The Acquired podcast released a new episode two weeks ago featuring an interview with Taiwan Semiconductor founder Morris Chang, lasting 140 minutes, plus the hosts Ben Gilbert and David Rosenthal's review and analysis of the interview, making for a full three hours of captivating content.

On December 10 last year, after 26 years, 93-year-old Morris Chang published the second volume of his autobiography, covering the period from 1964 to 2018, recounting his life experiences before and after founding Taiwan Semiconductor.

In this interview, he mentioned a few things, all of which are written in the book.

For example, in 2009, he had to return to the role of CEO. At that time, the 40-nanometer node was at a critical juncture, and Taiwan Semiconductor's process progress was slower than expected, causing clients, including NVIDIA, to lose time and money.

In the first four or five weeks after I resumed the role of CEO, I spent nearly half of my time dealing with how to resolve issues with NVIDIA.

Finally, Zhang Zhongmou quickly reached a settlement by compensating over 0.1 billion US dollars, on the condition that he clarified the key issues and calculated a fair price for both parties.

This is a very rational and wise business case, and it also reflects Huang Renxun's successful attributes.

This interview moved along with Zhang Zhongmou's memories, deeply presenting some details of past business events. Including:

1. A letter, the 'first encounter' with Huang Renxun

2. Returning as CEO in 2009

3. Resolving the dispute of over 100 million US dollars with NVIDIA

4. Key decisions on the 28 nanometer node

5. How Wei Zhejia 'unexpectedly' became the candidate for succession

6. The real reason why Apple chose Taiwan Semiconductor.

7. The story of IBM and Qualcomm: Insights into industry changes.

8. How to become a devoted believer in the learning curve.

9. Anticipating the emergence of the fabless model.

This interview was arranged by Ben Gilbert and David Rosenthal with the help of Jensen Huang, specifically flying to Taipei to meet with Morris Chang in his office. After a whirlwind 48-hour trip, they also visited Taiwan Semiconductor's headquarters located in the Hsinchu Science Park, where many of Taiwan Semiconductor's fabs are gathered. It is quite enviable!

A review and analysis of their interview has also been specially translated and attached at the end of the text, as it is equally splendid.

I really like their consistent 'analysis handbook' from their previous blogs, systematically presenting the solid research, investigation, and feelings of the interview process before the interview.

Smart investors share the translation with everyone, hoping that everyone has the same sense of gain.

1. A letter, the 'first encounter' with Jensen Huang.

David: We want to start this interview in a special way—by talking about the person who connected us. Could you share your perspective on your relationship with Jensen Huang and the special connection between Taiwan Semiconductor and NVIDIA?

Zhang Zhongmou: My relationship with Jensen Huang can be traced back to a letter.

I remember it was in 1997 when he wrote a letter to Taiwan Semiconductor. The letter was sent through the post office to Hsinchu and eventually reached my hands.

The letter stated that their company, NVIDIA, was still very small, but they had developed some very promising chips and hoped to find a wafer foundry to cooperate with. They had previously contacted Taiwan Semiconductor's office in San Jose but had received no response.

Therefore, Jensen Huang personally wrote this letter to me, requesting that I contact him because NVIDIA was very eager to collaborate with Taiwan Semiconductor.

Coincidentally, I will be going to the USA next week, and this letter piqued my curiosity. But at the same time, it also made me a bit annoyed. Because I have always instructed our sales team never to neglect potential clients, even if they are a very small company.

At that time, NVIDIA had only been established for four years.

I remember that Zhang Zhongmou they were facing bankruptcy, and I recall their employee count was probably only 50 to 60. By that time, Taiwan Semiconductor's employee scale had already reached several thousand.

I also remember that our revenue exceeded 1 billion USD in 1995. By 1997, in comparison, we had already become a fairly large company.

This is quite remarkable. After all, at that time Taiwan Semiconductor had only been established for 10 years, and it had already achieved 1 billion USD in revenue.

Yes. The second week I went to California and then called Huang Renxun without prior notice. I saw the phone number on the letter he wrote to me.

The call was answered by Huang Renxun himself, and there was a lot of background noise; he seemed to be arguing with his employees about something.

As soon as I introduced myself and said I was Zhang Zhongmou, he immediately shouted for everyone around to be quiet and said, 'Zhang Zhongmou just called me!'

Then, I scheduled a meeting time with him, which should be the next day or shortly after. I went to NVIDIA, and this was our first meeting.

He left a very deep impression on me—his communication skills were outstanding, and his optimistic attitude also impressed me, even though he candidly told me that NVIDIA was experiencing financial difficulties.

He also told me that they hoped to find Taiwan Semiconductor to manufacture this chip, which could not only save the company but also turn NVIDIA into an important client of Taiwan Semiconductor.

This statement was actually very bold. After all, at that time, our annual revenue had already exceeded 1 billion dollars, and to become a major client of Taiwan Semiconductor meant they would have to contribute at least 50 million dollars in revenue to Taiwan Semiconductor each year.

David, was that chip the RIVA 128?

Zhang Zhongmou, I don't remember the specific model, but it was a very successful chip. It shouldn't be called RIVA or something like that; it was a gaming chip. Of course, it did achieve tremendous success.

In fact, Jensen Huang was right.

This chip not only resolved NVIDIA's financial crisis, saving it from bankruptcy, but within two to three years, NVIDIA truly became one of Taiwan Semiconductor's top five clients.

II. Returned to the position of CEO in 2009.

This is how Taiwan Semiconductor and NVIDIA established a very close cooperative relationship—Taiwan Semiconductor is responsible for wafer manufacturing, and NVIDIA is responsible for chip design. To this day, the two continue to maintain this collaboration at a large scale.

However, cooperation is not always smooth sailing, and challenges have been encountered.

I want to discuss the 40-nanometer node in 2009. At that time, the progress of Taiwan Semiconductor's process was slower than expected, leading to losses of time and money for customers like NVIDIA. Could you share with us how this issue occurred and how it was resolved?

Zhang Zhongmou: At that time, I decided to hand over the CEO position to my successor while I retained the position of Chairman. In Taiwan, it is usually the Chairman who is the real decision-maker, even if another person holds the title of CEO.

The issue you just mentioned occurred during a time when I was not the CEO. Clearly, this was a manufacturing problem and also a quality issue. The CEO initially reported this problem to me.

The quality director of Taiwan Semiconductor at the time insisted that Taiwan Semiconductor was not responsible, and based on this claim, the CEO did not provide any compensation to NVIDIA.

However, the 40-nanometer issue was not just a predicament faced by NVIDIA; the entire industry was affected. Of course, NVIDIA was one of the most important customers at the 40-nanometer node at that time.

In the Semiconductors industry, the "yield issue" refers to the low proportion of qualified products that can be manufactured when factories produce high-quality chips.

Zhang Zhongmou said that's pretty much the case. This issue persists, and even when not being the CEO, impatience has started to grow. Simultaneously, besides the 40-nanometer issue, other problems have continued to arise. Therefore, I decided to resume the role of CEO.

In 2009, I resumed the position of CEO at Taiwan Semiconductor.

At that time, I faced multiple urgent issues to resolve, one of which was the dispute with NVIDIA.

I remember in my first few days back, I called all major customers, including Huang Renxun.

Also, Qualcomm was one of the major customers, right?

Oh yes, Qualcomm was too. The list of major customers back then isn't much different from now, with perhaps one exception.

Apple?

Zhang Zhongmou Yes, Apple joined later.

In the call I had with Huang Renxun, he was still very friendly towards me, but his tone was serious. He reminded me that the 40-nanometer issue still exists, which has had a significant impact on NVIDIA.

I told him, "I am aware of this issue, and it is one of my current top priorities. Give me a few weeks, and I will get back to you."

Of course, in addition to the 40-nanometer manufacturing issue and the disputes with NVIDIA, I was also facing other challenges at that time.

First, the price is declining faster than the cost. You do not want to see this situation because it means that the gross margin is constantly decreasing.

This means that you have already committed to customers about the timeline for price reductions, but in reality, the decline in manufacturing costs has not kept pace with this rhythm.

Zhang Zhongmou Yes, that is one of the issues.

Another issue is the direct reason that prompted me to return as CEO—I found that the previous CEO had conducted a round of layoffs, but he did not use the word "layoffs." Instead, he made excuses, saying that these people had poor performance evaluations, with a total of 600 to 700 people being dismissed.

However, Taiwan Semiconductor never does that.

During my leadership, the worst handling method was merely placing employees on a six-month observation period. Usually, after six months, most people return to their original positions, and a small number may be reassigned, but we almost never directly fire employees, even after they have gone through the observation period.

In other words, under your leadership, Taiwan Semiconductor has never conducted layoffs, and the purpose of performance evaluations is to help employees grow, not as a basis for layoffs.

Zhang Zhongmou: That's right.

I even specifically told the management that this should not be done. Especially in 2008, when the global financial crisis broke out, the entire semiconductor industry was affected, and Taiwan Semiconductor's revenue significantly declined...

At that time, I was not the CEO; I was the Chairman. However, I knew that any general manager or CEO without much experience would instinctively react in the same way when faced with such a situation - 'Oh, this is my test, I must save costs as much as possible, so I will lay off employees.'

David: But this is the semiconductor industry! According to Moore's Law, no matter what happens, businesses always need people.

Zhang Zhongmou: I understand. But in fact, people in the semiconductor industry also think the same way, and they will lay off employees as well.

I worked at Texas Instruments for many years and have rich experience, but I was not the CEO at that time, just one of the senior managers.

In the early 1970s, when the company decided to cut jobs, the CEO first discussed it with us senior executives. My colleagues' first reaction was completely the same as that of Taiwan Semiconductor's CEO during the 2008-2009 period - to determine the layoff list based on performance evaluations.

However, in the early 1970s, when the executives at Texas Instruments believed that layoffs should be based on performance evaluations, I was the only one who stood up to oppose it.

I said that this was not a reasonable way, and employees would not respect us if we decided who to let go based on performance ratings.

David, why is that?

Zhang Zhongmou, because it's too subjective.

Performance ratings are scored by each person's direct supervisor. If you want to lay off the 700 'worst performing' employees, then who gave them those low scores?

It was the 700 direct supervisors! It's completely subjective. This approach will not earn the respect of the employees.

Moreover, if you have to rehire these laid-off employees within a year, it indicates that layoffs should not have happened in the first place. The compensation cost of layoffs usually equals half a year's salary, and the training time for a new employee takes at least half a year. If a company needs to rehire within a year, it shows that the layoff decision was completely wrong.

So, when you resumed the role of CEO, how did you handle the issues regarding layoffs and client matters (like with NVIDIA)?

Let's first talk about the issue of layoffs.

I previously mentioned that out of the 600 to 700 laid-off employees, some later protested at my doorstep. At that time, Taiwan Semiconductor had already received notice and learned that hundreds of people would come to my house to demonstrate, so they informed the local police in advance. The police dispatched 50 to 60 officers to maintain order.

Ultimately, more than 100 protesters showed up, and my neighbors were affected when they came and went. But this was just the first protest. A month later, the issue remained unresolved, and since I was not the CEO at that time, the protesters returned.

This time, about 25 protesters decided to stay overnight, sleeping in a park a street away from my house. That night, my wife couldn't sleep at all; she kept checking the situation outside from the window.

The next morning at six o'clock, after my wife got up, she took a bodyguard to the nearby market and bought traditional Chinese breakfast—soy milk, deep-fried dough sticks, and buns, enough for 25 to 30 people. Then she took the breakfast to the park and distributed it to the protesters.

The protesters were very grateful and even decided on the spot that they would not protest at the presidential office that day. They told my wife, 'We won’t go to the presidential office today.'

This matter ultimately prompted me to take on the role of CEO again.

However, there was one more thing. At that time, I had already warned the former CEO before the layoffs. I knew his instinctive reaction was to cut jobs when faced with a crisis. So I said to him: "If you want to lay off employees, you must submit it for discussion by the Board of Directors. I will call for a special board meeting."

In fact, I had already thought it through. If the board meeting were indeed held, I would request the board to veto his layoff proposal. But he bypassed the board because he believed he was not laying off employees but was instead "penalizing underperformers."

As for the former CEO, I did not fire him; instead, I had multiple in-depth discussions with him and told him that he was still a potential successor for me in the future. I allowed him to retain the same level, salary, and bonuses, but his new position was to be responsible for new business.

At that time, we had high hopes for "new business," including CECEP Solar Energy and Led & Optoelectronics.

This matter is really ironic because Taiwan Semiconductor's core business—integrated circuit manufacturing—ultimately became the largest market opportunity globally, while the so-called "new business" did not bring about any changes.

That's right, in the end, Taiwan Semiconductor's core business became the largest market opportunity. But why do you find this ironic?

I have always found it interesting that many companies think: "We should explore new businesses." But in fact, the semiconductor industry itself has grown into a market with an annual revenue of 600 billion dollars, while the market size of solar energy and LED, in contrast, is trivial. Taiwan Semiconductor was already standing in the best market.

Zhang Zhongmou: I know this. I never really believed that CECEP Solar Energy or Led & Optoelectronics would replace the integrated circuit business. I am clear that the future of the integrated circuit industry will be very good. However, in 2009, we also thought that CECEP Solar Energy and Led & Optoelectronics were very promising businesses.

But in the end, neither succeeded.

In fact, the CECEP Solar Energy industry could have developed well, but massive subsidies from the mainland affected this market, and now they almost completely control the entire solar cell industry. Due to subsidies, prices are extremely low, and they remain very low, so the market did not really take off, and Taiwan Semiconductor's solar energy business also failed.

Led & Optoelectronics did not develop either, but for different reasons. The scale of the Led & Optoelectronics market is much smaller than that of CECEP Solar Energy, and patents in this field are mostly controlled by a few companies. These companies that hold the patents are completely unwilling to budge, resulting in difficulties for the market to break through.

So, a few years later, the CEO who was assigned to oversee the new business also realized that this direction was not viable at all, so he chose to resign.

Now he is the CEO of MediaTek, right?

Zhang Zhongmou: Yes, he is now the Vice Chairman and CEO of MediaTek.

Three, to resolve NVIDIA's controversy worth hundreds of millions of dollars.

Returning to 2009, after you reassumed the role of CEO, you invited all the laid-off employees back to the company and reestablished Taiwan Semiconductor's vision and Global Strategy. So, how did you resolve the dispute with NVIDIA?

Zhang Zhongmou: In the first four to five weeks after I resumed as CEO, I spent almost half of my time dealing with how to resolve the issues with NVIDIA. Of course, this isn't just about NVIDIA.

David: Not just NVIDIA, right? Qualcomm and Intel were also affected.

Zhang Zhongmou: Yes, because the 40-nanometer node is an important milestone in the development of the Semiconductors industry. If the issues with 40 nanometers cannot be resolved smoothly, we cannot advance to 28 nanometers, and 28 nanometers is the next critical node.

So, I convened all the sales staff who had direct contact with NVIDIA and gathered everyone involved with this issue. Ultimately, we concluded that this was fundamentally a 'money issue.'

From the manufacturing side, we have been doing our utmost to optimize the process, not just for NVIDIA, but for all of Taiwan Semiconductor. However, NVIDIA was the most severely affected customer, and their losses were the greatest.

So, this turned into a compensation issue.

I spent a long time studying all aspects of the problem, and then calculated a reasonable compensation amount.

I also know that NVIDIA's clients are putting pressure on them, so I compiled all the information and developed a response plan.

About a month after I resumed my role as CEO, I sent an email to Jensen Huang. I said, "I will be in Silicon Valley next week, and I'll arrive at your home at six o'clock that evening. Let's have salad and pizza."

We had eaten salad and pizza together many times before, so this felt like a familiar atmosphere. Jensen Huang immediately replied to the email, asking, "So when do we discuss business?"

David, did he ask who was paying the bill?

Zhang Zhongmou, he didn't ask that question (laughs).

But I had thought of that in advance. So I replied, "We start with salad and pizza at 6:30, and then we will go to your study to discuss formally at 8:00."

On the agreed day, I arrived at his house on time. We followed the plan: salad and pizza at 6:30, and the conversation was very enjoyable.

His wife Lori made the salad herself, and the pizza was ordered from outside, or they may have made it themselves; I don't remember clearly.

It's not surprising that they make pizza themselves.

No matter what, I have had this kind of meal at his house many times before. Then, at exactly 8:00, I looked at my watch and said, "Jensen Huang, let's go to your study to talk business."

You indicated at that time that the compensation amount was about 0.1 billion USD, right?

Yes, it was more than 0.1 billion USD.

I also added, "Our compensation plan is valid for 48 hours. We will not negotiate or continue to argue. If you do not accept within 48 hours, we can only leave it to the arbitration agency to handle."

In fact, before I took over as CEO, Jensen Huang had suggested to the previous CEO to resolve the dispute through arbitration. But the previous CEO didn't even provide a specific figure, he just told NVIDIA, "We will not compensate any amount."

But you definitely don’t want to end up in court with one of your most important clients, do you?

That's right, I absolutely do not want to go the arbitration route. I say this just to make my position clear. Because I spent weeks studying all the details, and the final compensation amount I came up with is fair for both parties.

So, did Jensen Huang accept your proposal?

Yes, he accepted it within 48 hours.

I think this case is very representative.

You had already established a cooperative relationship for many years, with close personal ties, and could even have a home-cooked meal first before discussing serious matters. Ultimately, you reached a settlement quickly with a compensation amount of over 0.1 billion dollars, and since then, cooperation between Taiwan Semiconductor and NVIDIA has created hundreds of billions of dollars in business.

This is a very successful collaboration case, with both parties benefiting from it.

I also really like this case, which is why I wrote it into my autobiography.

IV. Key decisions for the 28-nanometer node.

David After solving the problems of the 40-nanometer node, the next focus of development was the 28-nanometer node. From the history of Taiwan Semiconductor, the 28-nanometer node can be said to be a key point for Taiwan Semiconductor to officially establish its leading position in the industry.

How did the decision to heavily invest in 28 nanometers come about? After all, Taiwan Semiconductor has faced many challenges at the 40-nanometer node.

This reminds me of my experience at Texas Instruments. My highest position there was the head of Global Semiconductor Business.

Of course, Texas Instruments has a wide range of businesses, including defense, materials control, and their initial geophysical exploration business, among others. But the largest business remains semiconductors, and I was the head of the global semiconductor business.

At that time, as the head of global semiconductors, Texas Instruments' R&D budget accounted for 4.8% of the company's total revenue, which I believed was far from enough. I tried to raise it to 5.5%, but each request was rejected.

Returning to Taiwan Semiconductor, I decided to set a fixed percentage for the R&D budget, so there would be no need to debate how much to spend on R&D each year.

In 2008-2009, when I returned as CEO, our R&D budget accounted for about 6% or 7%, but this percentage had to be renegotiated every year, decided by the R&D head and the CEO. I wanted to end this uncertainty so the R&D team could focus on their work.

So, I almost made the decision directly—since we had maintained a 6%-7% budget, let's increase it directly to 8%, regardless of whether an economic recession occurred, we would maintain an 8% R&D budget share.

This decision was the best news for Taiwan Semiconductor's R&D team.

If you were to ask the research and development director at the time, he would definitely tell you that he has always believed this decision was the most important step in the history of research and development at Taiwan Semiconductor.

The R&D team no longer needs to worry about budget cuts; they know that, regardless of the market environment, their funding is stable. This is the core driving force that enables Taiwan Semiconductor to accelerate research and development.

In 2010, Taiwan Semiconductor not only significantly increased its R&D budget but also intensified its capital investment. Over the past decade, Taiwan Semiconductor has invested approximately 2-2.5 billion dollars annually in the construction of its fabs.

However, by 2010, Taiwan Semiconductor's capital expenditure directly doubled to nearly 6 billion dollars. What prompted you to make such a significant investment decision?

Zhang Zhongmou: This was driven by multiple factors. After I decided to fix the R&D budget at 8%, the R&D team began to have bigger ideas. They told me that 28 nm would become a key node, which they referred to as the 'sweet spot.'

The sweet spot?

Zhang Zhongmou: Do you play tennis?

I have played, but my skill level is average.

Zhang Zhongmou: Very well, I was also like this 40 years ago (laughs).

But I don't play anymore. However, I know that when you hit the ball with the sweet spot of the racket, the feel is particularly good. And 28 nanometers is that sweet spot.

At that time, I asked them: 'Why?' They gave me many technical reasons.

So, I chose to trust them. They now have the resources to fully push 28 nanometers.

In addition, we had already established a fairly mature market forecasting system at that time. I set up the Business Development Department, which is similar to a traditional marketing department but more oriented towards strategic marketing. At the same time, we also had a strong sales team.

In my view, the sales team operates at the tactical level, directly interacting with customers, while the marketing department is responsible for the strategic level, analyzing market trends.

From the analysis of various aspects such as marketing, business development, and technological research and development, 28 nanometers is that crucial turning point.

Thus, I quoted a line from Shakespeare in my autobiography: 'There is a tide in the affairs of men, which, taken at the flood, leads on to fortune. (Translated: In a person's life, there will be tidal opportunities; if you can ride the wave at its highest point, you can achieve success).'

(Smart Investment: This quote is from Shakespeare's play Julius Caesar.)

At that time, I thought that 28nm was the wave we had to catch. Of course, later there was 7nm, which was another "sweet spot." At that time, I recalled Shakespeare's quote again.

David "Ride the wave and break through the waves."

Zhang Zhongmou "Yes, ride the wave."

However, when the R&D budget was set at 8%, the Board of Directors had no objections. But when I suddenly tripled the capital expenditure (about three times), the Board raised many questions.

The system for our Board of Directors meetings at that time (and still is basically the same now) was as follows: Most directors were from overseas, mainly from the USA and United Kingdom.

Two weeks before the Board meeting, we would send the meeting agenda to them by email. Then, the night before the meeting, I would invite the independent directors to dinner.

The content of the conversation during this dinner would not be recorded, allowing the directors to ask me questions in an informal setting. In fact, more than three-quarters of the directors on Taiwan Semiconductor's Board are independent directors.

But regarding this plan, they began questioning it even before having time to discuss it at dinner.

This decision essentially puts the company's huge cash reserves at stake in the 28-nanometer process node.

Zhang Zhongmou responded affirmatively.

They directly called the company's legal officer (who is also the Board Secretary). The legal officer at that time was an American. They told him: "We want to talk to the chairman; we do not like this decision."

So about a week before the Board of Directors meeting, I communicated with them over the phone. I shared the same information with them that I just shared with you, including market forecasts, the R&D team's input, and the analysis from the new business development department, etc.

But frankly, they did not believe it. It is very hard to convince people regarding such matters.

In the end, I could only say: "I understand your concerns, but as CEO, I am the person ultimately responsible for the company's operations. You need to let me push this decision forward."

They ultimately accepted it.

So what was the result? During this period of 28 nanometers, what created such tremendous market demand?

The result is very good.

This exactly coincides with the rise of the Smart Phone era, right?

Yes.

When the Business development team was studying the 28 nanometer node, did you foresee the massive growth of the Smart Phone market? Did you realize that this market would bring such a huge opportunity for Taiwan Semiconductor?

No, I didn't fully see that at the time.

But the head of the Business development department (that’s another interesting story) might have known or at least I hoped he had a clearer perspective than I did at the time. Of course, this was not our only source of market forecasting; I had some other advisors.

Five, how Wei Zhejia 'unexpectedly' became a candidate for succession.

David, this brings us to the topic of Apple.

Ben, can you share with us how Taiwan Semiconductor connected with Apple?

Zhang Zhongmou, before discussing Apple, I would like to first talk about how C.C. (Wei Zhejia) became the head of the Business Development Department.

Ben, he is now the Chairman and CEO.

Zhang Zhongmou, yes. From 2005 to 2009, during the tenure of Rick Tsai as CEO, he divided Taiwan Semiconductor's operations into two main business groups - Advanced Process Group and Mainstream Process Group.

At that time, Wei Zhejia was in charge of the Mainstream Process Group, which honestly was somewhat a secondary one. Liu Deyin was in charge of the Advanced Process Group.

Each business group had a small business development department, consisting of about 30-40 people.

Ben, however, after you returned as CEO, you did not approve of this split model?

Zhang Zhongmou: In fact, I never thought that such a split was a good idea.

As early as 1996, when the president of Taiwan Semiconductor was Don Brooks, an American (at that time we didn't have the title of CEO).

Me: Don Brooks?

Zhang Zhongmou: Yes, that's him. He hoped to split Taiwan Semiconductor into multiple business units at that time.

In my opinion, he was mainly a bit tired of managing the company at that time. He initially only planned to stay at Taiwan Semiconductor for a year, but ultimately he stayed in Taiwan for 6-7 years. By the end, he was indeed a bit exhausted.

His idea was to reference the approach of Texas Instruments.

Texas Instruments once had different business units, such as the germanium transistor department, silicon transistor department, bipolar integrated circuit department, MOS integrated circuit department, etc.

Me: So that’s the traditional 'divisional structure' rather than a 'functional structure'?

Zhang Zhongmou: Yes, completely correct.

But I do not believe that Taiwan Semiconductor's foundry business is suitable for this divisional structure. Because our customers are almost the same group of people. If you want to adopt a divisional structure, how would you divide the company?

Don Bruker’s idea at that time was to divide it by factory, which was absurd. Customer Orders would switch between different wafer fabs, yet he wanted to divide the departments by factory.

Moreover, Taiwan Semiconductor now has 21-22 wafer fabs.

Zhang Zhongmou: Yes, at that time Taiwan Semiconductor only had 3-4 wafer fabs. But in any case, he did not agree with my point of view. He kept arguing with me. So I said, 'Why don't we invite a consulting company to take a look?'

And: McKinsey?

Zhang Zhongmou: Yes. We invited McKinsey to evaluate this issue. They studied for two months, spending several million dollars, and finally drew the same conclusion as I initially did - a functional structure is more suitable for Taiwan Semiconductor.

Then, Don Bruker was still unwilling to accept it, and he said, 'Then tell me, which large company adopts a functional structure?'

McKinsey immediately replied: "Boeing."

But in fact, Boeing also distinguishes between commercial and government business.

Zhang Zhongmou said, "That may be the case, but they won't split by aircraft model. If we divide Taiwan Semiconductor's organizational structure by factory, it would be like Boeing breaking down the 707, 747, and 757 into different departments, which obviously wouldn't work."

Don Bruker's attempt occurred in 1996, and by 2005, Tsai Lixing decided to try the same approach again. But this time, I did not stop him.

My management principle is that as Chairman, I sometimes have to let the CEO make mistakes and learn from them. Of course, this is on the condition that the mistake does not lead to the company going bankrupt, otherwise I must intervene.

So, when you returned as CEO, Taiwan Semiconductor was still divided into two business groups?

Yes, one was the advanced technology group and the other was the mainstream technology group.

At that time, I decided to merge these two business groups and establish a real marketing department.

However, I did not use the term "Marketing," but instead chose "Business Development," as it translates better in Chinese.

In 2009, when I decided to merge these two business units, the Advanced Process Group had about 0.01 million employees, and the Mainstream Process Group had slightly fewer, around 7,000-8,000 people...

This Mainstream Process Group has not been detailed yet in our previous discussions about this concept.

In simple terms, it is responsible for using older wafer fabs and high nanometer node processes, serving customers who do not require the most advanced technology. For example, it may manufacture chips for Auto Parts or provide foundry services for cameras' CMOS Sensors.

The task of this department is to maintain high production capacity utilization rates for wafer fabs of the previous generation process nodes, thereby extending the useful life of old facilities.

Zhang Zhongmou: Exactly.

But in reality, many customers need both mainstream processes and advanced processes simultaneously. For example, Qualcomm, I am sure they not only use the most advanced processes but also mainstream processes. Even Apple is the same.

David: If we closely analyze all the chips inside the iPhone, such as the A16 Pro processor, its manufacturing process is at the forefront of the industry. But in reality, there are many other chips inside the iPhone, not all of which are manufactured using the most advanced processes.

Zhang Zhongmou is correct.

So, you decided to merge the two business development teams into one, which has about 80 people, right?

Yes. At that time, we had Liu Deyin responsible for advanced processes and Wei Zhejia responsible for mainstream processes. But the question arose, who would be in charge of the entire merged team?

Because when the two business units are merged, only one leader is actually needed. Taiwan Semiconductor has many outstanding talent in operation management.

Here, operation refers to transforming the developed technology into large-scale mass production. However, in terms of market business development and marketing, the situation is different—Liu Deyin and Wei Zhejia both lack real marketing experience.

So, this was my biggest concern at that time.

We not only need a post-merger operation manager, but more importantly, we need a head of integrated market business development.

I first offered the position of market business development to Liu Deyin. He was then the head of the advanced process group, which means he managed Taiwan Semiconductor's core business.

I explained to him: "I know you haven't had much market experience in the past. If you're willing to take on this new position, it will be a great opportunity for you to learn this skill and expand your capabilities."

But his response was: "Oh my, I currently manage 0.01 million employees, and you want me to lead a department of only sixty or seventy people?"

Then, our conversation came to an end.

Your goal is to hope he becomes a well-rounded manager, and you want him to have the capacity to lead the entire company after completing this job, right?

That's right, I explained this to him at that time.

And although this department only has sixty or seventy people, they undertake the key tasks for Taiwan Semiconductor's future growth—responsible for finding the next core business of Taiwan Semiconductor.

That's right. In fact, I remembered the history of Henry Kissinger serving as National Security Advisor to the USA.

At that time, Kissinger had several hundred subordinates, while the Secretary of State (I can't even remember who held that position) managed thousands of diplomats globally.

But whose power is greater? There is no doubt, it's Kissinger.

This certainly is not the person whose name you can no longer remember (laughs).

Before this, David was actually responsible for Taiwan Semiconductor's market development and business expansion, right? It was you who uncovered potential high-quality clients like NVIDIA and Broadcom and made Taiwan Semiconductor their long-term partner.

Zhang Zhongmou: That's right.

At that time, you were almost always on the plane, maintaining relationships with the top 15 clients while also looking for the next 15 major clients.

Zhang Zhongmou: Correct. Except for those four years from 2005 to 2009 when I was not the CEO, essentially, most of my time was spent flying around the globe to establish client relationships for Taiwan Semiconductor.

This is one of my pleasures; I really enjoy this.

Zhang Zhongmou: Later, I offered the position of Director of Market Business Development to Wei Zhejia, who gladly accepted it, even one might say he accepted the position quite happily.

David, and now he is the Chairman and CEO of Taiwan Semiconductor.

Zhang Zhongmou, yes.

An unusual dinner - the first encounter with Apple.

David, when this happened, you had just returned home from the Board of Directors meeting, right?

Zhang Zhongmou, that's right. The Board of Directors meeting had just ended, it was already past six o'clock when I got home. It was in Taipei, and our Board of Directors meetings were held in Taipei at that time. In fact, it was in my conference room, you have seen it, right?

David, right across this hallway.

Zhang Zhongmou, yes. That day it was probably around 6:30 when I got home. My wife had anticipated that I would arrive home by 6:30, so she was specifically waiting for me at the door. This is quite uncommon. Obviously, she had something important to tell me.

She told me, 'Terry Gou called this afternoon and said he wants to come over for dinner.'

In order to help the audience better understand, can the introduction of Terry Gou be provided?

Tsu-Chung Chang Terry Gou is a distant relative of my wife Sophie. Strictly speaking, he is my wife's cousin; they have the same grandparents, so they are considered cousins (I think that is the case).

David Many Western audiences may not be familiar with this name; Terry Gou is the founder and CEO of Foxconn.

Tsu-Chung Chang That's right.

In order to introduce Foxconn, the parent company is called Hon Hai Precision Industry, which is the official company name for Foxconn in Taiwan.

Tsu-Chung Chang Yes, Hon Hai. I momentarily couldn’t recall that name (laughs).

Hon Hai is one of Apple's most important suppliers and is also a very large company. In fact, Terry Gou was considered one of the richest people in Taiwan at that time.

Sophie is gentle and lovely, but she doesn't know much about my business. I don't think she truly realizes the significance of Terry Gou coming over for dinner, let alone the fact that he brought along a vice president from Apple.

She may not be fully aware of the commercial significance of this matter and might not be very interested in this topic.

In fact, you and the business development team have spent months researching how to win Apple's business, right? Because the iPhone seems to be starting to succeed.

Zhang Zhongmou: Yes, but to say 'strategic layout' might be a bit exaggerated. I have just been thinking about this issue, but at the same time, I'm also clear that we can't do anything.

Apple is an extremely closed company. If you take the initiative to promote your services, they will just ask you to leave. That's always been Apple's style; they will only come to you when they are ready.

This was the case back then, and it still is now.

Sophie knew I wouldn't be home until after 6, so she informed Terry Gou, and he scheduled the visit for 8.

8 o'clock is a bit late for me, but I thought, never mind, I'll wait. After they arrived, I didn’t ask much.

Sophie just told me there was a vice president from Apple. But I thought to myself, Terry Gou wouldn't casually bring an ordinary vice president to meet me, so this must be a key figure; otherwise, there would be no reason to do so.

As a result, the 'vice president' who came here is actually Apple's Chief Operating Officer Jeff Williams.

Jeff is a very straightforward person and hardly engages in small talk.

David, this dinner is not like the previous meeting with Jensen Huang that involved 'pizza and salad', right?

Zhang Zhongmou: Yes, but it is also not a formal business meeting. Sophie just asked the chef at home to prepare a few more dishes.

Our chef only knows how to cook Chinese food and cannot make Western dishes. Foxconn's Terry Gou obviously grew up eating Chinese food, and I estimate this Apple executive would also enjoy Chinese cuisine.

However, what we eat during this meal is simply not important. The quantity and quality of food do not matter because Jeff Williams almost jumped straight into the topic as soon as he sat down.

Ben, how did Jeff, as an Apple executive, propose the intention to cooperate?

Zhang Zhongmou: He said it very directly: 'We hope you can manufacture chips for Apple.'

It's that simple. Very straightforward.

That night, I was mostly listening. Jeff spoke for about 80% of the time, while I only spoke 20% of the time.

If you exclude the small talk between Sophie and Terry Gou (which wasn’t much), it was basically Jeff leading the conversation.

Apple proposed economic terms during the first meeting, right?

Terry Gou said no, they didn’t propose any specific numbers. However, Jeff did mention, "We can help you maintain a gross margin of 40%."

At that time, I didn’t respond to him, nor did I make any statement.

Because TSMC's gross margin at that time was already 45%, and I had been pushing to raise it to 50%. This has also been a goal I’ve been striving for since returning as CEO.

But even by the time I retired, I was unable to fully achieve the 50% goal.

Of course, some events occurred later, such as the COVID-19 pandemic and the establishment of Taiwan Semiconductor's leading technology position. These factors allowed our gross margin to exceed 50%. However, we were still a bit short before my retirement.

In 2010, during the 28-nanometer node period, Taiwan Semiconductor had already entered the forefront of the industry, but the competition was very intense at that time.

It wasn't until the 7-nanometer period that Taiwan Semiconductor truly widened the gap with its competitors, right?

Zhang Zhongmou, saying this overlooks Intel. During the 28-nanometer period, Taiwan Semiconductor was undoubtedly the leader in the wafer foundry industry, even ahead of some semiconductor companies like Texas Instruments. But Intel was still a player that could not be ignored.

David, at that time Apple was also considering using Intel's chips, right?

Zhang Zhongmou, no, Apple was not actively considering Intel. That was a later development, and I believe we will have time to discuss this topic later.

David, then let's continue the story. In November 2010, you had your first conversation with Jeff Williams.

Zhang Zhongmou, yes. At that dinner, he told me that Taiwan Semiconductor's gross margin could be maintained at 40%.

At that time, I was thinking, "Hey, we are already at 45%." But I also knew that he might be trying to sound generous because Apple has always been an extremely strong negotiating party.

However, I did not respond to that question at the time because I believed that dinner was not the time to discuss pricing; there were many other more important things to talk about.

At that time, Taiwan Semiconductor had just entered the mass production stage of 28 nanometers. Although it was only preliminary mass production, basically all our research and development resources were focused on 28 nanometers. So, I told Jeff, “28 nanometers.”

His response really surprised me; he said, “We want 20 nanometers.”

This is a "half step" process, right?

Yes, but this "half step" is actually a "detour." The original research and development trajectory was to directly advance from 28 nanometers to 16 nanometers, but Apple now suddenly demands 20 nanometers.

At that time, my thought was, "This will force us to invest a lot of energy in 20 nanometers." Of course, the accumulation of R&D for 20 nanometers would ultimately help with the research and development of 16 nanometers, but it is still a detour.

If we had directly advanced from 28 nanometers to 16 nanometers, the time required would have been shorter.

However, the problem was that at that time, Taiwan Semiconductor did not have enough R&D resources to simultaneously advance two process nodes. Later, our resources became strong enough to simultaneously advance multiple process nodes.

Taiwan Semiconductor decided to only take half of Apple's orders.

At that time, Taiwan Semiconductor had just invested 6 billion dollars in capital expenditure on the 28-nanometer node and had not taken Apple into account at all. This meant that to meet Apple's demand, Taiwan Semiconductor would have to invest several billion dollars additionally.

This raised a question - "Is it worth investing so much in the 20-nanometer process to win Apple?"

Zhang Zhongmou said the situation was not that severe. When we decided to significantly increase the capital expenditure for the 28-nanometer process, Apple was not included in our calculations; Apple's emergence was an unexpected surprise for Taiwan Semiconductor, but it was not part of the original 28-nanometer plan.

But the question still remains whether Taiwan Semiconductor is willing to "take a detour" to do 20-nanometer, which may require an investment of up to 10 billion dollars in the coming years.

Zhang Zhongmou confirmed that this is why we established a relationship with Goldman Sachs.

Early in my management of Taiwan Semiconductor, I was already preparing for the future. I knew that one day we would need the advice of top investment banks. Therefore, we established a good relationship with Goldman Sachs early on.

In fact, I used to be a member of the Board of Directors of Goldman Sachs, did you know that?

David: Yes, I know.

Ben: I know too.

Zhang Zhongmou: The ADR (American Depository Receipt) of Taiwan Semiconductor is underwritten by Goldman Sachs, which also helped us establish a long-term relationship with them.

David: So that's the stock of Taiwan Semiconductor listed in New York.

Zhang Zhongmou: Yes, the ADR trades on the New York market, it's an independent market. In fact, currently the price of the Taiwan Semiconductor ADR has a 20% premium over the stock price in Taiwan.

Ben: Really?

David: Wow.

Zhang Zhongmou, however, to convert Taiwan Semiconductor's Taiwan stock into ADR, approval from the Board of Directors is required.

Otherwise, there will be an arbitrage opportunity.

Zhang Zhongmou, yes, we do not want that to happen. The Board of Directors must approve any conversion of common stock into ADR, and the Board usually does not approve easily.

David, so you were already prepared for such a moment.

Zhang Zhongmou, that's right. When the demand from Apple appeared, we had already decided to increase capital expenditures, but Apple's orders required us to invest more funds, so we needed to determine how to finance.

At that time we had several options: cut dividends (the dividend was not high at the time, but still there); issue new shares (which could be done in the Taiwan market or US ADR); issue CSI Enterprise bond Index; accept only part of Apple's orders, instead of all.

Ben, in the end, you chose the fourth option, right?

Zhang Zhongmou, we first made financial plans and decided not to cut dividends, nor to issue new shares, but to directly borrow (issue CSI Enterprise bond Index). This decision was also made at the suggestion of Goldman Sachs.

So, how many Apple Orders should be accepted?

I finally decided to only accept half of the Apple Orders.

Is this situation common? It seems that for the customers, they certainly hope Taiwan Semiconductor can fully satisfy the orders, but they do not need to bear any financial risk.

Zhang Zhongmou, of course, this situation happens frequently.

In the 90s, Taiwan Semiconductor was almost always in a state of tight capacity. Many customers would say, 'We need a lot of capacity,' but when it actually came time to pay, they might not really place that many orders.

So, I established a policy: customers need to pay a deposit, and if they ultimately do not fulfill the orders, we will confiscate the deposit.

When I first proposed this policy to Taiwan Semiconductor’s sales team, they were all stunned. I told them, 'You can directly tell the customers that if they ultimately do not execute the orders, we will confiscate the deposit.'

The sales team had never heard of such a strategy; they were shocked at first, but then they were overjoyed because they could finally take the initiative in negotiations.

Of course, we never actually confiscated any customer's deposit, but this policy ensures customer commitment.

Going back to Apple in 2011, we told them we could only meet half of the demand.

The first reaction of Apple's procurement team was: "Are you crazy?"

Wei Zhejia was responsible for first informing Apple's procurement team, and they were shocked, thinking how could Taiwan Semiconductor only accept half the orders? But Wei Zhejia did not respond directly, he just brought the message back to me.

Then, I personally went to Apple and met directly with Jeff Williams.

I told him: "We need to issue company bonds." I specifically used the word "cautious," explaining that after careful financial planning, we decided to only accept half of your demand.

After hearing this, Jeff was silent for a while, and his only suggestion was - 'You can cancel the dividend, the shareholders will understand.'

I immediately replied to him: "No."

In fact, I have researched this issue. This is also the value of Goldman Sachs providing top-notch consulting. About one third of our shareholders rely heavily on dividends, and if we really follow Jeff's advice, the stock price of Taiwan Semiconductor would plummet.

This would trigger large-scale sell-offs.

Zhang Zhongmou Yes. However, after listening to Jeff, he did not insist further. He just nodded, and then this topic ended.

In the end, we reached an agreement with Apple to only accept half of their orders. But even so, Taiwan Semiconductor still needs to borrow tens of billions of dollars to support this investment.

Six, the real reason Apple chose Taiwan Semiconductor.

After the massive investment in 28 nanometers exhausted the company's cash reserves, you decided to take on significant debt to build new fabs to meet Apple's demands. One could say this is a "betting the whole company" gamble, right?

Zhang Zhongmou Yes, I know people would say that—"betting the whole company."

But I never thought I would lose.

You sound like Jensen Huang.

That's exactly what Jensen Huang said back then!

I believe that the discussions between Apple's financial team and Taiwan Semiconductor were basically completed, until in February 2011, Jeff Williams suddenly called me.

This call was very short; he just said one sentence: 'We need to pause the discussion for two months.'

Because the top management of Intel had directly reached out to Tim Cook and asked him to consider Intel as a supplier.

At that time, Intel was still the main chip supplier for all Mac computers, right?

This was not the key issue, because in February 2011, Jeff Williams was talking about the iPhone, not the Mac.

There was still a close collaboration between Apple and Intel.

Zhang Zhongmou said, yes, I do not know how close their relationship is... but in any case, they definitely have cooperation.

However, I am not particularly worried because by 2011, Intel was no longer the company that commanded respect.

This is very interesting.

Zhang Zhongmou said, in the 1990s, Intel was definitely the dominant player in the Semiconductors Industry.

At that time, if you heard that Intel was your competitor, you might feel scared.

Of course, I am exaggerating (laughs).

David, Moore's Law, Intel's brand influence...

Zhang Zhongmou said, yes, in the late 1990s and early 2000s, Intel was indeed the undisputed industry giant.

This is also one of the reasons you decided to let Taiwan Semiconductor focus on the pure foundry model, right? You believe that Taiwan Semiconductor should not become an IDM (Integrated Device Manufacturer), but should compete in different arenas, as directly challenging Intel is unrealistic.

I never said 'will never catch up with Intel.'

Well said (laughs). Let's see how things are on this day in 2025!

Of course, I still accepted Jeff Williams' request to pause the discussions for two months. But I am still not too worried because I know the criteria Apple uses for supplier selection:

Technical strength: At that time, we believed Taiwan Semiconductor's technology was almost on par with Intel's, and possibly even stronger.

Manufacturing capability: We believed that Taiwan Semiconductor's manufacturing capacity had already surpassed Intel's.

Customer trust: We firmly believe that customers' trust in Taiwan Semiconductor is higher than that of Intel's customers in Intel.

So, I wasn't too worried that Intel might take away Apple's Orders.

But after a month, around mid-March, a decision was made to proactively contact Apple to check on the progress.

I emailed Jeff Williams to request a meeting.

"I happen to be going to Silicon Valley and would like to visit you. Is it convenient on a certain date?"

Jeff's reply was interesting; he said, "Come on over, but I won't be there. I'll have Tim Cook meet you."

Generally speaking, it is the superior who arranges for subordinates to meet clients, but this time it was the opposite—subordinates arranged for the CEO to host you!

Yes, such a "reverse arrangement" is quite rare in my career.

Typically, corporate executives would say, "My team will meet with you," but this time it was, "My boss will meet you."

Yes.

So, I went to Apple's headquarters in Cupertino, where Tim Cook personally received me and took me to the cafeteria for lunch. We each selected our meals and returned to his office with our trays.

He said a very brief but satisfying thing to me: "There is nothing to worry about. Intel doesn't know how to do foundry work at all."

That was the answer I most wanted to hear.

How do you interpret what Tim Cook said?

My understanding is that our technology and manufacturing capabilities have reached standards. But subconsciously, I think Tim Cook was referring to the third point: customer trust.

Prior to this, Intel had always dominated customers outside of Apple. All PC manufacturers in Taiwan were basically Intel's customers. But the problem is—no manufacturer likes Intel, not one.

Intel has always behaved as if they are "the only choice in the market" because they are indeed the only choice in the microprocessor field.

This is Intel's performance in the microprocessor business, but in the foundry business, the situation is different. Taiwan Semiconductor's core principle is to never compete with customers.

Even if Intel is earnestly engaging in foundry business, they will still face a structural conflict, as they are also designing chips themselves, which means they will directly compete with the chip design teams of clients like Apple and NVIDIA.

Zhang Zhongmou, but I really don't think Tim Cook was referring to this at that time.

I think what he meant was that clients will always have various demands, some of which are even 'crazy' or 'unreasonable', but we have learned to respond politely to each request.

However, Intel has never done this. In Taiwan, I know many Intel clients who wish for another supplier.

They neither trust Intel nor like Intel.

In the end, Apple's 20-nanometer collaboration went smoothly, right? But did it come at a cost? Did Taiwan Semiconductor pay a price in other areas for spending billions of dollars to develop the 20-nanometer process?

Zhang Zhongmou, yes, this story is certainly not over yet.

The pricing issue is not easy. Jeff Williams came personally to Taiwan Semiconductor to discuss pricing with me. Of course, we also did adequate preparations beforehand, conducting in-depth analysis on costs, pricing ranges, and so on.

Jeff directly quoted a price and explained that Apple needs to keep chip costs within a target range. In the end, we reached an agreement.

Once the pricing was settled, I said to Jeff, 'Let's go out for dinner.'

We went to a Michelin three-star restaurant in Taipei. Jeff joked with a smile, 'If you don't agree to this price, we might have to go to McDonald's.'

In fact, I never really thought about going to McDonald's (laughs).

Value distribution between Taiwan Semiconductor and clients.

David: What are the main considerations in pricing? I guess the yield rate of the wafers is also an important factor affecting pricing?

Zhang Zhongmou: Of course, the most core factor of pricing is cost.

Secondly, it's whether the client is willing to accept your pricing.

Currently, Taiwan Semiconductor's gross margin has reached 55%-57%, which is far higher than the levels during your tenure. On the other hand, many customers have gross margins as high as 70%-80%.

Taiwan Semiconductor has created a substantial amount of value, and chip design companies have also created significant value. So, how is it decided who gets how much value?

This is not something I can decide. Now, it is Wei Zhejia who is responsible for this issue; he has both the responsibility and the enjoyment of it.

In principle, one always tries to find a balance, but this balance is different for each CEO.

Every CEO will say they focus on long-term interests, but in reality, not everyone truly does. This is actually a very personalized issue.

Earlier mentioned cost, what does it refer to specifically?

This cost is actually very serious.

At that time, in 2011-2012, Taiwan Semiconductor's R&D resources were not strong enough to develop multiple process nodes simultaneously. Now we can do it, but we couldn't back then.

Accepting orders for 20 nanometers means we delayed the R&D for 16 nanometers. Then, Samsung took the lead by launching the 16 nanometer process.

This is because they directly skipped 20 nanometers, right?

Zhang Zhongmou: Yes, they don't have the burden of 20 nanometers, so they can directly progress to 16 nanometers. When I learned that Apple's first 16 nanometer orders were taken by Samsung, I was shocked.

That was indeed a huge shock!

We invested so much money in 20 nanometers, and although we only accepted half of their demand, we still invested billions of dollars. We were originally counting on getting Apple's order for 16 nanometers, but now Apple has chosen Samsung.

Can you imagine how I feel?

This certainly sounds terrible. It feels like being 'tricked'.

Zhang Zhongmou: I wouldn't put it that way, but I was truly shocked.

I immediately emailed Jeff Williams, saying that we have invested huge amounts of money to support 20-nanometer production and we have been counting on Apple to continue collaborating with us on 16 nanometers. But now we've heard that you have given the first orders for 16 nanometers to Samsung...

Jeff replied almost immediately, saying: "Don't worry, I will come to Hsinchu next week to explain to you in person."

This made me feel a little relieved, but of course, I was still not completely at ease.

Jeff did come the following week, and he explained to us: "As long as your 16-nanometer process is ready, we will give you all the orders."

This made me completely assured because this was our original goal.

It turned out that Jeff was a man of his word. When Taiwan Semiconductor's 16-nanometer process was ready, Apple still awarded most of the 16-nanometer chip orders to us.

David can imagine the shock you felt when you heard that Apple chose Samsung. But this once again proves the value of Taiwan Semiconductor's pure foundry model.

Samsung is Apple's biggest competitor, yet Apple is still willing to choose Samsung as a foundry in certain cases.

Zhang Zhongmou said, "As a foundry, we can see the trends in the Industry."

In my autobiography, I wrote, "Sitting in Hsinchu, as a pure foundry company, we can actually see market changes earlier than others."

Let me tell a story about IBM and Qualcomm.

David, please share!

Seven, the story of IBM and Qualcomm: how to insight into the changes in the Industry.

Zhang Zhongmou, we have always believed that Qualcomm is an important potential customer of Taiwan Semiconductor.

We are very eager to secure Qualcomm's Order because we know that Qualcomm is a company with strong technical capabilities.

Ben, in which year did this happen?

Zhang Zhongmou, this goes back to the early 1990s, until around 1997.

David, at that time, Qualcomm was one of the first wave of fabless companies, right?

Zhang Zhongmou, yes, when Qualcomm's founder Irwin Jacobs founded the company, it was even earlier than Taiwan Semiconductor.

From the early 1990s to around 1997, we were constantly striving to get Qualcomm to become our client.

But Qualcomm kept giving their major orders to IBM.

I often met with Qualcomm's Vice President of Operations, who was in charge of procurement. He was always polite to us, but gave us very little business.

Until 1997 or 1998, Qualcomm's attitude suddenly changed, and he started to say to me: "We are now willing to use Taiwan Semiconductor."

But he didn't directly tell me who their previous supplier was. However, I had long known through other channels—that their supplier was IBM.

Soon, Qualcomm's order volume for us rapidly increased, which immediately made me realize that IBM's Semiconductor division was in trouble.

IBM's wafer foundry business is almost entirely dependent on orders from Qualcomm and a few small fabless companies. Now, Qualcomm's business is flowing to Taiwan Semiconductor, which is a huge blow for IBM.

When IBM realized they were losing customers, their next move didn't surprise me.

They quickly reached out, inviting Taiwan Semiconductor to collaborate on the development of the next generation process—0.13 microns (130 nanometers) technology.

David, what year is this?

Zhang Zhongmou, it's 1999.

But I had already anticipated that IBM would take such action, so we unhesitatingly rejected their proposal at that time. In fact, even if I had not foreseen it, we would never agree to co-develop the technology with IBM.

IBM still sees itself as "the industry leader" and has always considered itself the "senior partner" in any collaboration.

Their cooperation model is that the partner must send engineers to work at IBM.

However, if we do this, we will lose our ability for independent research and development, and in the future, we will have to rely on IBM, which is absolutely unacceptable.

In addition, engineers from Taiwan Semiconductor who work at IBM will have to adapt to IBM's internal corporate culture, which will create huge communication barriers.

Therefore, we decisively rejected IBM.

IBM was very angry at that time; they still thought that Taiwan Semiconductor was just a "backward Taiwanese company" and that they were the powerful IBM.

So, they immediately turned to United Microelectronics Corporation (UMC).

UMC accepted IBM's cooperation proposal, but a few years later, they deeply regretted it.

This is exactly what I mean: as a foundry company sitting in Hsinchu, we can foresee changes in market trends in advance.

VIII. How to Become a Loyal Believer in the Learning Curve

David This may be a good time to return to the concept of the "Learning Curve". In the Semiconductors Industry, mastering one's core technologies and processes, ensuring self-control capability is crucial.

Can you explain the theory of the "Learning Curve" to us?

Zhang Zhongmou The theory of the learning curve itself is very simple, but if one only understands its superficial meaning, it is too shallow.

The most basic learning curve theory is this: as a company produces more of a certain product, its unit cost will decrease. This originated in the manufacturing industry of Autos and Refrigerators. With the accumulation of production experience, the company's production efficiency will continue to improve, and costs will decrease.

This is also why it is called the "Experience Curve". As you accumulate more experience, your efficiency will naturally improve.

This is a simple explanation of the learning curve.

But if you only stop at this simple explanation, thinking that the learning curve is merely "more experience yields higher efficiency", then you haven't really learned anything.

Back to the learning curve itself.

Bruce Henderson, who is now regarded as the father of strategic management...

David Bruce Henderson founded the Boston Consulting Group (BCG).

Zhang Zhongmou: Yes. There is a branch in business economics called competitive strategy, or something similar.

Michael Porter was once an important figure in this field, and he wrote three or four classic books on competitive strategy, each with 700 pages, all of which I bought.

The original memorandum of Michael Porter's 'Competitive Strategy' is only 20 pages long and remains one of the best business writings to this day. He was once a Director of Taiwan Semiconductor, right?

Zhang Zhongmou: Yes. I also wrote about his story in my autobiography. However, due to time constraints, we may not go into detail about it today.

However, we will talk about the story of Bruce Henderson.

Bruce Henderson is now considered the 'father of competitive strategy.'

He came to Texas Instruments around 1970, or more precisely, he first called CEO Mark Shepherd.

In the call, Bruce Henderson introduced Shepherd to BCG's 'Experience Curve Theory' and stated, 'This theory will greatly promote the development of the Semiconductors industry.'

At that time, Texas Instruments was the largest semiconductor company in the world, and Shepherd was very interested in this theory, so he invited Henderson to speak at Texas Instruments.

Bruce Henderson brought Bill Bain, whom you may have heard of, to Dallas (the headquarters of Texas Instruments).

He gave a very wonderful speech because Bruce Henderson himself was a person with great speaking talent.

Bill Bain stood by, looking like Henderson's assistant.

Mark Shepherd was very satisfied after listening and decided to have Texas Instruments adopt BCG's learning curve theory.

As a result, Henderson sent Bill Bain to Texas Instruments, while Shepherd assigned me as the point of contact for Texas Instruments.

Bill Bain worked at Texas Instruments three days a week, and he was arranged in a small office very close to my office because he needed to frequently request data from me, such as: the cost structure of Texas Instruments, product pricing, and the production situation of various integrated circuits and transistors.

Whenever he arrived at interesting or valuable conclusions, he would discuss them with me immediately.

We worked together very happily for about two years, even longer. Bill Bain flew from Boston to Dallas every Monday and stayed until Wednesday or Thursday evening before returning. Each time he returned to Boston, he reported the week's progress to Bruce Henderson.

This experience allowed me to deeply understand and master the learning curve theory, which I have continued to apply to this day.

I found that this theory is an extremely powerful thinking tool.

The impact of this learning curve on the semiconductor industry is very profound, isn't it?

In the semiconductor industry, companies need to quickly get through the low production phase, and it's best to completely avoid staying in a low production state for long periods.

The returns in the Industry will eventually flow to the companies with the highest production capacity, as they have the lowest costs, can attract all the Orders, and can reinvest profits into the next generation of processes.

This creates a positive cycle: having the lowest cost → gaining the most customers → further reducing costs → forming a long-term competitive advantage.

Therefore, from the perspective of the learning curve and globalization, ultimately there will only be one winner in the Semiconductor manufacturing Industry.

Zhang Zhongmou: Yes, the logic of the learning curve is just like that.

But then one day, Bill Bain came to me and said, "What I am telling you now, apart from BCG, you are the first to know."

He told me he was preparing to leave BCG and start his own business.

Ben: Why?

Zhang Zhongmou: I asked him the same question. He answered that it was imperative.

(In 1973, Bill Bain left BCG to establish Bain & Company. He aimed to create a new type of consulting firm different from BCG, emphasizing long-term partnerships with clients rather than just providing one-time consulting advice. Bain focuses on a few high-end clients and offers in-depth services.

In 1984, Bill Bain further founded Bain Capital. This is a Private Equity firm, with initial funding primarily coming from partners of Bain & Company.

Nine, anticipating the emergence of the fabless model.

Taiwan Semiconductor is almost the only company with a Market Cap of one trillion dollars outside the USA's West Coast. It occupies a crucial position in the global technology industry, creating an incredible success story.

Do you think such success seemed unlikely at the beginning?

Zhang Zhongmou, unlikely? That's your opinion, isn't it? Is Taiwan Semiconductor's success a 'low-probability event'?

I mean, you were already 56 years old when you founded Taiwan Semiconductor, which in itself is remarkable...

Zhang Zhongmou, I won't argue with you on this point, I'm just asking out of curiosity.

I have never felt that Taiwan Semiconductor's success was improbable. Of course, its scale and significance have indeed exceeded my initial expectations, but not by an order of magnitude.

But wasn't the original plan to stop after building Fab 2 (the second wafer plant)?

No. That was just an idea in the initial stages, but we never really intended to stop there.

We just talked about the learning curve, and if one truly understands it, it's impossible to stop after building only two plants.

If I knew nothing about the learning curve, I might say, 'Well, maybe building two plants is enough.'

But I am a firm believer in the learning curve, so we will never stop at Fab 2.

The reason I say that Taiwan Semiconductor's success is unlikely is that there were too many factors at the time that made it look like a bad business.

At the early stage of Taiwan Semiconductor's establishment, the 'fabless model' had not yet truly formed. Most of your early customers were IDMs, like Intel and those companies. However, they only outsourced products they did not want to produce to Taiwan Semiconductor, meaning that initially, Taiwan Semiconductor could only take on 'the orders that others did not want.'

Did you foresee the rise of the "fabless model" at that time? Or was it just a lucky opportunity?

Zhang Zhongmou: No, I anticipated the emergence of the fabless model.

In fact, just two months ago, I had dinner with the first person to pioneer the fabless model—Gordie Campbell.

He came to see me in the last few months before I left General Instrument. However, at the time, he did not know I was going to leave General Instrument, and in fact, I hadn't decided to leave myself.

(Note: Zhang Zhongmou briefly served as president of General Instrument after leaving Texas Instruments in 1983, but this experience was very short, lasting no more than a year.)

He came to see me to seek funding; he wanted General Instrument to invest $50 million to start a new semiconductor company.

I asked him, "Do you have a business plan?"

He said, "No, all the plans are in my head."

I said, "No, I need at least a business plan, as I have to report to the Board of Directors of General Instruments."

He agreed and said, "Okay, I will send it to you in three weeks." But three weeks passed, and the business plan did not arrive.

I was still interested because I knew he had a good reputation for starting companies. So, I proactively called him to ask why the plan had not been sent.

As a result, he told me on the phone, "Morris (the English name of Zhang Zhongmou), sorry, I didn't send it to you because I don't need you anymore."

I asked, "Why?"

He said, "I no longer need 50 million dollars; now I only need 5 million dollars, and I can easily raise that money."

I asked, "Why do you suddenly only need 5 million dollars?"

He said, "Because I decided not to build the wafer factory."

You see, this is when I first realized that the "Fabless model" was about to arrive.

Later, another person came to find General Instruments; he had founded a company called Atmel. Atmel is also a Fabless company, and they wanted General Instruments to help them with foundry services.

At that time, General Instruments' foundry had a lot of idle capacity, so I told the manager of the semiconductor division at General Instruments: "Okay, collaborate with them."

David Tang-Valentine has a classic saying; when he founded Sequoia Capital, he once said: "I have an advantage, which is that I know what the future looks like."

It sounds like you also foresaw the future of the semiconductor industry.

Zhang Zhongmou At least I had some insights. At that time, Atmel and General Instruments still had management differences.

Atmel wanted to operate the foundry in their own way, while the semiconductor manager at General Instruments insisted on managing the factory in their own way.

But after all, the foundry is an asset of General Instruments, and they certainly wanted to operate it in their own way.

This has allowed me to see early on the difficulties and opportunities that a factory must face:

The difficulty is that you must meet the different needs of multiple customers, and each customer wants the factory to operate according to their way.

The advantage is that you can serve a large number of customers instead of relying on just one.

Dr. Ben, we really appreciate you accepting this interview!

Zhang Zhongmou: Very good, very good. This interview was very pleasant. Although it has been a long time since I have done such a long interview, I enjoyed the process.

Thank you for being willing to share so many stories with us.

10. Postscript: Review of the Interview

(Note: The following content is the discussion between Ben and David after the interview)

We want to review this interview, discussing our analysis, conclusions, and which aspects have lingered in our minds during the past few days after crossing the ocean back home.

David completely agrees. For me, this interview is very important because it feels like recording history. I didn’t want to interrupt Morris Chang by inserting a discussion on business models; in such times, the best practice is to let him speak freely, and then we can analyze it later.

Yes. Fortunately, we have developed an analytical framework, so let's proceed according to this manual.

• Key to success: do not compete with customers.

Firstly, one thing that I can never forget is TSMC's 'pure foundry model', which appears to be a genius decision in retrospect. During our visit to TSMC's Innovation Museum, we also saw Chang's original business plan and his initial slides.

David: This was the business plan he presented to the Taiwanese government back then.

Yes, he first presented it to the government and then to the investors. This slide exists in two different versions, but one key point is—that TSMC will focus on being a pure foundry.

At that time, my feeling was that this decision was more based on realistic feasibility than on making TSMC the world's most important and valuable semiconductor company.

David: Yes. At that time, Taiwan Semiconductor did not have the capability to design chips and products, and Taiwan also lacked such an industrial foundation. They were completely unable to compete with customers, so they could only do foundry work.

Ben: Of course, Morris Chang must have thought, 'Can we compete with Intel?' But ultimately he gave up on that idea. I think it is because manufacturing is the area they excel in.

And the historical coincidence is that the pure foundry model ultimately became the optimal model—if measured by market capitalization, it outperforms all other semiconductor models, including Intel's IDM model.

David: That's right. And it turns out that this path allowed Taiwan Semiconductor to technically surpass all other IDM and quasi-IDM manufacturers, becoming the only leading company.

Ben: This reminds me of Steve Jobs' famous quote—'You can only connect the dots looking backward.' But looking forward is very difficult.

David: This may be the main reason for Taiwan Semiconductor's great success.

Ben: Because they do not compete with their customers.

David: They are the only pure foundry in the industry that does not compete with customers in leading processes. They have no end products, do not design chips, and only provide manufacturing services for customers. They do not participate in other parts of the value chain; they only do foundry work.

• The rise of ARM is crucial.

If you think about it carefully, you will find that the industrial landscape of this world has allowed a company that does not design chips, does not do architecture, does not develop EDA (Electronic Design Automation) tools (like Cadence or Synopsys), and does not produce equipment (like ASML), to become a trillion-dollar enterprise. What is the key behind this?

One factor, I believe, is the crucial rise of ARM. However, we did not delve into this issue during the interview.

If we look back in history, assuming Intel and the x86 architecture continued to maintain their dominance, there would not have been such a restructuring opportunity in the global semiconductor industry chain, and the division of labor in the industry would not have changed.

However, the emergence of the ARM architecture completely changed the rules of the game.

As we discussed with ARM CEO Rene Haas in the ACQ2 interview—ARM has achieved dominance from Smart Phones to PCs, Servers, and AI processors. Its popularity necessitated a dedicated foundry for producing ARM chips.

If the world were still dominated by x86, you would not see so many fabless companies emerging, and naturally, there would not be Taiwan Semiconductor serving these companies.

A very simple but important example is Apple. Without ARM, Apple would not have been able to create its own Apple Silicon (self-developed chips), and they would likely still rely on Intel to manufacture the core chips in the iPhone. After all, during the Mac era, all of Apple's Macs used Intel's x86 chips.

Of course, this assumes that Intel can manage mobile chips. However, perhaps the historical burden of x86 itself makes it unable to operate efficiently on mobile devices, so this is not an issue of Intel's capabilities, but rather a limitation of the architecture.

I agree. But in any case, without ARM, this industry chain would not have evolved the way it has today.

Our core point is that the division of labor in the global semiconductor industry cannot be separated from the collaborative development of ARM and Taiwan Semiconductor.

In fact, you checked before that many key companies were established almost at the same time, right?

• The ecosystem at the physical layer.

Yes, the late 1980s was a golden era – TSMC, ARM, Synopsys, Cadence, and ASML were all established within just a few years.

This brings us to the Hsinchu Science Park. We discussed in that episode about TSMC that even if you wanted to 'airlift' TSMC from Taiwan to somewhere else, you cannot replicate the entire industrial ecology.

Previously, we thought so theoretically, but actually driving through Hsinchu Science Park, that immersive feeling is completely different.

David: The ecosystem here is like a government-led super Silicon Valley, and there has indeed been a similar history, like the early Silicon Valley we discussed in the episode on Lockheed Martin.

Ben: Yes, it's the Silicon Valley from the Lockheed era.

David: Exactly. Today's Hsinchu Science Park is like Silicon Valley of that era. Everything is here, not only Taiwan Semiconductor but also their partners and clients. As we drove by, we saw buildings of Cadence, Synopsys, and ARM...

Ben: And Qualcomm.

David: MediaTek's headquarters is right here, just across the street. What shocked me the most is that there are actually two universities here.

Ben: Yes, right in the Science Park.

David: These universities produce a large number of doctoral graduates every year, and they directly integrate into this ecosystem after graduation. It's like building two universities at NVIDIA's headquarters specifically to supply talent for it.

Ben: Yes. People used to say that Taiwan Semiconductor's supply chain is 'highly integrated,' but it was only when I saw the geographical environment here that I truly understood this.

David completely agrees. The close connectivity of the entire ecosystem is simply hard to replicate.

The most impressive point for me is that we often hear people talk about the high integration of this ecosystem. For example, Synopsys must work closely with Taiwan Semiconductor to anticipate the next generation of process nodes, allowing chip companies using Synopsys design tools to smoothly adopt Taiwan Semiconductor's processes for manufacturing.

When you are in the midst of it, you immediately realize, 'Oh, I get it.' Because they are right across the street from each other, this extremely close communication is a physical reality.

Moreover, our flight experience gave us the same feeling—like a group of fabless chip design companies are pilgrimaging to Taiwan to meet with partners in this ecosystem.

David, my flight felt like a tech shuttle for the semiconductor industry, just like those commuter buses from San Francisco to Silicon Valley. The backpacks I saw on the plane were from Google, Amazon, ARM, and Marvell.

This raises the question of why Taiwan Semiconductor would build a fab in Arizona, or outside of Taiwan. Because this is not their most advanced process, nor is it a base for mass production, and it does not fully utilize the highly integrated geographical ecosystem established in Taiwan.

I remember Taiwan has three science parks; we visited the initial one, but there is a larger one in the south, which I believe is the Tainan Science Park.

But it is clear that Taiwan Semiconductor builds factories in other countries mainly based on the needs of customers and governments, rather than purely on the best commercial choices.

David, it is impossible to replicate such a complete ecosystem that exists on a physical level anywhere else.

Ben, indeed, recreating an ecosystem like the science park of Taiwan Semiconductor in Taiwan may take decades.

David, this is interesting; when we discussed in the car, we said this might be the most successful government-funded industry program in the world, no doubt about it.

Ben, at least in terms of promoting innovation and targeting specific industry development, it is the most successful.

David, I completely agree. This reminds me of the land grant university program in the USA. At that time, the US government consciously promoted the development of a specific industry, and what the Taiwanese government did was like a precise shot, deciding to promote the development of the semiconductor industry in this specific area, and in the end, it really succeeded.

Ben, ultimately, Taiwan Semiconductor became one of the ten most valuable companies in the world and is one of only two trillion-dollar companies whose headquarters are not on the US West Coast. I believe this program was indeed successful.

David, yes, it was successful.

Ben, and the scale is astonishing. We passed by a construction site where it looked like the building was only a quarter complete, and this is where they will produce 2-nanometer processes in the future, likely for the next generation iPhone.

Although no one has stated it outright, there is no doubt that a large number of NVIDIA GPUs and iPhone chips will be produced there.

The scale of this factory is enormous. The first phase is already in operation, and I believe that is only a quarter of the entire building, with three more phases of construction yet to be completed. I've heard they are planning to start small-scale trial production of 2-nanometer processes in the second half of this year and gradually expand production capacity.

David, just as you said, the physical scale of these fabs is indeed shocking. Standing there, it felt like looking at the Egyptian pyramids; the scale is enormous. Just one phase of the factory is equivalent to several football fields, and these buildings are truly massive.

• The wafer fab strategy of 'saving the country through curves.'

Well, let's return to what I've been thinking about since interviewing Dr. Morris Chang. I actually regret saying in the interview that TSMC's initial business model was not an ideal choice.

After all, in the early stages, their business model was essentially to take on excess capacity from Intel and other IDM manufacturers to produce the least critical, least advanced, and least attractive chips.

But that was indeed the case. At that time, Morris Chang believed that the fabless model would eventually become mainstream, but in the first five years, TSMC's main customers were almost all IDM integrated manufacturers, whose focus was simply: 'How cheaply can you provide manufacturing services?' This was not strategically significant for them; it was merely part of their additional income.

David, this is a significant difference between Intel and TSMC in their wafer fab strategies.

Intel constantly upgrades existing wafer fabs, converting them to the most advanced process technology. On one hand, this is a good thing as it maximizes asset value and ensures the production of the highest value products.

On the other hand, they also lose the ability to produce older process chips. However, the market demand for these older technology chips will not disappear immediately.

This demand will indeed decrease, but the process is slow.

David: Yes, it will decrease slowly. For instance, in many technology systems and products, even in the automotive industry, specific chips produced 10, 20, or even 30 years ago are still in use, and these chips rely on older process technologies for manufacturing.

When these chips fail and need replacement, the market still requires these identical chips. This was precisely TSMC's original business.

This is a fundamental conceptual difference. I remember that TSMC's first fab (Fab 1) was actually the asset of the Industrial Technology Research Institute (ITRI), and Morris Chang was the head of that organization at the time.

The second and third fabs of TSMC truly belonged to TSMC itself, and they are still running, having been in production since the late 1980s.

In addition to the replacement demand for outdated chips, many application scenarios still rely on older process technology, such as technologies above 40 nanometers, and even at the 1-micron level. There is still market demand for these lower precision wafer etching technologies.

For example, CMOS Sensors do not require a 2-nanometer process, but they still need to be manufactured using chip etching technology. The Sony sensors we are currently using are a great example. This demand has kept the foundries built by Taiwan Semiconductor’s early investments valuable.

More importantly, from an accounting perspective, these foundries have long been fully depreciated, so they are almost operating at 'zero cost'.

David Yes, all capital expenditures have already been amortized. Of course, some maintenance capital expenditures are still required, but the initial construction costs have been fully depreciated. Therefore, the products produced in Taiwan Semiconductor's older plants now have almost extremely high profits.

Ben This does not mean that Intel's strategy is better or worse, only that both have taken completely different paths.

Intel's approach is to close old factories, maintain a more streamlined production line, and focus on the latest and greatest technologies; while Taiwan Semiconductor chooses to take two steps forward, developing cutting-edge technology while maintaining old process capacity.

David Absolutely correct.

• Learning Curve Successful Practitioners

Ben However, one thing that confuses me is that Zhang Zhongmou clearly stated in an interview, 'No, I knew long ago that the foundry model would become mainstream.' This is quite interesting.

In some past interviews, he sometimes said, "The rise of the fabless model has a bit of luck involved." But from his statements in the interviews, it is clear that he had foreseen it long ago.

In the early years of Taiwan Semiconductor, their growth was actually not fast because they were waiting for fabless customers to emerge.

But essentially, he was making a long-term bet, willing to operate a low-margin, less attractive business just to build capabilities, establish capacity, and lay the foundation...

David and the construction of the foundry.

He wanted to be ready for the arrival of the fabless revolution. I believe the timing he predicted was within 12 months of when it actually happened.

But what is most surprising is that, especially when reviewing his experiences, you will find that in the early customer stories, in the fifth, sixth, and seventh years, the main portion of the business still did not come from fabless companies but from the least important orders handed to them by integrated device manufacturers.

David This essentially touches on the core of the learning curve pricing we discussed with Morris Chang.

We only briefly mentioned this topic in the interview, but it may now be worth delving deeper into what exactly the learning curve is.

The core idea of the learning curve was jointly developed by the Boston Consulting Group, Bill Bain, and Morris Chang.

By the way, this is really crazy! The founders of BCG and Bain & Company were actually the ones who elaborated and officially proposed this concept together with Morris Chang, or at least they named and promoted its development together at Texas Instruments.

David is absolutely right. The core insight of the learning curve is that the ultimate goal you are competing for is to become the player with the largest shipment volume in the industry. If you see this goal as a fact, then in an industry with extremely high fixed costs and evident economies of scale, you can dilute costs by maximizing your customer base.

The question is, how to rapidly maximize market share in a fiercely competitive early stage?

The answer is to accelerate the achievement of the industry's final pricing level.

This is why companies proactively lower prices, even selling products at a loss in the early stages. You can adopt lower pricing at certain points in the early stage of generation, even to the extent of selling at a loss, because the ultimate goal is to eliminate competitors, become the dominant player in the industry, and capture the most customers. Once you gather enough market demand, you can gain the advantage of economies of scale and ultimately achieve pricing power.

The key is to reach this position as quickly as possible—this is the essence of the game.

This is all derived backward, and it actually involves a significant amount of market scale forecasting.

Assuming a certain process node (such as 40 nanometers) has reached maturity, what is the expected market demand? What will be the total volume of chip orders for this node? If the goal is to offer the lowest price to customers, maximum production scale must be achieved.

So, the question becomes: how can production be ramped up to mass production as quickly as possible?

Everyone can intuitively understand the concept of economies of scale, but its impact on the entire business is profound, including pricing strategy, strategic financing, when to take on debt, when not to take on debt, and when to introduce more shareholders, among other key decisions.

David, to successfully achieve all of this, extremely precise coordination is needed, akin to a superb conductor's performance. It is almost similar to Costco's business model, requiring accurate calculations to operate smoothly.

That's right, just like Apple's case. We are about to secure a giant-level customer, but we must weigh the options—according to the learning curve principle, we should take on all of Apple's orders as much as possible to reduce costs quickly and deeply learn from the curve.

But the problem is, the risks of doing so are extremely high. If the company's own capacity cannot digest such large-scale production, any market fluctuations could pose an existential crisis for the company. So, is this bet worthwhile?

David, your market forecasting ability must be extremely precise, not only predicting your customers' (like Apple's) needs but also forecasting the demand from their end consumers.

In other words, you must accurately predict how many consumers will purchase the next generation iPhone to ensure the normal operation of the business.

In the case of NVIDIA, there is a need to predict the future market size of the AI industry. For a manufacturing company, this demand forecasting is nearly insane - they must not only focus on their own business but also accurately determine the future size of the market where their customers operate.

They truly need a crystal ball that can foresee the future because the entire industry's development relies on the accuracy of these forecasts.

David, because if your forecast has an error of 5% to 10%, it could directly destroy the profits of the entire process node and thus impact the company's.free cash flow.Ultimately leading to an inability to continue investing in the competition of the next technological cycle.

However, if you perform excellently in these areas, make accurate predictions, and execute flawlessly, once you truly grasp the core logic of the learning curve, then the rise of Taiwan Semiconductor is no longer an unexpected story, but a certainty. Of course, the company that accepts all orders and has the lowest cost will ultimately prevail.

David, of course, the entire industry will ultimately move towards a pattern dominated by one player, which is almost inevitable.

Currently, the cost to build a new wafer fab is about 20 billion dollars? In the future, this number could become 40 billion dollars, 80 billion dollars, or even 100 billion dollars. So, how many companies will have such capital capability left to spend 100 billion dollars to build a wafer fab? This market inherently has near-monopoly characteristics.

David, and this is just a part of capital expenditure, as we discussed with Dr. Zhang Zhongmou. In addition to the construction costs of the foundries, R&D expenditure is also key to determining the outcome.

To develop more advanced process nodes so that this huge capital expenditure can be effective, a massive influx of funds is equally required.

• Integrated circuits have become the Infrastructure of the modern world.

If one observes Taiwan Semiconductor's capital expenditure and Net income year by year, a very surprising phenomenon can be detected: their growth trends are nearly synchronous. In other words, they almost invest all their earnings into capital expenditure, certainly not all, but if you present this data in a bar chart, you will find that their growth curves are highly similar.

Moreover, David, this does not include R&D expenditure. If Taiwan Semiconductor observes the investment scale of other foundries and wants to compare how much it can invest, they can always invest more because their shipment volume is the largest.

In addition, they have also invested a substantial amount of money in R&D for advanced manufacturing technologies. That is why they are able to develop CoWoS—this proprietary technology for AI Chip packaging. Once you master the proprietary packaging technology, it becomes more challenging for clients to search for a second supplier or opt for a dual manufacturing model.

For mobile chips, they also have another set of packaging technologies different from CoWoS. Looking at the development trends of the entire Industry, leaders will only run faster unless major strategic errors or significant execution problems occur, otherwise the Industry landscape will become increasingly concentrated.

David fully agrees. I think the last key strategy point here is—Moore's Law is unbeatable.

Looking back at history, from Dr. Zhang Zhongmou's early career at Texas Instruments to his contemporaneous experiences with integrated circuit inventors like Jack Kilby and Bob Noyce — since the invention of the integrated circuit, the compound growth of the entire industry has become the dominant driving force.

Everything is just a continuation of Moore's Law. The world's demand for computing power is growing exponentially every 18 to 24 months, and that is the most important thing.

Of course, the technical definition of Moore's Law is no longer applicable, but from a core perspective, the world needs about twice the computing power every two years. This trend has continued for 50 to 60 years without any signs of slowing down, and therefore...

There are certainly no signs of slowing down, except for the repeated breakthroughs of physical limits.

David, I am talking about the demand side, and there are no signs of slowing down.

Of course, the growth in demand far exceeds twofold.

The essence of Moore's Law has never been about demand, but rather the speed of innovation in design and manufacturing. Now, this is more difficult than ever. To continue the continuity of Moore's Law, we have to incorporate more technologies, such as packaging technology, software optimization, proprietary interconnections, etc., which were not previously within the scope of Moore's Law.

David, my point is that this is essentially a self-reinforcing system. As long as there is still a demand for computing power in the market, as long as the world continues to crave twice the computing power compared to yesterday, the market will provide enough incentives to drive technological breakthroughs on the supply side.

This is why everyone is desperately pushing for its development.

Let's take a look at a set of data. Since Taiwan Semiconductor was established in 1987, the Global semiconductor market size has grown from 26 billion dollars to last year's 527 billion dollars. They have hitched a crazy ride.

David's crazy ride.

This is a super ride in the market. Moreover, as the entire semiconductor industry gradually shifts from a vertically integrated model (like Intel's own design + manufacturing) to a division of labor model (fabless + foundry), Taiwan Semiconductor has managed to establish itself as a foundry giant with a market cap exceeding 1 trillion dollars.

The scale of the data is truly unbelievable.

I've been thinking that they can spend 20 billion dollars to build a foundry, and the chips produced by this foundry are so valuable that the investment of 20 billion dollars can be recouped in just a few years.

I don't know if the exact investment recovery period is 3 years, 5 years, or 7 years, but what is certain is that they clearly know this is a foolproof investment.

Ultimately, it all points to the same fact - Silicon has become incredibly precious, and integrated circuits have become the infrastructure of the modern world.

Editor/lambor