Latest / Business Strategy Talks with Fexingo: Competitive Advantage, Market Positioning, and Growth Planning / How TSMC Turns Manufacturing Precision into an Unbreachable Moat
Transcript
- Lucas: If you follow chip stocks at all, you've seen the headlines about AI demand, export controls, the CHIPS Act money flowing into new fabs in Arizona and Ohio. But there is one company that sits at the absolute center of all of it, and its competitive advantage is so deep that even governments spending tens of billions have struggled to replicate it. Luna: TSMC. The Taiwan Semiconductor Manufacturing Company. It's probably the most strategically important company you can't buy a product from directly. Lucas: Right. Consumers never see their name on a box. But every iPhone, every NVIDIA GPU, every AMD processor — they all come out of TSMC's fabs. And the question I want to drill into today is: how do they defend that position? Because on paper, this should be a commoditizable business. Chip manufacturing is a service — you design the chip, TSMC prints it. So why hasn't someone else come along and undercut them? Luna: Before we get into that, I want to anchor with one number that always sticks with me: TSMC's share of the global foundry market is roughly 58 percent. Number two, Samsung, is around 13 percent. That's not just a lead — that's a chokehold. Lucas: Exactly. And that gap has been widening, not shrinking. So the moat here isn't just about having better machines. It's about the accumulated knowledge of how to use those machines at a scale and precision no one else can match. Luna: Let's talk about that precision. TSMC is currently manufacturing chips on a 3-nanometer node. For context, a nanometer is a billionth of a meter. The transistors they're printing are so small that the wavelength of the light used to etch them — extreme ultraviolet lithography — is barely larger than the features themselves. Lucas: Right, and getting that process to work reliably at high volume is an absolute nightmare. There's a metric in the industry called 'yield' — the percentage of chips on a wafer that actually work. When Intel tried to jump to its 10-nanometer node a few years ago, their yield stayed below 50 percent for a long time. TSMC, on the other hand, reached over 90 percent yield on its 3-nanometer node within nine months of starting production. Luna: And yield matters because it directly impacts cost. If you're only getting 50 good chips out of 100 on a wafer, your cost per chip doubles. TSMC's ability to ramp yield quickly means they can offer better prices than competitors while still making great margins. Lucas: That's the financial moat. But where does that capability come from? It's not just buying the best ASML lithography machines — anyone with enough capital can buy those. The moat is in the process recipe: the thousands of individual steps — deposition, etching, cleaning, doping — that have to be tuned to perfection. TSMC has been refining those recipes since 1987. Luna: So it's a classic learning-curve advantage. More cumulative production equals more data, more tweaks, more understanding of how the process behaves. And TSMC has produced more wafers than any other foundry by a huge margin. Lucas: That's exactly it. And this ties into something that people often miss about the semiconductor industry: capital intensity alone isn't a moat. Intel has spent tens of billions on fabs. Samsung has spent tens of billions. But they still can't match TSMC's yields because you can't buy experience — you have to earn it over years and years. Luna: Right. And if this kind of deep-dive into competitive advantage is something you find useful — maybe it shifts how you think about your own business, or gives you a framework for evaluating a company — these episodes stay ad-free because of listener support. If you want to help keep that going, it's buy me a coffee dot com slash fexingo. Really low-key, but it makes a difference. Lucas: Yeah, absolutely. We love doing these, and knowing people get value out of them — that's the fuel. Okay, back to TSMC. Lucas: So there's another dimension to this moat that I think is even harder to copy: TSMC's business model of being a pure-play foundry. They don't design their own chips. That means they never compete with their customers. Luna: Which is a huge contrast to Intel and Samsung. Intel designs and sells its own CPUs. Samsung designs and sells its own Exynos chips and memory. So when a company like Apple or Qualcomm goes to a foundry, they have to trust that their proprietary designs won't leak to a competitor. Lucas: Exactly. Trust is a massive part of this. A chip design can represent billions of dollars in R&D. You're not going to hand that to a foundry that might use it to build a competing product. TSMC has never done that, and that reputation took decades to build. Luna: It also means TSMC can focus all its engineering resources on manufacturing. They don't have to split attention between design and process. Intel, for all its engineering talent, has had to balance its internal design teams with its foundry ambitions. And that has created friction. Lucas: Right. Intel spent years trying to become a foundry for other companies, but the cultural shift has been rocky. Their internal design teams get priority access to new process nodes, which makes external customers feel like second-class citizens. TSMC doesn't have that problem. Every customer gets the same process, the same priority — as long as they pay for capacity. Luna: Which brings up another point: capacity allocation. TSMC's customers often compete for fab space. During the chip shortage of 2021, Apple basically reserved TSMC's entire 5-nanometer output. Smaller chip designers had to wait months. Lucas: That's a great example of how the moat becomes self-reinforcing. Apple, NVIDIA, AMD — they all want the best process technology, so they commit to TSMC years in advance. Those commitments give TSMC the revenue visibility to invest even more in the next node. And each new node widens the gap. Luna: Let's talk about the next node. TSMC is already working on 2-nanometer, expected to enter production in 2025. They're also investing in advanced packaging, like their CoWoS technology, which stacks chips vertically to improve performance. That's another layer of moat — packaging is becoming as important as the transistor itself. Lucas: Absolutely. And the cost of building a 3-nanometer fab is now over $20 billion. A leading-edge fab complex can cost $30 to $40 billion. That's a barrier to entry that gets harder with every generation. Very few companies can even afford to play in that league. Luna: And yet governments are trying. The US CHIPS Act allocated over $50 billion to boost domestic semiconductor manufacturing. Intel is building fabs in Arizona and Ohio. But even with that money, Intel is years behind TSMC. The moat isn't just about capital — it's about the organizational learning I mentioned earlier. Lucas: Right. You can't write a check and catch up on thirty-seven years of process refinement. TSMC's learning curve is embedded in the way their engineers think, the way they schedule maintenance, the way they handle contamination. It's tacit knowledge, not just explicit knowledge. Luna: There's also a geographic concentration risk that works as a moat in a strange way. TSMC's most advanced fabs are all in Taiwan, which creates a national security dimension. The US and other countries are desperate to diversify supply, but they also need TSMC's technology for their own defense and tech sectors. So they end up subsidizing TSMC's expansion rather than replacing it. Lucas: Exactly. TSMC is building a fab in Arizona, but it's for 5-nanometer — not their latest 3-nanometer. The most advanced nodes stay in Taiwan. And even with subsidies, the Arizona fab has faced construction delays and labor shortages. It's not easy to replicate the ecosystem. Luna: So the moat is really a combination of factors: pure-play business model, trust with customers, enormous capital requirements, and decades of accumulated process knowledge. No single factor is unassailable, but together they create something very hard to break. Lucas: And here's the thing — the moat is actually getting wider. As chips get more complex, the number of companies that can even design them is shrinking. And those that do — Apple, NVIDIA, AMD, Qualcomm — they all need the best manufacturing. So TSMC's customer concentration is also a risk, but it's also a sign of their indispensability. Luna: Which brings us to the interesting question for the future: can TSMC maintain this lead? Or will geopolitical pressure and the push for chip sovereignty eventually erode their advantage? Lucas: I think the most likely scenario is that TSMC remains the leader for at least the next decade. The next few nodes — 2-nanometer, 1.4-nanometer — are already in development and the barriers only get higher. Intel could become a credible number two, but catching up to TSMC's yields and cost structure is a monumental task. Luna: So if you're an investor, the lesson might be that moats aren't just about brand or network effects. Sometimes the deepest moat is in manufacturing — the ability to make something better, cheaper, and more reliably than anyone else. Lucas: Exactly. And that's a lesson that applies beyond semiconductors. In almost any industry, if you can achieve a 90 percent yield on a complex process while your competitors are at 50 percent, you win. It's just that for most of us, the process isn't measured in nanometers.