TSMC创始人 Morris Chang

Welcome to the spring 2025 season of Acquired, the podcast about great companies and the stories and playbooks behind them. I’m Ben Gilbert.

I’m David Rosenthal, and we are your hosts. Today, we have something very special to share with you. After becoming obsessed with semiconductors from our TSMC episode four years ago, Ben and I wound our way through the rest of the industry, studying fabulous companies like NVIDIA and Qualcomm, architecture companies like Arm, and chip-design software companies like Synopsys.

As we were thinking about what’s next in the world of chips, we threw the Hail Mary. We asked friend of the show Jensen Huang if he would ask Dr. Morris Chang, the 93-year-old founder of TSMC, if he would be open to an interview with us.

It is kind of insane and super cool that Jensen made time to help us with this. It’s not like he doesn’t have a lot of other things going on.

Listeners, it happened. Today’s episode is a conversation that we recorded in Taipei last week at Dr. Chang’s office. We flew to Taiwan for a 48-hour whirlwind, where we spent some time at TSMC’s headquarters in Hsinchu Science Park, where many of TSMC’s fabs are located.

It was super cool to see. So conveniently, Dr. Chang just published Volume 2 of his autobiography a couple of months ago, after a 26-year hiatus from Volume 1. Inconveniently, it is written in traditional Chinese and has not been published in the Western world.

We managed to get our hands on an unpublished translation of the book to prepare. What you are about to hear focuses on a few crucial stories from TSMC’s history that Dr. Chang shares in his memoir about Apple, NVIDIA, and the birth of the fabless industry.

And a big thank-you to Karina Bao, whom we were lucky to connect with after we set this up and who has been translating Morris’s memoirs with funding from Tyler Cowen and Emergent Ventures. Right now, the memoirs are not published in English, and we will let you know if and when that happens.

We thought that, as a fun way to start things off, we would actually talk about the man who introduced us. Could you tell us a little bit, in your own words, about your relationship with Jensen and TSMC’s special relationship with NVIDIA?

My relationship with Jensen started with a letter that he sent to me, I think in 1997. The letter was sent to the post office, and I received it in San Jose. The letter said that NVIDIA, the company that Jensen was the CEO of, was a small company, but they had developed some really promising chips. They were looking for a foundry and had approached TSMC’s San Jose office, but they got no answer from the San Jose office. Would I please contact Jensen, because NVIDIA really wanted to do business with TSMC?

I was going to the United States the next week anyway, so the letter raised my curiosity and also irritated me a little bit. I had always told our salespeople that we should never be negligent in talking to future customers, even if the customer seemed to be a very small one.

At this point, NVIDIA was four years old. They were facing bankruptcy, I think. They had maybe 50 or 60 employees, so TSMC, which at that time already had a few thousand employees, was relatively speaking a pretty big company. We had exceeded, I remember, US$1 billion in revenue in 1995, and this was 1997.

You were yourself only a 10-year-old company doing over $1 billion in revenue.

Right. The following week, I went to California and called Jensen back without advance notice. I looked up the telephone number—I think it was on the stationery that he sent me the letter on. Jensen himself picked up the phone, and there was a lot of background noise. He was arguing about something with his people.

As soon as I introduced myself and said, “This is Morris Chang,” he immediately shouted to those people who were making noise, “It’s quiet! Morris Chang is calling me!”

I proceeded to make an appointment to visit NVIDIA the next day. That was our first visit and our first meeting. He immediately impressed me with his articulateness. He also impressed me with his optimism, while he was very frank. He told me that NVIDIA was in financial difficulties, but the chip that he wanted to have made would not only save the company; it would also make NVIDIA a major customer of TSMC.

That was quite a bold statement. We were a billion-dollar company, and to be a major customer of ours, he would have to produce revenue for us of at least $50 million a year.

Was that chip the RIVA 128?

I forgot the number, but it was a very successful chip.

I don’t think it was RIVA anything.

It was a games chip, of course. It was successful. In fact, his prediction came true. Not only did it solve NVIDIA’s financial problems and prevent it from going bankrupt, it also started to make NVIDIA a major customer of TSMC. Within 2 or 3 years, they became one of our 5 biggest customers.

There was a great partnership forged there. TSMC would fab the chips, manufacturing them, and NVIDIA would design them. That is true all the way to today, at immense scale. But it hasn’t always been easy, and it hasn’t always been perfect.

I want to go to this moment in 2009, on the 40-nanometer node, where development was slower than TSMC had hoped and it was costing customers like NVIDIA time and money. Can you share the story of how this came to be and how it was resolved?

I decided to give the CEO job to a potential successor of mine, while I would still retain the chairmanship. In Taiwan, usually the chairman is the top man anyway, even though the CEO is another person.

The problem you just mentioned happened during the period when someone else was the CEO. Apparently, it was a manufacturing problem, and it was also a quality problem. The quality problem was what the CEO first reported to me. He insisted that TSMC was not at fault, based on the arguments of our director of quality, and on that basis he had not offered NVIDIA anything.

As far as the manufacturing problem was concerned, it was a yield problem. Everybody was suffering from it. Of course, NVIDIA at that time was perhaps the biggest customer of that node, the 40-nanometer node.

A yield problem in this context is when you are trying to make a bunch of very high-quality chips, but you just can’t get the percentage that actually work up very high.

Yes. The problem apparently continued. Even though I was not the CEO, I was getting a little impatient. Then, of course, some other problems popped up—other problems than this 40-nanometer NVIDIA problem.

So I decided to take the CEO position back. In 2009, I did that. There were several priority problems that I had to deal with when I took the CEO job back, and one of them was this continuing problem and controversy with NVIDIA.

I remember that in the first few days after I took back the CEO job, I called all the major customers, including Jensen. Qualcomm was a top customer as well.

In my call with Jensen, he was still very friendly with me, but he also reminded me in a very serious tone that we had the quality, delivery, and manufacturing problem on 40 nanometers.

And this was happening with Qualcomm as well?

Qualcomm was also a major customer. The top customers had not changed very much since then, except for perhaps one.

Apple.

Apple came later. In my call with Jensen, I said I knew about the problem and that it was one of my priority problems, but I asked him to give me a couple of weeks and I would get back to him.

I did have several problems aside from the 40-nanometer manufacturing problem and the argument we were having with NVIDIA. We also had the problem that pricing was dropping faster than costs. You don’t want to see that. Your gross-margin percentage keeps dropping because you have committed to a schedule of price reductions with customers, but you aren’t able to drive down your manufacturing costs at the same rate.

That was one problem. Another problem was the immediate one that triggered me to retake the CEO job. The previous CEO had laid off—he didn’t use the term “layoff”—the worst-performing people according to the performance review. There were about 600 or 700 of them, and he laid them off on the basis of poor performance.

We never did that. The worst we would do was place people on probation for 6 months. Quite often, at the end of the 6 months, everybody would go back to their old job. Some of them would get transferred because they were in the wrong jobs, but we almost never really fired people, even after the probation period.

Under your watch, you never did a layoff, and you never looked at performance reviews—which are meant to help coach people—as the means to determine whom to lay off.

Right. In 2008, of course, there was a financial crisis, and the semiconductor business was affected. Our revenue dropped and our business dropped pretty seriously.

I was not the CEO; I was chairman. I just knew that any general manager or CEO without much experience, in a situation like that, would have a knee-jerk reaction: “This is my test. I have to save as much money as possible, and I have to lay off people.”

But this is the semiconductor industry, and Moore’s law means that no matter what happens, you will always need people.

I know, I know. But semiconductor-industry people actually think the same way as I described. They all lay off people. They are all people, too.

I had a lot of experience at Texas Instruments. At TI, I was not the CEO; I was one of the top managers under the CEO level. When the company decided to have a layoff, the CEO conferred with the top managers, including me. Their first reaction was exactly the same. I’m talking about the early 1970s. Their first reaction about whom to lay off was exactly the same as what our TSMC CEO did in late 2008 and 2009: go by performance.

I was the only one at Texas Instruments in the early 1970s who said no, that would not be a credible way of doing it. People would not respect us if we laid off by performance ratings.

Why? Because performance ratings are done by everyone’s own supervisor. Seven hundred worst-performing people in the company—and who gave those 700 people the bad ratings? Seven hundred supervisors. It is very subjective. It is not something that people will respect.

If in a year you have to hire people back, you have to hire the laid-off people back, then you shouldn’t lay off in the first place. The layoff separation expense is usually about half a year of salary, and it takes at least half a year to train a person. If you need the people back within a year, you shouldn’t have a layoff.

What did you do when you came back as CEO, both about the employment issue and about the customer issue?

To finish the employment issue, the laid-off employees—there were 600 or 700 of them—came to my home to demonstrate and protest.

TSMC was forewarned that hundreds of people would appear in front of my home, so they notified the police department in my district. The police department sent 50 or 60 police officers to try to maintain order.

More than 100 protesters appeared, and my neighbors had trouble getting in and out. That was only the first time. A month or so later, the problem was still not solved. I was still not the CEO, so they appeared again.

About 25 protesters decided to spend the night in a little park about a block away from my home. My wife literally didn’t sleep that night. She would wake up and go over to the window to take a look and see what was going on.

Very early the next morning, at 6:00, my wife got up. She took one of the bodyguards to a neighborhood market and got Chinese-style breakfast—fried bread and other things. She took enough breakfast for 25 or 30 people back to the park and distributed it to the protesters.

They were thankful, and they decided not to go to the president’s palace that day. They told my wife that they would not do that. All of this precipitated my taking back the CEO job.

There was another thing. Before the previous CEO laid off the 600 or 700 people, I told him, “If you want to lay off people, bring it to the board. I’ll call a special board meeting.” I knew what I would ask the board to do, which was not to grant permission.

He decided to circumvent the board because he did not consider it a layoff. It was just punishment for poor performers.

As far as the CEO was concerned, I kept him. I had more than one nice talk with him. I intended to keep him as a potential successor to me. I kept him at the same job grade, with the same salary and bonus, but he was now the president of new businesses.

Back then, we had high hopes for the so-called new businesses, which were solar cells and LEDs.

It’s a great irony that your core business of manufacturing integrated circuits ended up becoming the largest market opportunity of all. You didn’t need any new businesses, and the biggest market opportunity was already in front of you. Why is that so ironic?

I knew that. I did not really think that solar or LEDs would replace our integrated-circuit business. I knew integrated circuits were going to be great. But at that time, which was 2009, we also thought that solar and LEDs were going to be very promising.

It is always interesting to me when companies think, “We should look at other new businesses,” when in reality semiconductors became a $600-billion-a-year market. Solar is a small fraction of that, and LEDs are a small fraction of that. You were already in the best market.

Solar could have been pretty good. However, China ruined it. They subsidized the hell out of it, and they now control that business. Solar-cell prices were extremely low, and they are still low. So it didn’t take off for TSMC.

LEDs did not take off either. The market is not as big as solar, and the patents are controlled by just a few companies. The few companies that controlled the patents would not let up at all.

A few years later, the CEO who was put in charge of the new businesses decided that his assignment wasn’t working out either, so he quit.

He is now running MediaTek, is that correct?

He is now the vice chairman and CEO of MediaTek.

Coming back to this moment in 2009, you offered to rehire anyone who had been laid off who was interested in coming back, and you were setting a new vision and strategy as CEO—or, in many ways, returning to the old one. How did you resolve the NVIDIA dispute?

In the first 4 or 5 weeks after I retook the CEO job, I spent almost half of my time on how to resolve the problem with NVIDIA.

As far as yields were concerned, we were doing our best. We had to do it anyway. It wasn’t just for NVIDIA; it was for TSMC. Qualcomm and Intel were also involved, and 40 nanometers was a very important node in the progression of Moore’s law. Only if we did 40 well could we do 28 well, because 28 was the next one.

I called the salespeople who had been in direct contact with NVIDIA. I called everybody who was somewhat involved in the problem. It became a matter of money.

We were doing what we could on the manufacturing lines. But because NVIDIA had borne the brunt of the problem, they had suffered the damage. I familiarized myself with all aspects of the problem and worked out a number. I also knew that NVIDIA’s customers were making demands on NVIDIA, so I used all the intelligence I could get.

About a month after I retook the CEO job, I sent an email to Jensen. I said, “I’m coming to Silicon Valley next week. I will be at your home at 6:00. Let’s have salad and pizza,” which was something we had done many times in the past.

Jensen immediately sent back an email asking, “When do we discuss business?”

Did he ask who was going to pay for the pizza and salad?

He didn’t ask that. I anticipated it, so I said, “At 6:30, we start having pizza and salad. At 8:00, we’ll go to your study and discuss business.”

On the appointed day, I showed up, and we followed the schedule exactly. At 6:30, we had pizza and salad. His wife, Lori, would make the salad, and the pizza was delivered from outside. Maybe they made their own pizza too; I forgot. I had had it many times at his home.

At 8:00, I looked at the watch and said, “Jensen, why don’t we go to your study?” I gave him the offer. It was more than $100 million.

I also said, “Our offer is effective for 48 hours. If you do not accept the offer within 48 hours, we have to go to an arbitrator.”

That was what he had suggested to the previous CEO anyway. The previous CEO had not even given him a number. He had given him zero.

You probably don’t want to go to arbitration with your best customer.

I didn’t want to, but I had to say that. The number we offered was arrived at after weeks of work on my part, and I thought it was fair to both sides.

Did Jensen accept the offer?

Yes, he did. I think he accepted it within 2 days.

It’s an amazing example of a situation where you had a strong partnership together for many years and had built a close personal relationship, such that you could have an hour-and-a-half family dinner and not talk business. You were then able to come up with a large sum of money—more than $100 million—to settle the dispute, and since then there have been many billions of dollars of business done together.

It’s a great example of working out your differences.

I liked it too. That’s why I included the story in my autobiography.

After the 40-nanometer node, after you fixed these problems, the next node was 28 nanometers. As we understand your story and the company’s story, 28 nanometers is when TSMC really started to take the leadership role at the leading edge of the industry.

How did you decide to commit so hard to 28 nanometers after having had all the problems at 40 nanometers?

I had a lot of trouble at TI. My peak job at TI was head of worldwide semiconductors. TI had many businesses—defense, materials and controls, and others—but the semiconductor business was the biggest, and I was the head of that worldwide semiconductor business.

At that time, our R&D budget was 4.8% of revenue. I thought that was not enough. I wanted to raise it to 5.5% of revenue, but my request was denied every time I raised it.

Coming back to TSMC, I wanted to set a percentage-of-revenue number so we wouldn’t have to argue every year about how much to spend. When I came back, we were running at about 6% or 7% a year, but it was negotiated every year between the R&D director and the CEO.

I wanted to stop that. I wanted to put him at ease so he didn’t have to argue or request the money every year. I literally picked a number. We had been running at 6% or 7% already, so I said, “Let’s pick 8%”—8% regardless of whether there was a recession or not.

That was the best news for R&D. If you asked the R&D director, he would tell you that this was really the best thing we did for R&D. He was not concerned about having his planned budget cut back or his allocation of people reduced. He has been working with 8% ever since.

That is what propelled our R&D effort during this period.

It wasn’t just ramping the R&D budget. It was also capital expenditures. You had had almost a decade of spending $2 billion to $2.5 billion building fabs every year, and in 2010 you ramped that to almost $6 billion.

What was it about the competitive environment and the 28-nanometer node that caused you to push all your chips in on it?

It was a kind of mutual feeding thing. As I settled the R&D budget at 8% of revenue, to the satisfaction of the R&D people, they began to have big ideas. They began to tell me that 28 was going to be the term they used several times.

The first time I heard them use it was when they said, “28 is going to be the sweet spot.”

Like a tennis racket. You hit the ball with the sweet spot of the racket.

Do you play tennis?

I have played tennis—not well.

I was like you 40 years ago. I don’t play anymore, but I know the feeling of hitting the ball in the sweet spot.

The R&D people gave me a lot of technical reasons why 28 nanometers was in the sweet spot, so I decided to believe them. They now had the resources to push it and do it as fast as they could.

As for capital spending, by then we had built up a pretty good organizational infrastructure. We had a good market-forecasting group, and I had set up the business-development department, which was like the marketing department.

We always had a strong sales effort, but to me, sales was the tactical side with customers. Marketing was the strategic side, looking at the outside world.

From all these inputs—from the business-development department, which was our strategic marketing group, and from the technical side, where R&D said that 28 was going to be the sweet spot—I decided that 28 was going to be our tide.

I quoted Shakespeare in my autobiography: “There is a tide in the affairs of men, which, taken at the flood, leads on to fortune.” I decided that 28 nanometers was going to be our tide.

There would be others. 7 nanometers was another sweet spot.

The R&D people told me that, and again I reminded myself of Shakespeare: take it at the flood.

Setting R&D at 8% of revenue did not invite opposition from the board. Suddenly increasing capital spending threefold did invite a lot of questions.

Our practice in board meetings was that most of our directors were from overseas, from the United States and England. We emailed the agenda to them 2 weeks before the board meeting. The night before the board meeting, I invited the independent directors to dinner.

The conversation at that dinner was not on the record, so the independent directors—more than three-quarters of our directors were independent—had the opportunity to ask me questions if they had any.

On the vastly increased capital spending, they didn’t even wait until that dinner. They called the general counsel, who was also the secretary to the board. At that time, the general counsel was an American. They said, “We want to talk to the chairman. We don’t like this idea.”

I talked to them on the phone about a week before the board meeting. I told them what I have told you: that the decision was based on inputs from the market-forecasting group, R&D, and the business-development department.

Of course, they didn’t believe it. You really can’t convince anybody about something like this. At the end, I had to say, “Look, I heard you, but I am still the person responsible for operating the company, so you need to let me go ahead with this one.”

They were satisfied with that.

What was the result? What happened around the 28-nanometer era that created so much demand?

The result was good. It was the smartphone era, and it coincided with 28 nanometers.

When the business-development group and you were looking at this, did you see how big smartphones were going to become and the immense opportunity that would unlock for you?

No, I didn’t. Maybe the business-development guy knew. At least I hope he had more detailed visibility than I did, and of course I hoped that at that time too.

It was not the only input. I had a few other advisers as well.

That takes us to Apple. Could you share with us how you ended up meeting Apple?

Before we do that, let me explain how we made C.C. Wei the business-development director.

When Rick Tsai was CEO, between 2005 and 2009, he split operations into two groups: advanced technology and mainstream technology. C.C. was the head of mainstream technology—the lesser one, really—and Mark Liu was the head of advanced technology.

Each group had a small business-development section, maybe 30 or 40 people each.

When I came back to be CEO, I never thought the split into 2 groups was a good idea. In fact, back in 1996, the president—he was not called CEO then—was an American named Don Brooks. He wanted to split the company.

He had become a little tired of running the company. He had originally planned to stay only a year, but ended up spending 6 or 7 years in Taiwan. Toward the end, he was getting tired of running it and thought he would organize it like TI.

TI had a germanium-transistor department, a silicon-transistor department, integrated-circuit bipolar, integrated-circuit MOS, and so on. It was a divisional organizational structure instead of a functional structure.

I did not think the foundry business was suitable for a divisional structure, because we had almost the same group of customers. How do you divide up that group? Don Brooks wanted to divide it by fab.

TSMC now has more than 20 fabs, so what would you have had—22 divisions?

He only had 3 or 4 fabs back then. But customers move from one fab to another. It is the same customer, not to mention that they use different fabs for different nodes.

He was not convinced, so I said, “Why don’t we get a consultant? McKinsey.”

We brought in McKinsey. After a month or 2—and a couple of million dollars, I guess—they told us the same answer: functional was best.

Don Brooks said, “Tell me one big company that is functionalized.”

McKinsey immediately answered, “Boeing,” which is a good answer.

Except it’s not true. Boeing has commercial and government divisions.

They probably have commercial and government, but they don’t divide by the 707, 747, and 757. They don’t divide the company that way. If we divided by fab, it would be like dividing the 707 from the 757 and 737.

Don Brooks’s attempt was in 1996. By 2005, Rick Tsai decided to cover the same ground. This time I didn’t stop him. My principle, when I was chairman and not CEO, was that sometimes you have to let the CEO make his own mistakes and learn from them—not if the whole company is going down the drain, of course. You have to interfere then, but only then.

That was the background. There were 2 groups when I came back as CEO, and each had a small business-development section. The advanced group had a larger group than the mainstream group.

I wanted to combine the 2 operations groups. I also wanted a real marketing organization. I didn’t call it marketing because I decided to use “business development” in English; it has a good translation in Chinese.

The advanced group had around 10,000 employees, and the mainstream group had 7,000 or 8,000 employees. The mainstream group was taking the older fabs and higher-nanometer nodes and finding customers that didn’t necessarily need the leading edge—automotive parts or CMOS sensors for cameras, for example—to keep utilization high on those older fabs.

The same customers often use both mainstream and advanced technologies. Qualcomm uses the most advanced processes as well as older ones. Apple does too. If you think about all the chips in an iPhone, the A-series processor is built on the leading edge, but there are many other chips in there.

Right.

So you combined them into one business-development organization of about 80 people, with Mark Liu in charge of the advanced group and C.C. in charge of the mainstream group. The question was who would be in charge of the combined organization.

We had a lot of operational talent. By operations, I mean manufacturing—taking the developed technology from R&D and converting it into mass production. We had a lot of talent there.

But we didn’t have much business-development or marketing talent, and neither Mark nor C.C. had any real previous experience in marketing or business development. That was my main worry.

We needed a combined operations manager, but even more importantly, we needed a good business-development manager.

I first offered the marketing and business-development job to Mark, who was in charge of the larger advanced group. I explained to him that I did not think he had significant marketing experience, and that this new job would give him an opportunity to become proficient in that area.

He declined. He said, “I have 10,000 people reporting to me now. You want me to take a job that has only 60 or 70 people?”

Your goal was for him to become a well-rounded executive, in hopes that he would lead the company after he had done that tour of duty.

I explained to him that these 60 or 70 people were very important. They were responsible for finding all the next businesses.

Before this period, you were doing the business development and marketing for the company. You were the one finding the NVIDIAs, the Jensens, and the Broadcoms—the next great customers and great markets.

That’s right, except for the 4 years when I was not the CEO. I was on a plane most of the time visiting customers. That was my pleasure. I really liked it.

I then offered the business-development job to C.C., and he accepted it. I thought he accepted it quite delightedly.

He is now the chairman and CEO of TSMC.

This had just happened, and you came home from a board meeting one evening.

The board meeting had ended at 6:00 or later, and I went home. Our board meetings were held here in Taipei, in the conference room across the hall.

It was about 6:30 when I got home. My wife knew I would not be home until around 6:30, because she met me at the door, which wasn’t very often. This time, she had something to tell me.

She said, “Terry Gou called in the afternoon and said he was coming to dinner.”

For listeners, Terry Gou is the founder and CEO of Foxconn.

Terry Gou is a relative—actually, a second cousin—of Sophie, my wife. They share the same grandparents. Terry was also the chairman of Hon Hai, which is Foxconn in the United States. It is a very important supplier to Apple and a pretty big company. Terry Gou is reputed to be one of the richest men in Taiwan.

My wife, Sophie, is lovely, but she doesn’t know much about my business. I don’t think she understood the significance of Terry Gou coming to dinner with a vice president from Apple.

You had been trying for months, strategizing with the business-development team about how to win Apple’s business. The strategy seemed to be working.

“Strategizing” is probably too strong a word. We were thinking about it and knew we couldn’t do anything about it. Apple is a very close-mouthed company. If you try to talk to them or offer your service, they just tell you to go away. They will come to see you when they are ready.

That is what I knew about Apple then, and I know the same thing now.

Sophie knew I would not be home until after 6:00, so Terry had set their arrival time for 8:00. That was a bit late for my dinner, but I said, “What the heck?”

They showed up. Sophie just said it was a vice president. I thought to myself that it wouldn’t be an ordinary vice president. There was no reason for Terry to bring just any Apple vice president to my home. It must be someone special for TSMC.

Jeff Williams came. He was not just a vice president; he was the chief operating officer of Apple. He was a straightforward person and didn’t spend much time on ordinary chit-chat.

This wasn’t the same pizza-and-salad period as before.

It wasn’t formal, either. My wife and I had a cook, and she was a pretty good Chinese cook. She didn’t cook Western food. Sophie just told her to add a few dishes.

Terry grew up on Chinese food, and I imagined the Apple executive would also like Chinese food. The food wasn’t important. The quantity and quality were not important, because Jeff almost immediately started his pitch as soon as he sat down to dinner.

What was the pitch from someone like Jeff Williams? “We would like you to provide wafers”?

It was pretty straightforward. I listened. That night, Jeff talked maybe 80% of the time and I talked 20%, not counting the relative-to-relative conversation between Sophie and Terry, which wasn’t very much either.

Jeff had proposed economic terms at this first dinner, right?

Nothing concrete. He just said that he would let us have a 40% gross margin.

I didn’t answer him, because our margin at that time was already 45%, and I was trying to push it up to 50%. There was an announced effort in the company to push the gross margin up. I had that effort for many years after I came back to be CEO, and I didn’t succeed even by the time I retired.

Of course, later there was COVID and we achieved technology leadership, so our margin jumped to over 50%. But when I retired, I was still slightly short of 50%.

When you say technology leadership, around the 28-nanometer node you were one of a select few at the leading edge, but there was fierce competition. Once you got to 7 nanometers or so, that’s when you really pulled away.

You are neglecting Intel.

I’m neglecting Intel. Fair enough. Look where we are in 2025.

At 28 nanometers, we were definitely the leader among foundries and perhaps among a few other companies, such as Texas Instruments, but not Intel.

Was Apple considering Intel?

Apple was not actively considering Intel at that time. That came later.

Take us there. After the initial conversation with Jeff Williams in November 2010, you had the financial discussion.

He said he would let us have 40% gross margin. My thought was, “We are already at 45%.” But I also thought he was trying to be generous.

This dinner was not the time to go into a pricing discussion. We had many other things to discuss. We were almost in production with 28 nanometers, so I asked him, “What node do you want?”

He said, “20 nanometers.”

That was a surprise and, frankly, a disappointment. The Moore’s-law progression after 28 was going to be 16 nanometers. Apple wanted a half-step. A half-step is a detour.

We would have to spend effort on 20, which of course would help us with the next node, 16, but it was still a detour from 28. If R&D could go directly from 28 to 16, it would take less time than doing 20 first and then 16.

Back then, R&D did not have enough resources to do 2 nodes at the same time. Later, we did.

You had just spent $6 billion in capital expenditures the previous year, going all in on 28 nanometers. You were asking Apple, potentially your biggest customer ever, what it wanted, and it said no—we want you to go do something you weren’t planning to spend money on.

You were left with the question: Is it worth it to land Apple as a customer?

It wasn’t that serious, because when we figured out that there was a very big market for 28 nanometers, we planned to increase capital spending. We did not have Apple in mind. Apple came as a surprise.

It was still a question of whether you were willing to spend potentially $10 billion over the next few years doing 20 nanometers for Apple when you weren’t planning to do 20 nanometers at all.

That is where our connection with Goldman Sachs came in.

I planted a lot of seeds when I ran TSMC. I knew that one of these days we would probably need top-level investment-bank advice, so we established a good relationship with Goldman Sachs very early in our existence.

I was, in fact, a board director of Goldman Sachs. We did the ADR with Goldman Sachs, which opened a good relationship with them.

That was the New York public listing of TSMC’s stock. ADRs are American depositary receipts.

It is a separate market. In fact, right now the TSMC ADR price has a 20% premium over the Taiwan price.

You need TSMC board permission to convert your shares to ADRs, otherwise you could arbitrage the difference.

We don’t want that. The board has to approve any conversion of ordinary Taiwan TSMC stock to ADRs, and the board does not give such permission easily.

As I was saying, we needed funds. The Apple request came after we had already decided to increase capital spending. Apple required even more capital spending, and we had to figure out where the cash would come from.

There were several possibilities. We were paying a dividend—not a big dividend, but a modest one. We could cut the dividend. We could sell stock through a new offering in Taiwan or the United States. We could borrow money, perhaps through convertible bonds.

Or you could fill only part of Apple’s order.

We did that. First, we did our financial planning. We decided not to cut the dividend and not to sell new stock. We decided to borrow.

I looked at the numbers, and after prudent financial planning, I decided to take half of what Apple said it needed.

It seems like it would be in a customer’s interest to come to you and say, “I need to buy zillions of chips from you. I need all your wafers,” because Apple has no skin in the game if you spend all the money. It is based on their words.

Back in the 1990s, in the first 10, 12, or 15 years of our existence, we were short of capacity almost all the time. What you just said happened all the time.

We figured out that we would require a deposit from the customer and would confiscate the deposit if the time came for the customer to take the wafers and they didn’t.

I told the salespeople in San Jose, “Tell the customer that we need a deposit from them. It is our money, and it is only their word. They may not want the wafers when the time comes. Tell the customer we will confiscate the deposit.”

The salespeople had never heard anything like that before. They were delighted. They could stand up and tell the customer that we might even confiscate their money.

Of course, we never confiscated any money. It did happen quite often, particularly in the 2000 or 2001 period. There was the internet recession. People were starting companies called Pets.com or something.

The recession trickled all the way back to semiconductors. TSMC’s revenue took 3 years to recover to the previous level.

A number of customers had placed deposits in anticipation of normal good times. We built the plants. In fact, we bought a couple of other companies, so their plants and fabs became ours.

The customers didn’t need the wafers anymore and didn’t need the output of those fabs. We did not confiscate their deposits. We let them delay their demand, and eventually every one of them used up the deposits.

At this point, in early 2011, you went to Apple and said, “We are prepared to serve half the number you told us about.”

First, the relatively new business-development director, C.C., had the privilege of telling the lower-level purchasing people at Apple. He got a response: “You must be crazy.”

C.C. did not comment on that. At least, he said he didn’t. He brought it back to me, and then I went to Apple myself and talked to Jeff Williams.

I said, “We have to issue corporate bonds. After prudent financial planning, we decided that we will take half of what you asked for.”

He was quiet. He made one suggestion: “I think you can eliminate your dividend. Your shareholders will understand.”

I said, “No, I don’t think so.” I had looked into that. That was another reason for having high-level consulting advice. About one-third of our shareholders were seriously interested in the dividend. If we did what Jeff suggested, our stock would drop like hell and trigger a sell-off.

When I talked to Jeff at Apple’s headquarters in Cupertino, he took it fairly willingly. The only suggestion he made was eliminating the dividend. I said no, and he let it lie there.

The issue was settled—how much demand we would take and how we would get the money. We still had to borrow billions of dollars, even with half of the demand.

This was, especially after the investment in 28 nanometers had depleted your reserves, a bet-the-company move. You were taking on a bunch of debt to build the fabs to make this happen.

I didn’t think I would lose.

You sound like Jensen. That is exactly what Jensen said.

The financial discussion with Apple had already happened when Jeff Williams called me in February 2011.

He said we needed to pause our discussions for 2 months because the highest level of Intel had approached Tim Cook and asked Tim Cook to consider Intel.

At that time, Intel was the major supplier for all Macs. Apple’s Mac line was all Intel.

That wasn’t an issue for the iPhone, but Apple had a close existing relationship with Intel.

I don’t know what relationship they really had. It must have been closer than mine.

I accepted Jeff’s request. I was not all that worried because, in 2011, Intel was no longer a name that, when you heard it, made you stand up and bow.

In the 1990s, in the late 20th century, Intel was a name in semiconductors that made you tremble with fear. If you heard that they were in competition with you, you would think, “My goodness.”

This is why you started TSMC as a pure-play foundry business. You didn’t want to compete head-to-head. You said that TSMC should not be an integrated device manufacturer, both designing and manufacturing chips, because you would never catch Intel.

I didn’t quite say that we would never catch them.

Fair enough. Look where we are in 2025.

I reviewed in my mind all the characteristics that Apple was looking for in a supplier. At that time, we thought we were almost at parity with Intel in technology. In manufacturing, I thought we were better than Intel. And in customer trust, we thought our customers trusted us more than Intel’s customers trusted Intel.

So I was not too worried.

I also thought that when Jeff Williams told me that the highest level of Intel had approached Tim Cook, he was talking about somebody like Andy Grove, who was retired by then. It turned out that he was talking only about Intel’s CEO.

Would that have been Bob Swan or Paul Otellini?

It was Paul Otellini.

Today, Intel doesn’t make the chips in the iPhone. TSMC makes all of Apple’s chips. What happened?

I wasn’t too worried, but it was still in my mind. About a month after Jeff called to tell me to pause for 2 months, I decided that I would pay Apple a visit and ask what was going on.

I emailed Jeff and asked for an appointment. I said I was coming to Silicon Valley anyway, which was normal, and would stop by on a certain day.

Jeff replied, “Come, but I won’t be here. I have asked Tim Cook to see you.”

That was an extraordinary privilege. Normally, someone says, “Someone on my team will see you,” not, “My boss will see you.”

Exactly. It was usually the other way around.

I showed up, and Tim was very nice to me. He took me to lunch in the cafeteria, where there was a lot of food. We each picked our food, carried our trays back to his office, and ate there.

He told me there was nothing to worry about because Intel simply did not know how to be a foundry.

That was a very short but satisfactory answer.

What is your interpretation of the meaning behind that statement?

I was thinking about the three characteristics I just described. Technology and manufacturing were two of them. Subconsciously, I think I interpreted Tim’s explanation as referring to the third one: customer trust.

Intel had always acted superior. Before Apple became our customer, I knew many Intel customers in Taiwan. All the PC makers were Intel customers, and none of them liked Intel.

Intel acted as if it were the only company, the only supplier of microprocessors.

That was for Intel’s microprocessor business. Here, we are talking about the foundry business, where TSMC’s core principle is that it does not compete with its customers.

Even if Intel was trying to do business in good faith, it had a conflict because it also designed chips that competed with Apple’s chip designers, NVIDIA’s chip designers, and other customers.

I don’t think Tim meant that. I think he meant that customers ask for many things. We have learned to respond to every request. Some requests are crazy, and some are irrational, but we respond courteously to each request.

Intel has never done that.

I knew many of Intel’s customers in Taiwan. None of them trusted Intel or liked Intel. They all wished there were another supplier.

To finish the Apple story, the short answer is that TSMC worked on 20 nanometers. Were there any trade-offs where pursuing 20 nanometers and spending the billions of dollars cost TSMC in some way?

There might have been a cost. The story certainly does not end there.

There was pricing. Everything was not easy, including pricing. Jeff came himself, and we talked about pricing. We had done our homework on costs and on what price we would accept.

Jeff came and told us to name a number. He gave us his reasoning: he had to make his component costs meet a certain goal.

Could you tell us a little more about what goes into pricing? I imagine things like the yields you think you will be able to get have a huge impact.

Of course. The main thing that goes into pricing is cost. The second thing is where your desired price will be accepted by the customer.

One thing that has occurred to me is that TSMC now gets mid-50% gross margins—55% or 57%—higher than in your time. Many of your customers have 70% or 80% gross margins. TSMC is creating a lot of value, and the designer is creating a lot of value. How do you sort out who gets to capture the value?

I don’t get the privilege of sorting it out now. C.C. has the pleasure and duty of sorting it out.

In general, you try to find a middle ground, which is different for every CEO. Every CEO who wants to protect his reputation says, “I worry about the long range,” but in truth, not everyone does. It is a very personal question how to sort these things out.

For many CEOs, there is no choice. As a supplier, you have to accept a certain price, particularly if it is a commodity.

We have not finished with Apple yet.

Please, let’s finish Apple.

You were asking whether there were any trade-offs. There was a significant trade-off, and that was the detour I mentioned. In 2011 and 2012, our R&D was not strong enough to do 2 nodes at the same time. Now it is, but back then it wasn’t.

The trade-off of accepting the 20-nanometer technology was that we delayed development of the 16-nanometer node.

Samsung came up with 16 nanometers. They had lost the 20-nanometer business, so they went ahead of us in the 16-nanometer department because they got to skip 20 nanometers.

I was shocked when I heard that Apple had placed its first 16-nanometer orders with Samsung.

That must have been a real shock. You invested so much, and even though you took only half of the original demand, it was still tens of billions of dollars. You were counting on converting at least 80% or 90% of that equipment to 16 nanometers. If Apple went to Samsung for 16 nanometers, where did that leave you?

I was really shocked. I emailed Jeff Williams right away. I said, “We invested in all this equipment, and we were counting on you to take the 16 nanometers from us. Now we find out that you are buying the first 16-nanometer chips from Samsung.”

Jeff replied immediately: “Don’t worry. I’ll be in San Jose next week and explain it to you.”

That relieved me a little, though certainly not completely. The following week, he showed up and explained. He said, “As soon as you are ready with your 16 nanometers, we will buy from you. We will buy all of our needs from you when you are ready.”

That completely relieved me. That was what we were supposed to do anyway.

What he said was true. We developed our own 16-nanometer process about 6 months later, and most of Apple’s 16-nanometer requirements still belonged to us.

Most of them.

Most.

It illustrates the brilliance of TSMC and the pure-play foundry business model. Samsung is Apple’s chip competitor.

I know. I said in my autobiography that, sitting in the foundry business, I see a lot of things before they actually happen.

Let me tell you the IBM-Qualcomm story.

Qualcomm was a prime candidate to be our customer. We really wanted Qualcomm because we knew they were a technology house.

What year was this?

This was way back, in the 1990s, when we started. Qualcomm was part of that initial wave of fabless companies. Irwin Jacobs started Qualcomm before I started TSMC, a few years earlier.

From the early 1990s through the latter part of the 1990s, we wanted Qualcomm to be a customer. I often saw their operations vice president—that is what our customers called their purchasing people. I saw him often, and he was always polite, but he gave us very little business.

I knew that their main foundry was IBM. Sometime in the latter part of the 1990s, perhaps 1997 or 1998, he started telling me that Qualcomm would use us. He didn’t tell me who our competitor was, but I knew from other sources that it was IBM.

Our business with Qualcomm rapidly increased after that period. I immediately knew that IBM Semiconductor was in trouble. IBM had its own fabs, but its main business was supplying Qualcomm and a few other small fabless companies.

The next step IBM took was to ask TSMC to co-develop the next generation of technology, 0.13 microns, or 130 nanometers, in 1999.

Since I anticipated that, it was no problem for us to refuse. Even if I had not anticipated it, we would never have accepted that kind of co-development. IBM still considered itself the senior partner in any partnership. We would have been the company sending engineers to IBM.

If we did that, we would lose our ability to develop our own process and would have to depend on the co-development. The co-development would have had a lot of difficulties. Our people had a different culture.

We declined without having to think about it. IBM was angry. They thought we were still a small, backward Taiwan company, while they were big.

They immediately went to UMC, and UMC accepted the proposal, only to seriously regret it.

In 1999, was UMC a peer of TSMC in Taiwan in terms of volume and size?

Not by 1999. They were already smaller.

This is a good time to go back to the learning curve, speaking about the importance of owning your own technology and process at the leading edge and controlling your own destiny. You developed the learning curve.

I did not develop it, and I certainly did not initiate it. I had a role at TI in refining it to the point where a semiconductor company could use it effectively. That was my role.

How would you explain it to a novice?

Explaining learning-curve theory is simple, but one would be foolish to take the simple explanation and think that is all there is.

The simple explanation is that, as you make more of one thing—anything—your unit cost goes down. It started with refrigerators and cars. If a company makes more cars, its cost per car goes down. That is why it is also called the experience curve: you gain more experience and become more efficient.

If one takes that simple explanation and thinks that is all it is about, one really hasn’t learned anything.

Bruce Henderson, who is now considered the father of strategy, founded the Boston Consulting Group. There is a branch of business economics called competitive strategy, and Michael Porter was at one time considered a major figure in that field. He wrote several big books—700 pages each. I have all of them.

His original Competitive Strategy memo, I think, was about 20 pages. It is still some of the best business writing ever.

Michael Porter was a director of TSMC at one point. I had a story about him in my autobiography, but because of time we probably won’t go into it.

We will talk about Bruce Henderson. He is now considered the father of competitive strategy. He came to Texas Instruments around 1970. He first called the TI CEO, Mark Shepherd, and told him that the Boston Consulting Group had an experience-curve theory that would benefit the semiconductor industry.

TI was the largest company in the semiconductor industry then. Mark Shepherd invited him to make a presentation. Bruce Henderson brought Bill Bain with him. Mark Shepherd invited the CEO and me to attend the presentation.

It was an eloquent presentation because Bruce Henderson was a very eloquent man, and Bill Bain was there on the side.

Mark Shepherd was impressed and decided that TI would work with BCG on the learning-curve theory. Bruce Henderson assigned Bill Bain to work most of the time at TI—about 3 days a week—and Mark assigned me as TI’s person.

Bill Bain and I became partners. I assigned him a small office close to mine in the same building. He needed a lot of things from me, including permission to get our costs and prices. We had many families of integrated circuits and transistors, so he had many requests.

It was easier when he was nearby. Every time he arrived at an interesting or useful conclusion, he would discuss it with me. We had a very pleasant association for 2 years, maybe even more.

He would fly to Dallas every Monday and go back to Boston either Wednesday night or Thursday night. Every time he went back to Boston, it was to tell Bruce Henderson what he had done that week.

After a couple of years, Bill Bain came to see me one day. He said, “You are the first person I’m telling outside the Boston Consulting Group. I am leaving BCG to start my own consulting company.”

I asked why. He said, “There is a world imperative.”

That was the first time I heard that term.

He meant a world imperative for himself?

For himself, apparently.

It seems fundamental to the industry that you want to get through the low-volume period as fast as you can. Ideally, you spend no time in the low-volume period. Over time, the winner is the one with all the volume, because that gives you the lowest prices.

There is a flywheel: once you have the lowest prices, you get all the business, and then you can reinvest in the next node.

Right.

You couldn’t have known that TSMC would be the winner, but once you internalize the learning curve and globalization, you can intuit that in the future there will be one winner in semiconductor manufacturing.

I couldn’t have told you that TSMC would be the winner, but I understood the learning curve.

As our time comes toward a close, one question David and I wanted to ask is this: TSMC is essentially the only trillion-dollar company in the world not on the West Coast of the United States. It is an incredibly important company in the world, an unlikely success of grand scale. Was that unlikely in your opinion?

There are many things I’m not going to argue with you about.

I merely ask as a point of curiosity. You started it when you were 56.

I didn’t realize it was that unlikely. TSMC’s size and importance exceeded my expectations, but not by an order of magnitude.

Wasn’t the original plan to stop building after Fab 2?

No, that was only the very initial plan. We were never going to stop there. We were just talking about learning. If I didn’t know anything about learning, I might have said, “Maybe we stop after 2 fabs.” But I was a serious student of the learning curve, and I would never stop at only 2 fabs.

Here is why I say it was an unlikely success. There were so many reasons why the original incarnation of TSMC was kind of a bad business. Fabless was not a thing yet, so your initial customers were integrated device manufacturers—the Intels of the world—and you were taking their worst excess capacity.

You were their second-source supplier for the least critical, least leading-edge, least interesting stuff that they didn’t want to make on their own. Did you see fabless coming, or was that a lucky thing?

I saw it coming. In fact, I had dinner 2 months ago with the first guy, Gordon Campbell. Have you heard his name?

Gordon Campbell came to see me at General Instrument during my final months there. He did not know I was leaving. Frankly, I did not know I was leaving when I saw him.

He wanted funding from General Instrument—$50 million. He said he wanted to start a new company. I asked, “Do you have a business plan?”

He said, “No, it’s all in my head.”

I said, “I need at least a business plan. I have to go to the board of General Instrument.”

He said, “All right. I’ll send it to you within 3 weeks.”

Three weeks later, there was no business plan. I was interested because I knew he had a good reputation for starting companies. I called him, and he said, “Morris, I’m sorry I didn’t send you anything, because I don’t need you anymore.”

I asked, “How come?”

He said, “I don’t need $50 million anymore. I need only $5 million, and I can gather that up very easily.”

I asked why he needed only $5 million.

He said, “I’m not going to build a fab.”

That was the start for me. I saw that there would be fabless companies.

Another guy came to General Instrument and said he had already started a company called Atmel. They did not have any fabs, and he wanted General Instrument to make the wafers for them.

Back then, General Instrument had empty fabs. I told the semiconductor manager of General Instrument to go ahead and work with him.

Don Valentine, who I’m sure you knew, had a great quote when asked about starting Sequoia. He said, “I had an advantage: I knew the future.” It sounds like you knew the future too.

At least I had a glimpse of it.

At Atmel, the customer wanted the fab to be run his way. The General Instrument semiconductor manager wanted to run the fab his way. General Instrument owned the fab, for heaven’s sake.

That was a very early situation in which the difficulty and advantage of running a foundry business appeared. The difficulty was that you have to satisfy a lot of customers, and everyone wants the fab to be run his way. But you can run a fab only one way, in a way that more or less satisfies all the customers.

The advantage, of course, is that you have a lot of customers.

Morris, we can’t thank you enough.

Thank you. It was my pleasure, even though it is the first time in a long time that I have talked for so long.

We appreciate it. Thank you for doing it with us.

David and I are coming at you now from our home studios back in Seattle and San Francisco. We wanted to do a little postgame on that interview—a little analysis, our conclusions, and the things that are still sitting with us a few days later after crossing the ocean.

This felt essential to me because it felt like we were recording history with Morris. I didn’t want to interrupt him to make a business-model point. It felt like we should let him talk, and then we could do our part afterward.

Fortunately, we have a model for doing analysis at the end of a story: our Playbook. Let’s do it.

The first thing I can’t shake is this idea that, in hindsight, it was genius not to compete with your customers—to be the dedicated pure-play foundry.

We saw Morris’s original pitch at the TSMC Museum of Innovation. It was his original slide deck, his original business plan, which he pitched to the Taiwanese government and then to investors. One of the bullet points is right there: “Dedicated pure-play foundry.”

At the time, I get the sense it was more about “What can we win at?” than “What will be the most important and valuable semiconductor company in the world in the future?” They didn’t have the capabilities to design chips and products, and that capability didn’t exist in Taiwan, so it was impossible for them to compete with customers.

This was all they could do.

It crossed Morris’s mind that they could compete with Intel, but he scrapped that because the thing they were good at was manufacturing.

It is almost an accident of history that the pure-play foundry ended up being the best way to do this. At least it is the best way as measured by market capitalization versus other foundries and integrated device manufacturers such as Intel.

It is also the path that has led TSMC to operate essentially alone at the leading edge. It has surpassed all the other integrated and quasi-integrated chip foundries in technology.

That is the first thing. You can connect the dots looking backward, as Steve Jobs said, but looking forward is difficult.

The primary reason TSMC has worked so well is that it doesn’t compete with customers. It is truly the only foundry at the leading edge that does not in any way compete with its customers. It doesn’t have its own end-product division, and it doesn’t design its own chips. It truly only serves its customers.

If you ask how the world arranged itself so that there could be a trillion-dollar company that does not do chip design, chip architecture, EDA tools like Cadence or Synopsys, or its own equipment, the answer involves the rise of Arm.

If Intel and the x86 architecture had maintained their dominance, you wouldn’t have had this window for the value chain to rearrange itself. Arm became a dominant architecture in phones, then computers, then servers, and now all these AI chips are built on top of it.

You had a standalone architecture company, a standalone manufacturing company, standalone EDA companies, and standalone designers like Apple and NVIDIA. A lot of that is due to Arm.

Arm and TSMC are coupled at the hip of history in terms of when this came to be.

One enormous example is Apple. If Arm hadn’t become such a viable CPU architecture and Apple hadn’t standardized its Apple Silicon on Arm, Intel might be making all of the leading-edge chips that go into your phone.

Apple already had its Intel relationship for the Mac. Macs were running on Intel x86 chips. If that had remained the dominant architecture, the common case would have been a fully integrated Intel.

You have to keep peeling the onion. This supposes that Intel could have gotten its act together and made a chip for mobile phones that was performant. Maybe the baggage from x86 structurally prevented that from happening.

I think all of this is true, but if Arm hadn’t existed, there would have been nowhere else for this vector of innovation to go.

The point is that there is now a standalone architecture company, a standalone manufacturing company, standalone EDA companies, and standalone chip designers. That is partly due to Arm.

Which brings us to Hsinchu Science Park. Going there in person was very different.

We talked about this on our original TSMC episode. Even if you wanted to, you couldn’t airlift TSMC and this capability out of Taiwan and recreate it somewhere else. We talked about that in an abstract way before, but driving around the Science Park made me feel it physically.

It is like if Silicon Valley were all in one government-sponsored industrial park. It isn’t just TSMC there. All of its partners and customers are there too.

We were driving by a Cadence building, then a Synopsys building, then an Arm building. Qualcomm is there. MediaTek is headquartered there. The whole ecosystem is physically instantiated in one place.

The craziest thing is that there are 2 universities right there in the Science Park, producing PhDs every year who are absorbed into the ecosystem.

It would be like having 2 universities on the NVIDIA campus, cranking out people directly for the companies there.

You always hear people talk about how integrated this ecosystem is. Synopsys has to be closely tied to TSMC to understand what the next node will look like, so it can make it easy for people using Synopsys tools to design chips that can actually be manufactured on TSMC’s process.

You understand it when everyone is walking across the street to each other and having extremely close communication.

Ben, both of our flights felt like pilgrimages for chip-design and fabless companies going to Taiwan to meet with this ecosystem. My plane felt like the semiconductor version of the technology buses that go from San Francisco down to Silicon Valley every day.

The backpacks on the plane were a giveaway: Google, Amazon, Arm, Marvell.

It raises the point of TSMC’s Arizona fab and its fabs outside Taiwan. Why is TSMC doing that? It is not the leading edge, not big volumes, and not leveraging the close geographic ecosystem in Taiwan.

There are customer and government reasons to build fabs in other countries, but you are not going to recreate the magic of the ecosystem physically instantiated in Taiwan.

It would take decades to recreate the ecosystem in the Science Parks.

As we were driving around, we said this must be the single most successful government-funded industry initiative of all time—at least anywhere in the world to spur innovation with this particular mandate.

The land-grant universities in America are somewhat comparable, but this was a rifle shot: “We are going to spur semiconductor-industry innovation in this industrial park, in this location.” It worked.

There you have one of the 10 most valuable companies in the world and the only—or one of the only—trillion-dollar companies not on the West Coast of the United States.

It worked. The scale is incredible too.

We drove by a construction site where it looked like a quarter of the building was done. This is where they are making the 2-nanometer process, which presumably will be in the next iPhone.

No one said that, but I guess they are making lots of NVIDIA GPUs and lots of iPhone chips there.

The first phase of the 2-nanometer facility was open. I think it was a quarter of the building, with 3 other phases not even ready for prime time yet. They are doing small production runs and getting ready to ramp the 2-nanometer process in the second half of this year.

The scale of the physical buildings smacked me in the face. I felt like I was looking at the Sphinx in Egypt. They are huge—many football fields in size. Each phase of a fab is an enormous building.

I felt a little bad for saying that the original TSMC business plan was kind of a bad one: taking excess capacity from Intel and other IDMs and giving them a place to manufacture their least critical, least leading-edge, least interesting chips.

But that was true. Morris believed fabless was coming, but for at least the first 5 years, the only real business TSMC had was IDMs willing to say, “How cheaply can you give me some manufacturing capacity?”

It wasn’t strategic, but it was revenue.

That is a major difference between Intel’s fab strategy and TSMC’s.

Intel is constantly taking its existing fab footprint and repurposing or upgrading it for the leading edge. On the one hand, that utilizes the assets for the highest-value products. On the other hand, Intel loses the manufacturing capabilities for older process generations.

Demand doesn’t go away for those chips. It goes away slowly.

Replacement parts are a great example. There are technology systems, products, manufacturing equipment, and automobiles built 10, 20, or 30 years ago that have specific chips made on older process technology. When those chips break and need replacing, you need those exact same chips.

This is the business TSMC started in.

That is the fundamental philosophical difference. Fab 1 belonged to ITRI, the government organization where Morris was president before taking the helm of TSMC. Fabs 2 and 3 were the first TSMC-specific fabs, built in the late 1980s, and they are still running.

In addition to replacement parts, there are still applications for older nodes. CMOS sensors are a great example. The cameras we are talking into right now have great Sony sensors. They don’t require a 2-nanometer process, but they do require etching silicon in the same way you etch a chip. That is a specialty use case for TSMC’s older fabs.

Those fabs are fully depreciated on an accounting basis.

There is maintenance capital expenditure, of course, but the initial capital expenditure is fully depreciated. You are getting very high-margin dollars out of those old fabs.

It is not necessarily a better or worse decision than what Intel has historically made, but it is a different one. Intel keeps closing the old stuff so it can own a smaller footprint and focus everything on making the latest and greatest. That is not what TSMC does.

I am obsessed with the idea that Morris went on the record and said, “I knew fabless was coming.” He had a couple of great anecdotes about that.

In older interviews, he sometimes says the timing was lucky. He also said that TSMC’s growth wasn’t very high in the early years because they were waiting for customers to emerge.

But it really is the idea that he saw the future and made a bet. He did a crappy business to build up competency, capability, volume, and capacity—to build literal fabs and be there when the fabless revolution happened.

Exactly. To build up literal fabs.

I don’t know whether he was within 12 months of when he thought it would happen, but in his memoir, you are reading the story of the early customers in years 5, 6, and 7, and the majority of the business is still not fabless. It is someone else’s worst orders.

That gets to the heart of learning-curve pricing, which we only discussed tangentially with Morris. It is worth dwelling on.

What is the learning curve?

The core insight of the learning curve, from BCG, Bill Bain, and Morris, is that the goal is to become the largest-volume player at the end of the game.

If you take it as a given that the largest-volume player wins in a fixed-cost, economies-of-scale business, because it can spread fixed costs over the maximum number of customers, then the question becomes: How do you get to the maximum number of customers?

In the early stages, when the game is competitive, you accelerate your pricing toward where you think it will end up. You can even start out unprofitable on a given node generation because the goal is to outcompete everyone else, become the dominant player, and get all the customers.

Once you aggregate that demand, you get economies-of-scale pricing.

It works backward from the end state. It involves a lot of market sizing. At maturity, what do we think demand for 40 nanometers will be? How many chips will be made on 40 nanometers?

To have the cheapest price for customers, you need the largest orders. Then it becomes a question of how fast you can get into volume production.

Everyone intuitively grasps economies of scale, but the implications across the whole business are enormous. There is a strategic-finance question of when to take on debt, when not to, when to issue shares, and how to orchestrate everything.

The Apple example makes it concrete. You are about to get the absolute whale customer. The learning curve says you want to take the deepest possible position and accept all of Apple’s orders.

But that exposes you to existential risk when you are not yet within spitting distance of producing that volume on your own.

Is it worth betting the entire company? You have to be precise in forecasting ultimate market demand. In Apple’s case, that means forecasting how many customers will buy the next-generation iPhone.

In NVIDIA’s case, it means forecasting how big AI will be. It is a crazy thing for a manufacturer to have to do: maintain a crystal ball into the end markets.

If you are off by 5% or 10%, that can tank the profitability of the entire node generation, which tanks free cash flow and means you can’t play the game on the next node.

If you are good at all of this and execution is flawless, then once you internalize the learning curve, the story of TSMC goes from surprising and unlikely to inevitable.

Of course the company taking on all the orders to have the lowest price becomes the winner. It is inevitable that this industry ends with a dominant player. It now costs on the order of $20 billion to build a new fab. Eventually it will cost $40 billion, $80 billion, or $100 billion.

How many players will be left standing with the ability to deploy $100 billion to build a building full of machines? This market has natural-monopoly characteristics.

That is just the capital-expenditure side. As Morris discussed, there is also the R&D side required to create the next process node that can be built on that capital equipment.

It is crazy that if you look at TSMC’s capex and net income each year, they grow in very similar ways. The company spends an enormous amount of money, even before accounting for R&D.

If TSMC looks at competitors and asks how much they can invest, TSMC can invest more than anyone else because it has the most volume.

On top of that, TSMC spends separately on R&D for its manufacturing processes. That is how you get CoWoS, the packaging technology it uses for AI chips. That is its proprietary technology.

Once you have proprietary packaging, it is even harder for customers to go and double-source elsewhere. TSMC has similar technology for packaging mobile chips that does not use CoWoS.

It seems like this is a market where those in the lead are only going to get further ahead, absent a major strategic mishap or execution mistake.

Totally.

The last Playbook theme for me is that Moore’s law is undefeated.

At the end of the day, going back to Morris’s career at TI and his contemporaries Jack Kilby and Robert Noyce and the invention of the integrated circuit, once the integrated circuit was invented, the compounding growth of the industry was all that mattered.

Everything else is downstream of the fact that the world is going to demand more computing at this monotonically and exponentially increasing pace—every 18 to 24 months.

The technical definition of Moore’s law expired a long time ago, but spiritually, the world demands roughly twice the computing power it had 2 years ago, every 2 years. That has continued for 50 or 60 years and shows no signs of slowing down.

No signs of slowing down, except that we keep hitting theoretical physics limits.

I meant that the demand side of the equation shows no signs of slowing down.

The demand side is far more than 2 times. Moore’s law has always been about how much can happen on the innovation side of design and manufacturing, and that is getting harder.

We are having to call more things Moore’s law. Packaging was never part of the original Moore’s law. Software improvements and proprietary interconnects are part of it now.

My point is that it is a self-reinforcing system. As long as the demand is there—if the world wants twice as much computing as it had yesterday—there will be market incentives to drive the supply side. That is why people work so hard to make it happen.

Here is the statistic: Since TSMC was founded in 1987, the world semiconductor market has grown from $26 billion to $527 billion last year.

They rode a ridiculous tailwind.

A ridiculous tailwind. As the industry reorganized away from the vertical integration of the Intel world, TSMC was able to build a trillion-dollar-value foundry.

The scale of the numbers is staggering. TSMC can spend $20 billion to build a building, and the stuff that comes out of it is so valuable that the $20 billion was a profitable investment within a few years—whatever the payback period is.

They know for sure that it is worthwhile to make that investment.

The whole thing comes down to this: Silicon has become incredibly valuable. Integrated circuits are the fabric of our world today.

For my first carve-out, I have a recommendation for anyone who is not an AAA member. I highly recommend it.

I had a spectacular AAA experience. I went to fill up the air in my tires before a road trip, and there was something wrong with the pump at my local gas station. I ended up draining the air in my tires to an unsafe level, so the car was not drivable.

I had my baby in the back seat, and my wife and I were trying to figure out what to do. I signed up for AAA while sitting in the gas-station parking lot. Within an hour or an hour and a half, they had a mobile tire inflator drive out and fill up the air in my tires.

It was a long weekend, a holiday weekend when other people weren’t working. They got us on our way and didn’t ruin the weekend. It cost around $100 or $150 to become a member, and the service itself was free for something as trivial as this. You get 3 service calls a year. It was a phenomenal experience.

My second carve-out is a YouTube channel called Defunctland.

You turned me on to this.

It is an entire YouTube channel about defunct theme parks. If you like Acquired and wish you had something similar that was visual and about history, intellectual property, and people trying crazy things, this is perfect.

Some of the craziest entrepreneurs and executives within companies decided to build theme parks. It is fun to see the weird old Nickelodeon hotels and Action Park in New Jersey, the wildly unsafe park from the 1960s, 1970s, and 1980s.

You could get lost for hours watching Defunctland.

Those were the days, when you could take unreasonable amounts of risk and nobody thought there was anything wrong with it.

My carve-out is a movie. Speaking of it being 2025, I finally watched Everything Everywhere All at Once for the first time. I can’t believe I hadn’t seen it before, but with 2 kids under 3 and a half, there isn’t a lot of time for movies.

It’s so good.

It is so good. I think this was your carve-out when it came out a couple of years ago. It is truly excellent, lived up to the hype, and deserved every award that it won.