all of the purposes that we use electronics or on a daily basis well and if you want the newest you also got to pay the highest prices right so for many products it's actually not necessary to have the newest ones it depends on the computing power a product needs so for example the latest three or five nanometer processes are mostly used by companies building high-performance chips. Think Apple's A-series processors, NVIDIA's AI GPUs, or AMD's CPUs. So these are performance-sensitive applications where power efficiency and die size really matters.

36:17But many industries like automotive or other industrials still run perfectly fine on mature nodes like 28 nanometers or even 65. So when we talked about Mercedes and how they want to increasingly turn their cars into digital products, most of the notes they need don't need to be cutting edge. But as we've seen in the revenue breakdown, Mercedes is not the typical TSMC customer, TSMC is mostly active in the high-end market. So they focus on the Apples and NVIDIAs of the world. And I guess that's where this Lollapalooza mode comes together that you used as a term a minute ago. And we know it comes from Charlie Munger, and as you kind of touched on, refers to having multiple advantages compounding together at once.

37:08That sort of makes them untouchable. And competition in the lower quality nodes is more intense, and margins are therefore less attractive. And meanwhile, TSMC's margins look more like a software business than a manufacturer's business, which is just incredible. Yes. Competitors outside of the cutting edge have operating margins between 50 % to 25%, and sometimes also lower than that, especially in cyclical downturns. We've mentioned it earlier today, but there are basically only two companies in the high-end market, and that's TSMC and Samsung. Intel is trying to get probably back into it, but it's struggling.

37:53Global Foundries and UMC used to compete in that segment, but they stopped some years ago because they were just realizing it's impossible for them to achieve the same level of technology and accuracy in building and manufacturing these processes. And on paper, Samsung is the only company that has kept pace in terms of announcing new nodes. But in practice, even they couldn't really deliver. Yields were significantly lower, which hindered volume. and that again impacted profitability. The Foundry division was operating at a loss of over$700 million in recent quarters and it actually had to shut down about half of its three nanometer lines just due to underutilization.

38:36And big companies like the Apples and NVIDIAs of the world, they need stability and reliability. The entire product timeline depends on getting TSMC's product on time. When they would shift to another company, they fail to deliver on their timeline due to lower yields and less volume, well then Apple shelves are empty as well and NVIDIA can deliver GPUs to Amazon, Alphabet or Microsoft. And as shareholders of Alphabet, that's the last thing that we want, right? And asked about whether NVIDIA plans on reducing its reliance on TSMC, its CEO, Jensen Huang, actually said this. Do we have any intentions to reduce our dependency on TSMC?

39:16No. The reason why we depend on TSMC is because they're the world's best. And they're the world's best not by a little bit. They're the world's best at what they do, like we're the world's best at what we do, by a mile. I mean, it's incredible how good they are at what they do. And it's not just technology. It's technology, it's scale, but it's also the deep trust that we have I don't think there's anything I can add to this. I mean, the trust he's talking about is often underestimated too because it's nothing tangible. But as I said, companies need to rely on TSMC or their entire operations fall apart.

40:07And especially back in the day, they needed to rely on TSMC to not just take their ship design and then end up stealing it or copying it. Today, that's a given, but it wasn't at the beginning of the Foundry business model. And getting to Intel, they are going through a pretty dramatic pivot. Historically, Intel manufactured chips only for itself. And they were really the gold standard for quite a while. But they fell behind mainly after delays with the 10 nanometer and then seven nanometer nodes. And now they are trying to basically reinvent themselves. They've committed over$100 billion in CapEx for new fabs in the US and Europe.

40:50And some years ago, they actually plan to be back on top of the industry by now. But looking at the numbers, there's nothing that would support that ambitious goal. I mean, last year, Intel's founder division lost$13 billion on less than$18 billion in revenue, while TSMC generated$41 billion in operating profit on$90 billion in revenue. Intel is no small company, But if these type of losses keep adding up, I think it can get even worse for them than it actually has been in recent years. Perhaps that's also why there have been rumors of a joint venture between Intel and TSMC. But right now, TSMC denies any rumor about that.

41:34And to continue the theme of acronyms in today's episode, what about SMIC? which is i believe china's largest and maybe most advanced boundary we already saw in our uber amazon and visa episodes that china is sort of always doing its own thing and doesn't really ever want to depend on another country's tech if they can avoid it so is tsmc's position so strong that not Not even China can duplicate what they do or provide a competitor here. Or is it true that SMIC is perhaps the Chinese TSMC? Well, I can definitely say SMIC is not yet China's TSMC. And it's also not looking like that anytime soon.

42:21So SMIC is growing fast and they're getting tons of domestic support, especially as U.S. export controls restrict China from getting the chips that TSMC produces. But SMIC still lags significantly behind TSMC in terms of technology and scale. And the company primarily produces mature node chips. So that's 28 nanometers and above that. And they're heavily constrained by those sanctions. They can't buy EUV equipment, so these ASML machines. And that's huge because there's just no alternative to those machines. Those EUVs are also so complex that it's impossible to reverse engineer how they work.

43:04So even espionage, which allegedly wouldn't be completely new to China, wouldn't help them catch up. So SMIC remains a strategic player for China's domestic chip supply, but it's years behind TSMC and I don't really see how they can get into cutting edge nodes and challenge TSMC anytime soon. Okay, well, we've talked a lot about TSMC's dominance, but that only matters if there's enough demand to support their business. And right now that demand is clearly there. You've got AI, autonomous driving, and high performance computing, and all of those and more rely on advanced chips. And that shift toward demanding more and more computing power puts TSMC at the center of a booming industry.

43:54And a few years ago, there were, I guess, real concerns about whether chip demand could hit a ceiling, especially as it became clear that it would be more difficult to keep up with the expectations of Moore's law and advancements in computing power going forward. And really the slowdown and kind of stagnation and growth in the smartphone market added to this and was a big worry. The smartphone segment was, for a period, the biggest part of TSMC's revenues. And I think that's not the case anymore. And I guess that debate just feels outdated now. We've moved past smartphones as being kind of the core thing that semiconductor companies are focused on powering.

44:40Just about everything is becoming a semiconductor product, it feels like even in a lot of industries, you might not expect. For anyone who listened to my John Deere pitch on this episode way back early on when we first started doing it, you will know that increasingly their tractors are powered by machine vision. And so of course, you're going to need cutting edge chips for something like that. So just to bring it all together here, how do you see the growth opportunities for TSMC going forward as long term investors in what we think is a really high quality business. How do we think about what type of opportunities will be before them and how that can change over time?

45:21The secular growth trend is as strong as it could get. And as you mentioned, that debate, which was basically held in 2020 and 2021, feels like ages ago, because nowadays you couldn't even imagine how the world would look like without TSMC's products. So to understand where TSMC's growth will come from, we really do not have to look much further than AI. But one sentence on the fear that Moore's law could come to a hold, this would have actually happened if not for ASML's EOV machines. And I know I bring them up a lot today, but they're just at the core of what makes this so impenetrable for both ASML and TSMC.

46:02And it really seemed impossible to build those machines. As a competitor Nikon didn't even try because they actually thought this is just impossible. So ASML is the reason why TSMC can do what they do. And TSMC in turn is the reason that pretty much everybody else can do what they do. High performance computing, which includes everything from NVIDIA's GPUs to Apple's M series chips, now makes up over 50 % of TSMC's total revenue. And a big part of that is being driven by AI accelerators. Basically, the entire AI supply chain relies on TSMC. That's part of the reason why you saw sales growth immediately spiking after investments.

46:46Additional capacity and newer generations of nodes are in instant demand. In 2024 alone, TSMC said that revenue from AI-related chips more than tripled and they're expected to double again in 2025. And even the sectors that we have only mentioned as demand drivers for lower quality chips like automotive and you just mentioned John Deere will become TSMC customers more and more and modern day EVs have hundreds of chips in them. Sensors, power management, cameras, connectivity, ADAS systems. Some of these will still run on mature nodes But the more advanced driver-assistant platforms, they will definitely push into 7 nanometers and below.

47:32So TSMC's automotive revenue actually hit a new high last year. And it outpaced smartphone growth by two or three times in many of the last years. Companies like Tesla, Waymo, which I know you are very bullish on as well, and even more traditional car makers are working on self-driving stacks. And the further they get, the more they are basically becoming semi-companies. i was in austin a few weeks ago for a wedding and and it was so cool to see those waymos up close and personal just watching them zip around the city with no drive is something out of a movie the future is here daniel we are living in it it'll soon be just another mundane part of our lives that we take for granted to see these self-driving cars all over the place i know that i think they're coming to dc next year which is the area i grew up in and they're going to be all over the world probably in the next five years.

48:26But for now, it just feels so innovative and exciting. And I can only imagine how many semiconductor chips are needed to power these vehicles. And obviously, that's only going to ramp up over time. Waymos are only going to expand into more cities and have bigger fleets. And all of that is incremental new demand for TSMC. Absolutely. And you said it before, TSMC used to rely on smartphones, and that still makes up a third of their business. But what we are seeing with this diversification of growth engines with AI and high performance computing leading the way, this diversifying shift also reduces the cyclicality, which we've seen in the semiconductor industry for quite a long time.

49:10And when you look at all their customers and the fact that only TSMC can provide all those services, it feels like they're actually the bottleneck in the semiconductor industry. And that's a great position to be in if you are TSMC? My main concern is that if the AI hype cycle ends up looking anything like the dot-com bus, where everyone got way ahead of themselves about new technology, then that pullback could be damaging for TSMC, even if long-term the promises of AI are legitimate. We're in such a dynamic period right now. All we can really do, I think, is just be humble about what's going on and how quickly things are changing.

49:53But up to this point, the TSMC story otherwise sounds almost too good to be true. They've got the tech lead, they've got the customer trust, and they're sitting right in the middle of some of really powerful secular growth trends. And the part that we cannot ignore is the geopolitical risk. For all of TSMC's strengths, it has that one glaring vulnerability. Almost all of its production happens in Taiwan. And in today's world, Taiwan is not just a manufacturing hub. It's also, unfortunately, the scene of one of the biggest geopolitical tension points. And we've already scratched the surface of this a little bit in our Alibaba episode way, way back.

50:39And while I don't expect that we will find an answer to a question that simply doesn't have answers, we still have to address this risk, especially today, since this tends to be the reason why many of the best investors in the world have bailed on TSMC or just never owned it. And for example, Buffett bought a stake in TSMC back in 2022 for about$4 billion and sold it again after just a couple of months. And we know that that is very untypical for Buffett. It's very unlike him. And I guess it just shows that he didn't feel comfortable owning it. Maybe he was losing sleep after buying it. I don't know.

51:17Something got him out of it. Yes, Buffett praised the company a lot. But it proves the famous saying, at least to me, that you need to own a company to actually know it. I've experienced that multiple times. And it seems like it has been the same for Buffett here, too. And I guess he looked at the company and he just couldn't resist buying it. I mean, few companies fit the phrase a wonderful company at a fair price better than TSMC. But the fair price is accompanied by a tail risk that could result in a total loss. So you cannot really say there's a huge margin of safety here, no matter how the financials look.

51:54And Buffett addressed this, although in his typically reserved way in the 2023 shareholder meeting. And it's probably best if we listen to him directly. So here's the clip. Taiwan Semiconductor is one of the best managed companies and important companies in the world.

52:17And there is not, and I think you'll be able to say the same thing five or 10 or 20 years from now. I don't like its location and I've re-evaluated that. I mean, I don't think there should be any place but Taiwan, although they will be obviously opening up chip capacity in this country. And actually, one of our subsidiaries that we got in Allegheny is participating in their Arizona construction activities. But we would rather have the same kind of company. There's nobody in the chip industry that's in their league, at least in my view. And the man that was a 91-year-old or so that connected with us, that I think I played bridge with in Albuquerque.

53:16and marvelous people, marvelous company, but I'd rather find a marvelous people, and we'll find it in the ship industry, marvelous people and marvelous competitive position and everything. I'd rather find it in the United States. I feel better about the capital that we've got deployed in Japan than in Taiwan. I wish it weren't so, But I think that's a reality. And I re-evaluated that in the light of certain things that were going on. Charlie? Well, my view is that Warren ought to feel comfortable if he wants to.

53:59It's just typical Munger how he only said one or two sentences, but you immediately get so much from it. I think he would have held on to the position if it were up to him. But Buffett was seemingly not willing to take the risk that a worst case scenario could wipe out their investment, which is ultimately the biggest threat to compounding wealth over time is having some kind of major wipeout that just sets you back. And actually, funny enough, I was there for the 2023 shareholder meeting, which was Charlie's last. And in fact, there were definitely a few questions that I can remember related to TSMC and just people trying to figure out rightfully so, hey, what caused this pivot?

54:39And my gut feeling at the time is it's just none of the answers were all that satisfying. As Berkshire shareholders, we really did not get a lot of resolution on this one. I mean, to be fair, Buffett in the last couple of years has kind of made it a habit to build a position and then sell out of it pretty quickly. So we don't know exactly what happened with his TSMC position. We can only speculate it has been because of geopolitical tensions. That's also what they mentioned. But one of our other portfolio companies like Ulta Beauty, he also invested in and sold out of it pretty quickly. So we never really know what is their thought process with these positions.

55:18But we also touched on this whole topic in one of our calls in the intrinsic value community. And it just showed me how important it is to gather opinions from many intelligent people and diverse backgrounds, because both of us talk about it every now and then, not every day, but quite often. And when we then can also hear a dozen other opinions on it and look at angles that we haven't yet considered, it's just extremely helpful. And maybe to summarize the China-Taiwan situation very briefly, in case any listener isn't aware of what we're even talking about, Taiwan is a democracy and it operates independently despite many people having Chinese roots.

55:58It's not a part of China and it doesn't want to be a part of China. And the Chinese government sees that a bit differently and has made reunification with Taiwan a long-term national goal. And China has repeatedly held something like military drills close to Taiwan, violated their airspace and run cyber attacks. And China also said it would consider military action to take control over Taiwan if no peaceful reunification can be achieved. And by now, TSMC is making serious efforts to diversify globally, building new fabs in Arizona, in Germany, and also in Japan. And the Arizona project is by far the biggest.

56:38$40 billion of investments that will likely turn into more than$100 billion in the years ahead. And the Arizona plants are supposed to produce not only those older, less quality nodes, but also those high-end nodes. And that should already start in 2028. The investments in Germany and Japan, those are smaller and the focus will lie on nodes for the automotive sector, consumer electronics, and also industrials. I almost wonder if this expansion for as good as it sounds isn't a good thing for TSMC because you know if you have all this almost trapped expertise in Taiwan and then all of a sudden you start diffusing it's a good thing for TSMCMC because you know if you have all this Obviously, with these global expansion plans, even still, the majority of the most advanced production will probably stay in Taiwan, I think.

57:46And I mean, we just talked about how incredibly complex this process is and how TSMC built an entire ecosystem around its factories in Taiwan with suppliers, advanced engineers, having all these processes perfectly dialed in. So this is not a short or even intermediate term risk that I'm kind of musing on. And if no one else, not Intel, not Samsung could catch up, even with deep pockets and world-class talent, maybe it's fair to say that it would actually be hard for TSMC itself to just copy and paste that environment somewhere else to try to recreate its own playbook. So even as they expand abroad, how much of the volume for these cutting edge nodes can actually be produced outside of Taiwan, if any, anytime soon?

58:35You just mentioned one of the biggest problems, which is replicating that ecosystem in the US or basically anywhere else in the world will be incredibly difficult, even for TSMC. So we have already seen some delays with the Arizona fabs because TSMC had to fly in engineers from Taiwan to the US to kind of teach them how to go about the machines and all the processes to increase yield, improve margins, and also increase volume. And at this moment, all of TSMC's AI-related nodes are still manufactured in Taiwan. And this is supposed to change in the next years when everything remains on schedule with the Arizona fabs.

59:15But even if that's the case, Arizona will only build a single-digit percentage of the entire cutting-edge nodes in the US. And despite sounding counterintuitive, there's also an argument to make that keeping production in Taiwan is actually part of what keeps them protected. There's this theory, which is called the silicon shield. That's basically the idea that Taiwan's strategic importance in the semiconductor industry actually deters China from taking more aggressive action. A war would not just hurt the West, but also China, since they rely on TSMC chips to power their own economy as well. Although to a lesser extent, due to the sanctions that we already discussed.

59:59So while TSMC is still allowed to sell to Chinese customers, there are US export controls in place, and especially when it comes to those advanced chips. And one of the biggest impacts was the loss of Huawei as a major customer in 2020, which just losing that customer alone cost TSMC billions in annual revenue. And since then, the rules have tightened further. So limiting what TSMC can produce for Chinese companies using US tools and US technology. So while China still accounts for a decent chunk of TSMC's revenues, there's always been that risk of new restrictions and that could in turn hurt TSMC's business.

1:00:40But the importance to the West also gives Taiwan and TSMC in general leverage to demand help or protection if the situation with China should get worse. because Western countries just have such a strong interest in Taiwan being independent and in TSMC to operate freely. And thus, in theory, this should kind of help Taiwan to stay independent. And in fact, TSMC's CEO has said outright that if a war would break out, the fabs would become basically useless and the people, the tools, the supply chain, they would all be gone. So taking TSMC by force just doesn't work out. We actually talked about this a bit on a call with our intrinsic value community one time.

1:01:26And after going down the rabbit hole in Taiwan and China and the risks of a U.S. intervention, it just feels like, wow, are we even talking about investing anymore? Like, what are we doing here? If you have to seriously debate the potential of World war three and not in some like loony paranoid way but in a legitimate way where we're the company we're betting on is literally at the center of the discussion around what may or may not prevent or cause a war between the u.s and its allies in china it just feels like you immediately should go in the too hard pile i could come up with a bunch of reasons to comfort myself into believing nothing bad will happen but unfortunately i had that hope with ukraine in 2022 and that optimism proved to be very wrong.

1:02:12And look, I know that China and Russia are completely different countries with much different interdependencies with the rest of the world. And so obviously, I'm not saying that I expect the same thing here. But again, that experience just really humbled me, at least, to consider that you can't just set geopolitics aside, especially when you have these great power struggles that are unfolding across decades and centuries and are not driven by the kind of rational economic logic that we are considering when we're looking at businesses. So there's just very, very different and complex forces that can drive these things.

1:02:52And I guess this is a long winded way of saying, I'm not sure I feel comfortable always trying to be in that position of explicitly or implicitly making a bet on those forces in a very direct way, where at least with other companies, the impact is a little less direct. So I want to be open-minded here, though, and I don't want to rule out TSMC yet. So maybe I will try not to let my paranoia keep me away from appreciating what is truly an excellent business with TSMC. I think we will always end up where we have to analyze the risk. and when it comes down to whether one is okay with this ever-looming risk or not.

1:03:35For me, what matters a lot in answering this question is valuation. While this situation is different for a Chinese company, the question also played a role when I looked at Alibaba and at$70, I was fine with the risk. But since the upside was so large, that just changed the whole conversation. So the question we have to answer now is, is TSMC's upside equally large to Alibaba back then? The first thing that we can say is TSMC is not as cheap as Alibaba was. Although it does trade at a significant discount to companies that should be comparable, not necessarily because of their operations, but because of their dependency to TSMC.

1:04:17Apple cannot produce its iPhones without TSMC. NVIDIA cannot produce their GPUs and the entire AI industry would come to a halt. For some reason, NVIDIA still trades at a PE of 47, Apple trades at a PE of 32, and despite only growing revenues at 8 % CAGR in the last five years. And iPhone's not growing anymore since 2021. So if you believe that TSMC is as important as every industry expert says, I think the PE shouldn't be at 19, especially if you consider its five-year revenue CAGR of 22 % and the net income CAGR of 26%. I mean, there are other factors that play a role here as well. That's completely obvious.

1:05:01And this sort of relative valuation is also always dangerous because if the other companies are just overvalued, then at some point, if they get revalued, the entire investing case for TSMC would break down. But it is an interesting angle to at least understand just how differently the market chooses to price in the risk of this China-Taiwan situation for different stocks. There's this famous saying that during the gold rush, the people who made the most money, they were the ones selling the shovels. And during the AI rush, it might as well be TSMC. They don't care which AI model wins, which company comes out on top or if the market will be commoditized as we discussed in previous episodes because they would make money selling their leading edge products to every company or to every industry.

1:05:52They are always there and demand for their product will be as well. Well, okay, I'm excited then let's do it. We've done the relative valuation. We talked about the geopolitical risk. We've gone through the whole business. Try and value TSMC. hey, let's see if there's a way to price in that looming political risk. And maybe you can help me feel more comfortable with it because I always seem to be a little extra nervous about these things just because of some past investments that I made that were impacted by geopolitical risk. So relative to the actual cash it's generating and likely to generate, let's think about how we can account for that political risk.

1:06:33How did you go about valuing this company, Daniel? I basically broke down revenues into three main drivers. The first one is wafer volume. That's simply how many wafers TSMC actually chips. After a rough patch in 2023, the volumes dropped about 20 % due to industry-wide inventory corrections. Things started picking up again in 2024. And TSMC shipped around 13 million wafers. That's up 7.5 % from the year before. and 2025 will be a lot stronger. And the second driver is the mixed shift. Every year, a bigger slice of the pie moves to 3 nanometer and soon 2 nanometer nodes. Those waivers cost a lot more than the older nodes.

1:07:17And even if the posted price never changes, just moving share from 7 nanometer to 3 nanometer added roughly 12 % to wafer revenue in 2024 by itself. Going forward, I dial that back to a bit more sustainable 3 % per year for mix alone. And again, that might sound very conservative, but it takes time to upgrade to new nodes again and again. So after one new generation, the shift will naturally slow down for a couple of years and then draw back the average. And then the third driver is the price per node. TSMC has a lot of pricing power for the reasons that we have discussed. and a 3 nanometer wafer today goes for about$20 ,000.

1:08:01For comparison, 5 nanometer and 4 nanometer wafers are typically in the 16 to 18k range. And older nodes go for much less. So even though total wafer volume is still below 2022, TSMC's revenue per wafer has jumped. And that's what helped drive over 30 % revenue growth in 2024. And going forward, that pricing trend is likely to continue, especially in the leading edge nodes, TSMC plans on another round of price hikes in 2025 with 4nm and 5nm wafers, increasing their prices by as much as 10 % and about 4 % for the higher end knots. So I give them 4 % average selling price lift per year. And again, the best case scenario can be higher here, but combined, this already gives us revenue growth of about 18 % per year.

1:08:51And that's in line, actually a bit lower than most estimates. And TSMC projects their revenue growth for 2025 to be 24 to 26 percent. It's more than twice the growth of the broader foundry industry. That kind of top line growth strongly implies a return to double digit wafer volume growth, especially with surging demand for AI accelerators and a more normalized inventory environment. So Daniel, before your pitch, and we just said the numbers pulled up on screen a minute ago, I was not aware of just how good these numbers are. I mean, these are insane numbers. A company of TSMC size should not be growing at rates north of 20%.

1:09:36And if they do, they should not be trading at a PE below 20. You just don't see this stuff anywhere in the world. I mean, this would have been like some of the MAG7 companies a decade ago. Not to say in terms of valuation and growth. Not to say that TSMC will have the same results. but just that's kind of the idea here it doesn't make sense and i think this clearly reflects that the market is obviously applying risk premium to this company i'm not saying anything new with that but perhaps though it's not fair if we compare it to nvidia or apple considering their reliance on tsmc but at the same time that this is not the first time we've seen such a big discrepancy in valuation between u.s companies and chinese companies or in this case taiwanese companies And again, the same argument applies to Apple when we talk about its dependence on Chinese manufacturing for its iPhones.

1:10:30It seems like the market thinks that Apple or NVIDIA will eventually find a different supplier when or if geopolitics become a bigger problem. Whereas with TSMC, the market recognizes correctly so that it's just in a much more difficult position to shift its production and dependencies if it needed to in a crisis. And in the long run, that's probably just true. But imagine these companies struggle to make products for just four or five years. In the end, the case for TSMC is based on the idea that replacing TSMC will be incredibly difficult. And that's what protects it, both in terms of market position and perhaps also geopolitically.

1:11:14And we already talked plenty about the qualitative factors that show that TSMC must have a pretty impenetrable mode, the numbers paint the exact same picture, I mean we are still talking about a manufacturer. A manufacturer with margins that look like it's a software company. Gross profit margins in the mid to high 50s, operating margins in the 40s, and net income margins in the same region. For the rest of my model, I just asked myself whether I want to model it based on EPS, so earnings per share, or free cash flow. I was asked a couple of times how we decide which approach to choose for what company.

1:11:50So I thought this might be a good opportunity to elaborate on the process because the answer is not always easy. I hope this is not too boring. And for everyone who wants to dive a bit deeper into the models, as always, you can download those models for each company in our free newsletter and in our community where there's a space where you can find and download all the models in just one place. In TSMC's case, earnings per share are a pretty clean yardstick. It already absorbs a lot of capex through depreciation because with TSMC, roughly 20 % of revenues flow straight into depreciation and that's booked.

1:12:28So basically accounted for in costs of goods sold. So earnings per share doesn't ignore the investments that TSMC makes. The upside of using earnings per share is that it tracks the earnings power of the business way better. TSMC reinvests at world-class returns on capital north of 20%. So every new fab more or less immediately grows their business. And that compounding shows best up in earnings. And due to its less volatile nature, it makes applying a range of exit multiples a lot easier as well. Now, the flip side is to turn around and take free cash flow per share. Free cash flow is brutal honesty.

1:13:08Net income plus depreciation minus every dollar of capital expenditure. It tells us how much exactly is actually left to pay dividends, buy back stock, or pay down debt. The problem is that TSMC's CapEx is pretty lumpy. In a heavy build-out year, free cash flow looks anemic and even though that spending is what protects the mode, valuing the stock through free cash flow can make the business look chronically expensive when in reality the cash is being plowed into three nanometer and two nanometer capacity that will earn significant returns later on. If TSMC wouldn't have a huge cash pile of almost 90 billion dollars and would the returns on capital be lower then I would argue you should use free cash flow because it's the more honest metric and it can assess the risk of TSMC not being able to invest into the business, pay dividends or buy back shares.

1:14:04But since that's not the case, I personally prefer the EPS approach because it smooths the CapEx cycle without ignoring it. And I believe free cash flow tends to understate TSMC's earnings power if we look only over a five-year time horizon due to the cyclical nature of the business. If you would model out 10 years, that would be a bit different, but I'm always hesitant to make predictions over such a long timeframe, especially in an industry like semiconductors. Very insightful, Daniel. It's a very subtle point you're touching on, but in certain cases, focusing more on reported earnings can make more sense than others.

1:14:44Just looking at free cashflow is a better picture of reality. And to me, that's one of those things set is the hallmark of a sophisticated investor like yourself, Daniel, of thinking through what version of the numbers is painting the most accurate picture of reality. And then I'm going to focus on that. I've seen that you still include free cash flow in your model that I was just kind of looking through it while you're talking and kind of just to double check whether CapEx and dividends can be financed by the cash that TSMC earns, I'd imagine. So it's a good compromise to still look at both. And doing all of that, what do you say?

1:15:20What does your model suggest for an intrinsic value and a fair value? And how attractive is the opportunity with TSMC in your opinion? So using a 12 % discount rate to account for the significant risk coming from geopolitical tensions and the next multiple of 25, which to be fair is higher than today, but I really think this company is worth it. Well, then we would get a present fair value around$190. So basically today's price. Due to the dividend payments, we could expect an annual return of around 14%. As always, this is just an estimate based on my assumptions, and this is not guaranteed in any way.

1:16:02Change up some assumptions and you can get totally different results. But based on this, TSMC, at least in my opinion, looks really attractive. Even at an entry price of$220 on paper, you could still achieve a double-digit return. I think it would be interesting, Daniel, to see in your valuation and maybe rather than just doing a higher discount rate, actually trying to add in a weighting where you say, okay, let's say there's a 5 % chance that TSMC's factories get destroyed in a conflict and the stock basically goes to zero. And how does that affect the valuation? because really the company maybe doesn't deserve this high discount rate just based on the quality of its earnings in a way.

1:16:44I know we use a higher discount rate with Nubank, but that's also because this is a company operating in an emerging market where there's some concerns about the quality of their earnings and the financial risks they're taking by making consumer loans to lower income populations in an emerging market, not to relitigate that whole conversation. And with the point being though, Taiwan is a very developed country and this is as high quality of a business as we've seen. TSMC's clients are companies like Apple. So on paper, their earnings are just, I mean, they're high quality. The only thing is you have this black swan risk.

1:17:18And I don't know, as I'm trying to think about it in real time here, I'm just, I'm not sure whether I feel that adding a 12 % discount rate is the right way to account for of that, just to push back on you a little bit, Daniel, I would almost prefer, and maybe you did this already, I don't know, but maybe I would almost prefer a scenario analysis where I kind of explored what the company might be worth if you use a more normal eight or 9 % discount rate, like we typically do for the high quality, large cap companies that we look at, that we feel like have good notes and actually just show the effect on the current valuation.

1:17:53If you had this kind of sensitivity analysis of where you said okay but with x percent chance of the company going to zero in a worst case wartime scenario how does that affect the present valuation so if there's a 25 chance a 10 chance a five percent chance and maybe a one percent chance it's kind of up to the investor to decide what the range of probabilities looks like which again is why everybody can look at this and come up with totally different numbers for what they think tsmc is worth today because obviously a 25 % chance of the company go to zero where you're like, you think there's a one in four chance in the next decade that China invades Taiwan versus a 5 % chance.

1:18:31The math is just very, very different, obviously. And personally, I'm not going to pretend like I know the right percentages, but it's helpful to say, okay, even with including a seemingly very high probability of the truly worst case scenario, if the stock still looked undervalued, that would just be very, very compelling to me. And so my question for you, Daniel, is did you do anything like that? I actually did that. And it's also because on a prior call in our community, we actually discussed this approach back then for a Chinese company, but I thought, why not apply it now? So to once again, quickly explain why I also use an approach where I just increased the discount rate.

1:19:12I did that as one way to account for the heightened risk, because you can use the discount rate as a way to figure out how much returns market participants expect from a stock to get compensated for the risks if you back solve it. Of course, under the assumption that it doesn't come to a black swan event, this helps to understand what discount investors apply to TSMC compared to, for example, Nvidia and the Apples of the world. But getting to your method, what I did is that I estimated the probability of a total loss. So basically the worst case scenario, which would be a military conflict between China and Taiwan.

1:19:49And then I calculated the compounded probability of a complete loss over five years. You could also use 10 or 15 years, whatever the time horizon you want to use. Of course, the longer the time you forecast, the more the likelihood of a total loss compounds and the lower the fair value of the stock. And since we're now accounting for the Black Swan event explicitly, I used an 8 % discount rate instead of the 12 % from before. As you can see in this table. At everything below 2.5 % probability of a total loss, today's price still represents a significant upside. Around 5 % you reach today's price level and a 10 % likelihood of a total loss for each year for the next 5 years.

1:20:31That would result in a downside of about 28 % or an annualized return of about 4%. These probabilities might sound low but personally, I wouldn't consider a 5 % likelihood every single year going forward as low. And since I was already in the process of modeling, I also created a quick three-state model that accounts for a third scenario, which is neither a total loss of capital nor the status quo. Essentially a scenario of severe disruption without total loss. Maybe China takes over Taiwan and the stock crumbles, but TSMC can actually still keep producing, or they have already diversified enough to keep international fabs running.

1:21:13There are no boundaries to your imagination here. I just assumed a salvage percentage of 35%. So if you want to change any numbers or you want to see how different probabilities change the stock's fair value, then you can download it for free in our newsletter. When there's a 3 % disruption probability and a 1 % wipeout probability over five years again, then the fair value of the stock would currently be about$200 per share. I know all of this sounds somewhat arbitrary, but it may help you give a sense of the risks and how they impact the valuation. That is very, very interesting, Daniel. I think that would kind of change my inclination to invest in something like this if the implied probability of loss has to be that high for it to be a negative expected value bet.

1:22:03I think I could get around to being willing to bet on the quality of a company like TSMC while also betting that the odds of a worst case scenario are, let's say, lower than 15%. And I don't know, I can't believe I'm really saying that because I really do believe the point of investing is to minimize the risk of total loss. But I can see with sub-tempting why TSMC is attractively valued and such a great business. and I've talked to a lot of people too who are really smart about semiconductors and are like, hey, this company has a moat. And to some extent, you're kind of deferring to other people's expertise, which is okay.

1:22:44And the other thing is that we can size the position accordingly. And that's another way to manage the risk. I'm going to say, go ahead and guess that we don't want to make this a 10 % portfolio position. We might not even want to make it a 5 % portfolio position, but I don't know. How do you think of it? TSMC is such a phenomenal business that it might be worth taking some extra risk on this. They are diversifying operations and it's strategically so important that it's in the best interest of pretty much all countries to keep them running. And yet, if one believes China is taking Taiwan over by force and that's imminent, no expected value math could justify an investment.

1:23:23Personally, I feel more confident with investing in China than in Taiwan just because it feels like Taiwan is this pawn between China and the US. In a scenario of more threats, the world and especially the US can sanction China. And that's what you often say. The risk for an American of not being allowed anymore to own Chinese stocks is definitely there. And that's also a risk for a European investor or everyone outside the US. But it might be slightly lower. So for me, the main risk to focus on is an escalation and pretty much the beginning of a war. And what then happens is that I think China and Chinese companies will likely survive.

1:24:04And it's so important. China is so important for global trade that I don't see the same isolation happening that we have seen with Russia, for example. Taiwan, on the other hand, does not hold such a strong position. Another, although to be fair, smaller factor is the always-luminum risk and that investors apply a constant discount to TSMC's value because of that. That discount can be higher in times of more uncertainty and it can be lower when the overall relationships between China, the US and Taiwan improve. Since TSMC is priced cheaply but not to the extent that investors currently expect an immediate conflict, investing at current prices implies that one does not expect major further turbulences going forward i would be fine with taking a two percent position to buy a phenomenal business and keep a close eye on the situation but i know that you're more critical of the china situation than i am and even i have relatively strong concerns here so what do you think i'm not going to bury the lead up front, I'll say, I think I can get around to doing that.

1:25:16But then just for the sake of just putting everything out there, I will maybe say some of my concerns. And obviously, we've talked a lot about the China stuff. And to some extent, people who know me might say it's somewhat intellectually inconsistent for me to endorse Taiwan and investment in Taiwan here, because at various points said that I tried to avoid Chinese companies and not necessarily because I think that they're going to, you know, there's imminently going to be a war between the U S and China, but there just seems like there's a lower threshold there where, you know, you could have some kind of fallout in relations where I could imagine a scenario where the U S and China aren't maybe at war, but there's some sort of sanctions being levied where even if these Chinese companies are undervalued, at least from an American's perspective, you could still be penalized.

1:26:10And there could be some sort of sanctions that make it difficult, if not illegal for you to own, you know, maybe not not even all Chinese companies, but specific tech companies, for example, like if you're investing in Alibaba and Tencent. And so that for me is really the concern of China is not that they say anything bad about these companies or to even bet that there's gonna be a war between the US and China. But just to say that there could be a lower threshold where the US government, as for me, as an American, makes it painful for me to invest in these companies. And with Taiwan, I look at it differently because I don't think the US is going to sanction investing in Taiwan.

1:26:45And so truly, it would have to almost be the worst case scenario. So I see the risk reward as a little bit different and where I see the risk is actually being a little bit greater with Chinese companies where you have the risk of a war, but also you have the lower threshold of just some sort of like targeted sanctions that could impact you as specifically an American investor. What really kind of puzzles me with this investment or gives me pause is to think about the fact that for anyone who's read The Innovator's Dilemma by Clayton Christensen, I think the semiconductor industry is literally the example that they use of the fastest changing industry.

1:27:28And this is something Buffett has always talked about, of when an industry is so fast changing, you know, even the widest moats can disappear on relatively short timeframes. If you're investing in a company on a 50 year time horizon, then, you know, a fast changing industry that like this, like semiconductors, literally the textbook example of perhaps the most dynamic industry in the world where, you know, competitive advantages can be the most short lived. that really gives me some almost more anxiety than the geopolitical situation with tsmc okay i don't know how to underwrite the quality of their moat and then on the other hand as we've kind of seen with adobe it's like you know adobe is this great company that has dominated their niche for over 30 years and so the same is true with tsmc is for as fast as the semiconductor industry is TSMC has continued to dominate.

1:28:25So that's a really long way of saying, I have concerns, not totally sold, but I think I could get on board with a two or 3 % position. And as we continue to learn about the company more over time, continue to get more comfortable with the risks, we may exit the position if we're not sleeping well at night, or we might double down on it. How about we do a deal? We already mentioned that there are quite some experts in our community on this sector. And I still wanted to talk to some of them. What do you think if we still have some time until this goes out, if we still talk to those industry experts, perhaps we're both getting sold a bit more on the moat and why TSMC, even in an industry like semiconductors, can remain on top like they did for decades now.

1:29:12and then we have or decide on our final portfolio position when the newsletter and this episode is coming out. Okay, that sounds good to me, Daniel. Yeah, so we, for context, we typically record the podcast maybe a couple weeks before we end up writing the newsletter. So, well, I guess we'll have the final decision in the newsletter. Make sure you sign up for that to actually see the conclusion here. and it is a free newsletter and we condense a lot of the information that we do in the podcast into it, give it out for free to the world. It's about a 10 minute read once a week and you can find the signup link in the notes below or at theinvestorspodcast.com.

1:29:56We had a lot of technical details today and I hope it has been an interesting pitch nonetheless, but now it's time to wrap it up and to get finally to the easier part of the pitch, which are once again, your hints for the audience about the next episode. Okay, well, we've had a lot of fun digging into TSMC today. But yes, the train keeps moving forward, Dana. No breaks for us. On to the next. So I will be sharing another company that is also sensitive to war and societal tension, but in a very, very different way. And I'll also say that this is not a high quality compounder in the same way. And the last hint is that the company is very cyclical, but the cycles are not necessarily economic, not economic cycles.

1:30:44And I'll give a bonus hint. And if anyone doesn't want the bonus hint, just skip ahead like 15 seconds right now. Okay. And so the bonus hint is that the company has a truly iconic brand name spanning back to the days of the wild west. Okay, so that was the bonus hint. It's probably too much of a giveaway, but please leave a comment if you have any guesses. All right, so now I will leave you with a quote from, of course, Warren Buffett. And he said, quote, time is the friend of the wonderful business and the enemy of the mediocre. TSMC definitely belongs to the first category. Buffett still didn't come around to give it a long time in his portfolio.

1:31:27Let's see how this will turn out for us. Have a great day.

1:31:59Thank you.