Part IThe Story
On September 4, 1985, fourteen days after he landed in Taipei with unused American vacation days still on the books, Morris Chang was summoned to the office of Li Kwoh-ting. The minister skipped the pleasantries. The government wanted a plan, something that "could allow Taiwan to enter the semiconductor industry," and it wanted the plan from the man it had just installed as president of the Industrial Technology Research Institute, a state laboratory in Hsinchu on an island then better known for plastics, textiles, and cheap electronics assembly than for anything that required a clean room. Li had already offered Chang what amounted to a blank cheque. He now wanted to know what Chang intended to write on it.
Chang had his answer ready. For two weeks he had been working on the problem at a desk on Chongqing South Road. The street was named for the wartime capital where, as an eleven-year-old, he had sheltered from the Japanese with his family.
— Morris Chang, Autobiography, Vol. 2 (2024)Two weeks later (September 4), Minister Li Kwo-ting wanted to see me. He brought up the matter immediately. I had already spent nearly two weeks thinking about this problem, so at that time, I proposed the concept of a 'Common Wafer Fab.'
The phrase meant nothing to anyone in 1985, because the thing it described did not exist. In the semiconductor business as it was then constituted, you designed your own chips and you built them in your own factories. Intel did both. Texas Instruments, where Chang had spent a quarter century, did both. If you were a small firm with a clever circuit and no fab, you begged an integrated manufacturer for spare capacity and hoped it would not steal your idea. Chang's proposal was a factory that would design nothing, sell nothing under its own name, and compete with no one: a manufacturer of other people's ambitions. It would be very expensive, technically demanding, and commercially uncertain. It would also be built in a place with almost no semiconductor tradition by a man who had not lived in it for more than a few weeks of his adult life.
Four decades later the company that grew out of that conversation, Taiwan Semiconductor Manufacturing Company, sells more than $120 billion of wafers a year. It holds roughly 70 percent of the world's contract chipmaking market and makes the processors for Apple, Nvidia, Qualcomm, and Broadcom, which means it makes the substrate of nearly every argument currently being had about artificial intelligence. The Taiwanese call it the Sacred Mountain of Protection. The superstition is that a world unable to live without TSMC will not let anything happen to the island it sits on.
The paradox is in the biography. Before he turned eighteen, Morris Chang had lived in six cities, attended ten schools, and been uprooted by war three times. He was a refugee, then an immigrant, then an American citizen, then a returnee to a country that was not quite the one he had left. The most fixed object in the global economy, a cluster of factories so tied to one patch of coastal plain that Washington has spent tens of billions of dollars failing to replicate it, was built by a man who belonged nowhere for most of his life, and who came to believe, more firmly than almost anyone in his industry, that some things cannot be moved.
By the Numbers
The Common Wafer Fab, Forty Years On
$122.4BTSMC revenue, 2025
~70%Share of the global foundry market
55Chang's age when TSMC was incorporated in February 1987
$58MPhilips's founding investment, for a 27.6% stake
135xChang's estimate of Philips's return when it sold out in 2008
$265BTSMC's total planned U.S. investment as of July 2026
83,825Employees, 2024
Six Cities, Ten Schools, and a Father Who Said No
The father was a banker. Chang Wei-kuan, born in 1906, graduated from Kwang Hua University in Shanghai with a Western education and spent his career managing other people's money, first as the official in charge of finance for the Yin county government around Ningbo, later as a bank manager. His son was born in Ningbo on July 10, 1931. Six years later the Second Sino-Japanese War began and the family started moving: Nanjing, then Guangzhou, then British Hong Kong, where the boy spent most of his primary-school years between the ages of six and eleven. When the Japanese took Hong Kong in December 1941 the family went back to Shanghai and Ningbo for a few months, then made the long overland passage to Chongqing, the inland redoubt from which the Nationalist government ran its war.
After the war came Shanghai, and a stretch of adolescence that Chang later remembered as being spent mostly at cards and tennis. Then the civil war, and in 1948 Hong Kong again.
The boy wanted to write. He imagined himself a novelist or a journalist, and he read widely in English. His father, who had watched the family's fortunes swing with every change of regime, saw no future in it. "My father said to me, 'you will starve as a writer,'" Chang told an interviewer in 1998, half a century after the argument and still, apparently, able to quote it verbatim. He enrolled instead in a university banking program in 1948, the most obvious concession available to a banker's son, and was soon forced to abandon it when the war reached him again. In Hong Kong, father and son reached an agreement. There were no universities in the colony dedicated to science or technology, so the boy would go to America and study something useful.
An uncle, Chang Ssu-hou, a professor at Northeastern University in Boston, handled the applications and steered his nephew toward Harvard rather than MIT. "He knew MIT was a special school and thought I was an uncertain kid," Chang recalled decades later. "He felt I should spend a year figuring out what to do." In the summer of 1949 the uncertain kid arrived in Cambridge as the only Chinese student among Harvard College's 1,100 undergraduates. He finished his freshman year in the top ten percent of the class, having spent much of it indulging the literary passion his father had tried to talk him out of.
Then he did the math on his own life. "The only really serious… middle-class profession that a Chinese-American could pursue in the early Fifties was technical," he would later say. The sentence is as clear an account of the immigrant's calculus as you will find in the literature of American business. It contains no bitterness, only an accurate reading of where a young man with his face and his accent would be let in. He transferred to MIT as a sophomore and declared mechanical engineering.
At MIT he was poor and in a hurry. He won a scholarship and took jobs as a copy typist and as a research assistant to the professor Joseph Kaye. He carried the maximum course load the Institute allowed and stayed through the summers to take more, finishing his undergraduate degree in two years instead of three. He liked the economics and history courses as much as the fluid mechanics and heat transfer. "A liberal education is very helpful," he said much later. "All disciplines are related, and you can only start to connect things when you understand the basics of several disciplines." He met Christine Chen, a student at Boston University, and married her in 1952, when they were both twenty-two. A bachelor's degree came that year and a master's in 1953.
He was, by his own later description, "penniless."
The Failure That Became a Method
Here the story is supposed to turn upward, and it does, but only after a fall. Chang wanted a doctorate. He sat MIT's doctoral qualifying examination and failed it. He sat it again and failed again. He left the Institute in 1955 with a third degree, a master of engineering, in place of the PhD he had planned on, and with what he described sixty years later, addressing graduates at Asia University in Taichung, as an impulsive decision to go get a job.
That speech is one of the more revealing documents in the Chang record, partly because it is so unglamorous. He told the graduates that the missed opportunity had changed how he thought about learning. Before, he had treated it as a way to earn a degree. Afterward he treated it as a discipline he would keep for life, something to be done in a "disciplined, systemic and organized manner," after hours, harder than when he had been a student. Then he did something few commencement speakers would dare and gave the graduates a budget. A week has 168 hours. Work no more than 50 of them. Give 20 to 30 to family, friends, and hobbies. Spend 10 to 20 learning and seven or eight exercising. Rest with what is left. The man who later ran one of the most demanding manufacturing organizations on earth was, on the evidence of his own arithmetic, a fanatic about the evening hours.
The first job offer came from Ford Motor Company. He turned it down for a small semiconductor division of Sylvania Electric Products that was trying to make germanium transistors. His assignment was yield, the share of devices on a production line that actually work, which then as now separated chip companies that made money from those that did not. He knew essentially nothing about solid-state physics, so he taught it to himself, with help from a colleague, at night.
The word yield will come back in this story often. A yield problem is unglamorous. It is about contamination and temperature drift, and about the thousand small disciplines of a production floor that do not show up in a patent filing. It rewards patience, repetition, and the slow accumulation of tacit knowledge in a specific group of people standing in a specific room. Chang's first professional problem was the problem his life's work would turn on.
In 1958 he moved to Texas Instruments in Dallas, then rising fast in the young industry. Within three years he had become manager of an engineering section, and in 1961 TI did something unusual for an American corporation and paid to send him back to school. He went to Stanford, studied under John L. Moll, William E. Spicer, and Gerald Pearson, and in 1964, about two years after starting, received a PhD in electrical engineering for a dissertation titled "Optical absorption and photoluminescence of doped GaAs and (InxGa1-x)As." The man who had twice failed MIT's qualifying exam earned his doctorate in roughly the time it takes most students to choose an advisor. He became an American citizen in 1962, in the middle of it.
He returned to Texas, where he would spend the next two decades. The first volume of his autobiography, published in Chinese in 1998, ends in 1964, with the degree and the return. He chose to close the first half of his life at the moment the failure had been redeemed.
Pricing the Future at Texas Instruments
Two episodes from the TI years matter more than the promotions, though there were many promotions. Chang later explained his rise with characteristic flatness: in each job he had "accomplished something significant," which in his early years usually meant dramatically raising a production line's yield.
The first episode looks like a footnote. At TI, Chang worked on a four-transistor project whose manufacturing was done by IBM, one of the earliest instances in the industry of one company designing a part and another fabricating it. It was a curiosity at the time, an arrangement between two giants. Chang filed it away.
The second episode was a seminar. The founder of a young firm called the Boston Consulting Group asked to present research to TI's own founding chief executive, and the CEO asked Chang to sit in. The research concerned what BCG called the experience curve: unit costs fall predictably as cumulative production rises. A company that understood this, the argument ran, should not price its product at today's cost. It should price at the cost it will reach later, once volume has driven the learning, and use the low price to win the volume that makes the learning happen. Losses early would become dominance later. The CEO was persuaded and let BCG expand the research inside TI. BCG sent a protégé named Bill Bain to do the work. For two years Bain worked out of an office Chang had arranged near his own, Monday through Thursday. At the end of it Bain resigned and founded Bain & Company.
Chang had found his doctrine. At TI he became the leading advocate of pricing semiconductors "ahead of the cost curve," a practice that was controversial when he started pushing it and common across the industry by the time he had finished. TI's regular, aggressive price cuts on integrated circuits helped make it the era's largest and most profitable chipmaker. By the late 1970s Chang was group vice president responsible for TI's worldwide semiconductor business, and under his watch the company became the world's leading producer of integrated circuits.
"I was a serious student of learning curve," he said in 2025. Of all his self-descriptions, it may be the most precise.
The learning curve has a corollary that Chang did not fully absorb until the early 1980s, and when he did it reshaped his thinking. Learning happens somewhere. While still at TI he saw that the company's factory in Japan was achieving roughly twice the chip yield of its factory in Texas, with the same company, the same processes, and the same products. The difference was the people. Japanese staff and technicians were better qualified and stayed longer, and when Chang tried to recruit people of the same caliber in the United States he could not find them. He concluded that the future of advanced manufacturing was likely to be in Asia. At the time this sounded like heresy. It later became the premise of his career.
By then TI's leadership had turned its attention to calculators, digital watches, and home computers, consumer businesses Chang regarded as a distraction from the semiconductor core. He felt his career had hit a dead end. When the company wanted to move him into its computer products operation, he declined. He later said he had felt "put out to pasture." He was sure he would keep a job at TI. He did not want the job TI had in mind. In 1983, a man in his early fifties with a daughter in school and a marriage that had quietly ended (he and Christine separated at the end of 1981 but did not divorce until 1991, shortly before their daughter graduated from college), he walked out of the company that had made him without another offer in hand.
He landed as president and chief operating officer of General Instrument Corporation in New York, a job he held for a little more than a year. Then the government of the Republic of China came calling.
A Factory That Makes Nothing of Its Own
The recruitment came from the top. Sun Yun-suan, the premier of a government that had lost the mainland and now ruled from Taipei, asked Chang to become chairman and president of ITRI. The job paid less than General Instrument. Chang took it anyway, and his explanation, given in 2007 to the industry association SEMI, belongs among the cleaner statements of motive in modern business: he went because the opportunity "was so new, and appeared to be so challenging."
"You need to follow your interests, not where you think the big money is," he said in the same interview. "Back in 1985, the big money was thought to be in financial venture capital, or maybe even continuing to manage a company in the U.S., but I felt that the Taiwan opportunity appealed to me from the interest point of view." Then he added the line that has since been quoted on a thousand motivational slides, which in context reads more like a wry shrug: "As it turned out, when you don't chase money, money comes to you."
He was supposed to stay a few years. He arrived on August 21, 1985, three decades after the civil war had scattered him, and fourteen days later he was in Li Kwoh-ting's office describing a factory nobody had built.
Getting it built meant finding partners, and at first there were none. Chang approached Texas Instruments, his old employer, and Intel. Both turned him down. The economics were brutal: enormous capital, uncertain demand, and a location with no track record. Only Philips, the Dutch electronics conglomerate, would sign. It put up $58 million, transferred its production technology, and licensed its intellectual property in exchange for 27.6 percent of the new company. The Executive Yuan's National Development Fund supplied another 48.3 percent of the startup capital. In his autobiography Chang put the total share capital at $145 million.
The remaining fifth or so came from eight private companies controlled by some of Taiwan's wealthiest families, firms in plastics, textiles, and chemicals: Formosa Plastics with 5 percent, CAPCO with 5 percent, Taiyuan Textile with 1 percent, and others. These investors had been, in the delicate phrasing that appears in more than one account, "asked" by the government to participate. The last 4.1 percent came from the investment arm of the ruling Kuomintang. From its first day TSMC was less a startup in the Silicon Valley sense than an instrument of state industrial policy, capitalized by a plastics fortune and a political party and technically underwritten by a Dutch conglomerate that, as it turned out, was slowly getting out of the business of making its own chips.
The company was incorporated on February 21, 1987. Chang was fifty-five. He was not, at first, its chief executive. That job went to James E. Dykes, a Philips North America veteran supplied by the Dutch partner, who left after about a year. Chang took the role himself and kept it, through one interruption, until 2013.
The early customer list did not look like the future. The founding thesis rested on a fabless design industry that barely existed in 1987, so TSMC's first customers were integrated manufacturers like Intel offloading overflow production to a subcontractor. Asked in 2025 whether the original plan had been to build two fabs and stop, Chang corrected the premise: "It was only the very initial plan." He was a student of the learning curve, and the learning curve rewards scale and punishes hesitation. "I would never stop at just two."
He kept building. Philips held its stake for twenty-one years and sold the last of it in August 2008. Chang estimated it recovered about 135 times its capital, and the Dutch company used the proceeds in its long pivot toward healthcare.
In 1989 a young engineer named Rick Tsai came back from the United States and became deputy director of TSMC's second fab, a six-inch plant. When the company built its first eight-inch fab in 1992 and quickly followed with a fourth, Tsai brought the more advanced plants in on schedule. Another returnee, Mark Liu, ran Fab 4. The people who built TSMC had something in common with the man who founded it: they had gone abroad, learned the trade in other people's factories, and come home to a place that offered them more than prestige. Chang recruited a cadre of senior engineers from the United States, most of Taiwanese descent, to get the company going. Taiwan's first chip generation was a diaspora that came home.
The company listed on the Taiwan Stock Exchange in 1993. In 1997 it became the first Taiwanese company listed on the New York Stock Exchange. By then it had passed a billion dollars in annual revenue, and its founder had laid down the rule that would define it: TSMC would never design and sell its own chips. It would never compete with a customer. Every design that came through the door could be trusted to remain the property of whoever sent it.
— Morris Chang, Computer History Museum Fellow profile, 2007In my career, I have done best by following my interest, rather than going where the big money was supposed to be.
Morris Chang Is Calling
In 1997 a nearly bankrupt graphics company in Northern California sent a letter to TSMC's sales office in San Jose asking about becoming a customer. Nobody answered.
The company was Nvidia. It had perhaps fifty or sixty employees and, by one account, was down to about a month of payroll. Its co-founder and chief executive was a thirty-four-year-old Taiwanese-born engineer named Jensen Huang, who had come to the United States as a child, washed dishes in a Denny's (a detail Huang himself has made famous), and now ran a company whose survival depended on finding a manufacturer for a chip it could not afford to fabricate. When the sales office ignored him, Huang wrote directly to Morris Chang in Taiwan.
Chang was irritated, though not with Huang. "I told our salespeople we should never be negligent in talking to future customers, even if they're small," he recalled in 2025. He was flying to California the following week anyway, so he picked up the phone. There was a racket on the other end, people shouting, an argument in progress. Then Chang introduced himself.
"He immediately shouted at those people that were making noises," Chang remembered, more than a quarter century later and still enjoying it. "He said, quiet. 'Morris Chang is calling me.'"
The two men met. Huang made a promise that a sober observer would have filed under bravado: the chip he wanted made would not just save Nvidia, it would make Nvidia a major TSMC customer. Chang was skeptical. TSMC had a few thousand employees and more than a billion dollars in revenue, and by his reckoning a "major" customer meant at least $50 million a year. Within two or three years Nvidia was there.
The phone call has become a parable in Silicon Valley, the founder who returned the small customer's call and was repaid many times over. That is true as far as it goes. What the parable leaves out is that Chang's attention to small customers was structural, not sentimental. A pure-play foundry with no products of its own lives entirely on other people's ideas, and it cannot know in advance which of those ideas will grow. The integrated manufacturers TSMC had begun with were a temporary market. The fabless startups (shoestring outfits with a few dozen engineers and a design they could not build) were the market Chang had bet the company on in Li Kwoh-ting's office in 1985. Ignoring Huang's letter would have meant ignoring the thesis.
Huang and Chang became friends of a sort uncommon between supplier and customer. On October 17, 2007, Huang conducted a long public interview with Chang at the Computer History Museum in Mountain View. For seventeen years it was the last long-form English-language video interview Chang gave. The relationship would nearly break two years later. It would also, eventually, lead Chang to offer Huang his own job.
The Founder Comes Back
In 2005, at seventy-three, Chang handed the chief executive's role to Rick Tsai, the engineer who had come home in 1989 to run Fab 2. Chang stayed on as chairman. The succession looked orderly.
Then came 2008. The financial crisis hit the semiconductor industry hard and fast, as it always hits a business whose costs are mostly fixed and whose demand is mostly derivative. TSMC's revenue fell, and the company did something it had not done before: it laid off employees on the basis of performance reviews, some 700 people by Chang's later account. The cuts went badly. Laid-off workers protested. And in the middle of the downturn, TSMC's new 40-nanometer process was struggling with yields, the old Sylvania problem returning on a far larger scale, and the customer hurt worst was Nvidia.
In June 2009 Chang took the chief executive's job back. He was seventy-seven.
His view of the layoffs is worth hearing in full, because it shows a mind that treats labor as an asset with a learning curve of its own. Recalling an earlier round of cuts at Texas Instruments, he said he had been the lone voice in the room against using performance reviews to decide who went. "People would not respect us if we lay off by performance," he said. "It's very subjective." And then the arithmetic: "The separation expense is usually half a year, and it takes at least half a year to train a person, so if you need people back within a year, you shouldn't lay off." In a business where Moore's Law guarantees that next year will need more engineers than this year, layoffs are usually a way of paying twice for the same knowledge.
The Nvidia problem needed a different kind of arithmetic. The 40-nanometer yields had cost Huang's company badly, and the dispute threatened the relationship Chang had personally started a dozen years earlier. Chang ran the numbers himself and arrived at a settlement figure of more than $100 million, serious money for a contract manufacturer at the end of 2009. He presented it to Huang over dinner with a condition attached. There would be no negotiation. Huang had forty-eight hours to accept, or the matter would go to arbitration. He accepted within the window.
The story has a coda that complicates its moral, and it comes from the trade. Daniel Nenni, founder of the industry site SemiWiki, who worked on a different company's 40-nanometer graphics chip, has argued that the yield failure was not really TSMC's fault. TSMC's design kit at that node offered two sets of rules: a minimum set that maximized performance, and a more conservative "design for manufacturing" set that sacrificed some performance for reliability. Nenni's team chose the conservative rules and their chip yielded. Nvidia, in his telling, chose the aggressive ones and did not. "TSMC took a horrible bashing over this," Nenni wrote, and at 28 nanometers TSMC offered only one set of rules: "lesson learned." He still called Chang's settlement "a diving catch." If Nenni is right, Chang paid more than $100 million to settle a dispute TSMC might have won. He was not paying for fault. He was paying for the next twenty years of the relationship.
The second thing Chang did on his return was fix TSMC's research spending at roughly 8 percent of revenue, every year, regardless of conditions. This ended the annual budget fight, which in a downturn always ends with the research labs giving ground. It let the R&D organization plan years ahead, which is the timescale on which process nodes are won and lost.
The third thing was the bet. TSMC's researchers told him the coming 28-nanometer node would be a "sweet spot" between performance and cost, the kind of process generation customers stay on for years. Chang decided to build for it at a scale that alarmed his board. TSMC roughly tripled its capital spending to about $6 billion. The company also announced plans to raise R&D spending by almost 39 percent, to NT$50 billion, in 2011 and to expand capacity by 30 percent. When the board pushed back, Chang reached, as an old literature student might, for Shakespeare. He quoted Brutus in Julius Caesar:
There is a tide in the affairs of men, which taken at the flood, leads on to fortune… On such a full sea are we now afloat, and we must take the current when it serves, or lose our ventures.
He did not know, he later admitted, exactly where the demand would come from. Asked in 2009 about the next wave, TSMC was talking about solar panels and LEDs, neither of which panned out. The demand came from smartphones and, later, data centers, markets Chang held no strong opinion about. What he did know was the physics: the nodes ahead would need new transistor structures, new materials, and new lithography, and whoever mastered them first would become the indispensable manufacturer. He could be wrong about the market and right about the factory.
He was also helped by his competitors' mistakes. At 28 nanometers an alliance of IBM, Samsung, and Chartered Semiconductor (later folded into GlobalFoundries) pursued a "gate-first" approach to a critical new transistor structure and struggled to yield. TSMC followed Intel's "gate-last" approach, and it worked. In Nenni's blunt summary, that was "the end of the Common Platform Alliance." TSMC won the node by a wide margin. Its first big node win came in the same years the smartphone remade the industry, and it was about to matter enormously, because a California company wanted to stop buying its most important chip from a rival.
Dinner at the Chang House
In 2010 Terry Gou came to dinner at Morris Chang's home in Taipei and brought a guest.
Gou was the founder of Foxconn, the largest contract manufacturer of consumer electronics in the world. He had built his fortune in the same trade TSMC practiced, making things other companies designed and branded, only at the assembly end of the chain rather than the silicon end. He was also family. Chang had married Sophie Chang in 2001, and Gou was her cousin. ("He is second cousin of Sophie's, Sophie's my wife," Chang explained on a podcast in 2025, in the tone of a man noting the weather.) Gou's guest was Jeff Williams, Apple's head of operations. Apple wanted TSMC to make the processor for future iPhones.
What followed was one of the defining commercial negotiations of the smartphone era, and Chang's telling of it has the texture of a long, careful, sometimes absurd courtship. Williams proposed that instead of the 28-nanometer process TSMC was then preparing to ramp, Apple's chips should go on a 20-nanometer node, a detour that would require TSMC to build substantial dedicated capacity. Apple named its price. It would allow TSMC a 40 percent gross margin, a figure below what TSMC was already earning. "Apple 'allowed' TSMC to have a 40% wafer margin," Nenni later wrote, with sardonic quotation marks around the verb.
Chang accepted. The capacity Apple needed was enormous, and TSMC had to finance it. At one point Chang explained to the Apple side that the company would have to issue corporate bonds to raise the money. Apple's reply, as Chang recounts it: "I think you can eliminate your dividend."
He did not eliminate the dividend. He issued the bonds.
Then Apple went quiet. For two months Williams paused the talks. Intel's chief executive had approached Tim Cook, and Apple was considering whether the world's most famous chipmaker, with its superior process technology and vast fabs, might be the safer bet. In the end it was not, and the reason Cook later gave Chang lies at the heart of the foundry business. "Intel just does not know how to be a foundry," Cook told him. Intel was a company built to design and sell its own chips, with a culture formed by decades of being the best in the room. A foundry has to be something humbler.
Chang described what he thought the difference was in a sentence that sounds like customer-service boilerplate until you consider how much of the company is in it. "We have learned to respond to every request," he said. "Some of them were crazy, some of them were irrational — but we responded to each request courteously. Which we do."
Even after TSMC won the business, the relationship had its shocks. Apple placed its first orders on the next-generation 16-nanometer process not with TSMC but with Samsung, Chang's most feared competitor. In his memoir and in conversation he described the news as a genuine blow. Williams flew to Taiwan to reassure him in person: when TSMC's 16-nanometer process was ready, all of that business would come to TSMC. It did. Trial production of Apple chips at TSMC was reported in 2011. By 2014 Apple was sourcing its A8 processors there and had become TSMC's most important customer. In time TSMC became Apple's sole foundry. Today, according to Wired, a single TSMC plant, Fab 18 in Tainan, etches a quintillion transistors for Apple every six months.
In March 2017 TSMC's market capitalization passed Intel's for the first time: $168.4 billion against $165.7 billion. In 2018 TSMC's 7-nanometer process entered volume production and the company took the lead in the most advanced manufacturing technology on earth from the firm that had invented the microprocessor. Asked at the launch of his second memoir in December 2024 about Intel's troubles, Chang offered a cool diagnosis: the company's board needed a core strategy before it needed a new chief executive. He said Intel's earlier chief executives had been his friends. Pat Gelsinger, who had tried to rebuild Intel as a foundry, had not been. He had seemed, Chang said, "hostile."
The deal that remade the company began at a family dinner, brokered by a cousin by marriage who had built the world's largest assembly business on the same proposition TSMC was built on: we will make your thing, and we will not compete with you for the credit.
Learning Is Local
In 2013 Chang tried to hire Jensen Huang.
He offered Huang the chief executive's job at TSMC, by Chang's account twice. Chang praised Huang's integrity, vision, and professionalism. Huang declined. He had a company of his own to run, and Chang later noted that Huang's 7.5 percent stake in Nvidia would have been a problem, since TSMC would have required him to sell it to avoid any bias toward one customer. Chang, by his own account, held no major stake in any TSMC customer. At the time Nvidia was worth about $9 billion. Within about a decade it had passed $3 trillion.
So in 2013 the job went to two insiders as co-chief executives: C. C. Wei and Mark Liu, the former Fab 4 manager from the early 1990s. On June 5, 2018, at eighty-six, Chang announced his retirement. Liu became chairman and Wei chief executive. Chang had led the company for thirty-one years. That September President Tsai Ing-wen awarded him the Order of Propitious Clouds.
Retirement did not mean silence, and it did not mean he would stay out of politics. He had already represented President Chen Shui-bian at the 2006 APEC summit in Hanoi. Tsai sent him as her envoy six times between 2018 and 2023, each time under the name diplomacy permits Taiwan at such gatherings: "Chinese Taipei." The man who had been a refugee from the mainland, then an American citizen, then the founder of the island's most valuable company, now represented that island at summits where it could not use its own name. In 2024 Tsai awarded him the Order of Dr. Sun Yat-sen.
In April 2021, at ninety, he gave an hour-long speech in Mandarin to Taiwanese government and business leaders, with Liu and Wei in the audience, titled "Cherishing the Advantages of Taiwan's Semiconductor Wafer Manufacturing." He opened by stating his position precisely. "My status is a TSMC retiree, and I no longer have any authority inside the company, which is my choosing." Then he laid out an argument he would repeat for years, with growing force, as the United States, Japan, and Europe began spending tens of billions of dollars to pull chipmaking back onto their own soil.
The argument came in parts. Taiwan had technical talent, engineers and technicians who wanted to work in manufacturing. Taiwan had professional managers, though Chang warned they would not necessarily perform as well abroad. Taiwan had geographic concentration, an entire supply chain within a few hours' drive. And Taiwan had low turnover. TSMC's employee turnover ran around 3 to 4 percent a year, and in the mid-2010s, during Chang's second stint as chief executive, the company had run a round-the-clock research operation known internally as the Nightingale program, which an observer called "unthinkable in the US."
He made the case most vividly at MIT on October 24, 2023, before a capacity crowd of more than 425 people in Room 10-250, a few hundred yards from where he had once worked as a copy typist. The learning curve, he said, the doctrine he had absorbed beside Bill Bain in Dallas, has a geography. "It works, the learning curve, the experience curve, it works only when you have a common location," he said. "Learning is local."
He had come to this view through his own losses. In 1996 TSMC and three American partners put $1.2 billion into a fab in Camas, Washington, about twenty miles outside Portland. It was then the largest startup investment in the state's history, and it began production in July 1998 with a 0.35-micron part for Altera. TSMC bought out its partners in 2000. The company had assumed the plant's costs would be comparable to Taiwan's. That assumption, Chang told the Brookings Institution in April 2022, had been "naive." TSMC changed managers and engineers and improved the plant's performance, and still its chips cost about 50 percent more than the same chips made in Taiwan.
So when Washington began spending heavily to bring chipmaking home, Chang was not polite about it. "I think it will be a very expensive exercise in futility," he said. America still led the world in chip design, he noted, but after the 1970s its best technical talent had drifted into finance and consulting. "I don't really think it is a bad thing for the US actually," he added. "But, it's a bad thing for trying to do semi manufacturing in the US." He said the decision to build TSMC's $12 billion fab in Arizona had been made by Mark Liu after he retired, at the insistence of the American government. And to the question everyone asks about Taiwan, he gave an answer that is either an old man's stoicism or a refugee's realism: "If there is a war, we all have a lot more than just chips to worry about."
— Morris Chang, Brookings Institution interview, April 2022If there is no war, then I think the efforts to increase onshore manufacturing of semiconductors is a wasteful and expensive exercise in futility. If there is a war, we all have a lot more than just chips to worry about.
The company he no longer ran built in Arizona anyway. Construction there cost four to five times what an identical plant would cost in Taiwan, and TSMC warned it would need to bring in Taiwanese workers because American ones with the right skills were too scarce. The Biden administration awarded it $6.6 billion under the CHIPS Act. On March 3, 2025, at the White House, Donald Trump announced another $100 billion in TSMC investment and said of C. C. Wei, standing beside him: "He is the most important man in this room." By July 2026 TSMC's planned American investment had reached $265 billion, for a projected twelve fabrication and advanced-packaging facilities in the Arizona desert. Fabs have also gone up in Kumamoto, Japan, and in Dresden, where the building cannot house extreme ultraviolet lithography machines because they are too tall for it. In 2017, over the objection of Tsai's own government, TSMC had invested $3 billion in a plant in Nanjing, a short distance from the city the Chang family fled in the 1930s.
Chang had said it was futile. Whether he was proven right or wrong by the money that followed, the arc of his career had brought him here. The engineer who had watched TI's Japanese fab double the yield of its Texas fab, and who had concluded in the early 1980s that the future of manufacturing lay in Asia, now watched the world's most powerful government pay many billions of dollars to bring that future back, and he told it the price would be too high. "Frankly," he said at MIT, "the advantages that Taiwan enjoys today … were enjoyed by the U.S. in the '50s and '60s." Then he suggested that decades from now the advantage might move again, to India, Vietnam, or Indonesia. Learning is local. Places change.
The Iceberg
The writer his father had forbidden finally got his book.
The second volume of Morris Chang's Autobiography was published in Taipei on November 29, 2024, twenty-six years after the first. Chang began it after retiring in 2018, and the pandemic and his health slowed it. At ninety-three he finished a manuscript of nearly 200,000 words covering fifty-four years, from 1964 to 2018. The two volumes together run close to 300,000 words. The book went through three printings before its release. "It's done, and I feel relieved!" he told Global Views Monthly in the Taipei office he shares with his wife. He would write no sequel.
He told interviewers he admired Hemingway, and the iceberg theory in particular: seven-eighths of meaning lies beneath the surface, and simple words carry the most weight. "This book is my autobiography, not TSMC's history," he wrote in the preface, a sentence that is either modest or quietly territorial. At the Taipei launch on December 9, 2024, reporters asked him about the turmoil around his company: tariffs, export controls, a world in which, as he had recently put it, "globalization is dead" and "TSMC has become a battleground." He answered, more than once, "I'm not involved anymore."
For a while nobody in the West could read the book. It was published only in traditional Chinese. At the end of 2024 Brian Potter of Construction Physics bought both volumes from a Taiwanese bookstore, ran them through his browser's built-in translator, and published a long review calling the absence of an English edition "extremely stupid." The machine handled the prose tolerably and the names badly. One American surname from Chang's Texas years came out variously as Hegdi, Haig, Haigidi, Haggard, Hagdi, and Hagrid. Karina Bao, with funding from the economist Tyler Cowen's Emergent Ventures, began a proper translation. The Acquired podcast hosts used an unpublished draft of it to prepare for their January 2025 interview in Chang's Taipei office, his first long-form English video interview since the conversation with Jensen Huang in 2007. Huang had helped arrange it.
Reporters who saw him in those years noticed small things. A Wall Street Journal columnist who reached him by video found him at his desk just after lunch in Taipei, at one in the morning New York time, sipping a Diet Coke, ready to talk for ninety minutes. Chris Miller, author of Chip War, described him as a pipe smoker who "doesn't fit the image of a hoodie-wearing Silicon Valley tech bro." MIT Technology Review noted that he and Sophie live in Taiwan, and that he keeps a lifelong interest in music "from the era of Bach to Mahler." In 2009, the year he took back the CEO's job, he voiced the role of Master Dragon in the first episode of Let's Go Guang!, an animated program teaching Chinese to children. It is the only acting credit in a résumé otherwise made of fabs.
In January 2026 the IEEE announced that its Medal of Honor for that year would go to Jensen Huang. Chang had received the same medal in 2011. On January 29, 2026, Huang landed in Taiwan and went straight to Chang's residence in Taipei's Dazhi district. That evening the two men were photographed at dinner in a well-known local restaurant. Chang was ninety-four, in a wheelchair, his hair white, the high cheekbones unchanged. Nvidia was worth close to $5 trillion. TSMC was the first company in the world to mass-produce 2-nanometer chips.
Some of those chips were being made in Hsinchu, where the company's newest fabs climb into the hills around the science park. To make room for one of them, in the Baoshan area of Hsinchu, workers were relocating the ancestral grave of a Zheng family. It had been built in 1844 and had stayed in that ground for 181 years.
Part IIThe Playbook
What follows are twelve principles drawn from Morris Chang's career. They are not generic virtues. Each comes from a specific decision, sometimes a costly one, and each runs against the received wisdom of the era in which he made it. Chang's methods were usually counterintuitive at the time and obvious only afterward. The aim here is to describe them precisely enough to be used.
The principles
- 1.Build for the customer who doesn't exist yet.
- 2.Never compete with the people who feed you.
- 3.Price at the cost you will reach, not the cost you have.
- 4.Treat yield as the business.
- 5.Learning is local, so concentrate.
- 6.Make R&D a constant, not a negotiation.
- 7.Bet on the physics, not the forecast.
- 8.Return the small customer's call personally.
- 9.Settle fast, settle generously, set the clock.
- 10.Don't lay off by performance review.
- 11.Trade margin for the anchor customer, then answer every request.
- 12.Let expertise compound before you found.
Principle 1
Build for the customer who doesn't exist yet
When Chang proposed the "Common Wafer Fab" in September 1985, the fabless chip-design industry it was meant to serve was barely a rumor. TSMC's first customers were integrated manufacturers like Intel dumping overflow production, the least strategic business in the industry. A founder who listened to the market as it stood would have built a subcontractor. Chang built the infrastructure for a market he was confident would appear once the infrastructure existed. The design companies that later came to define the industry, Nvidia, Qualcomm, Broadcom, and Apple's silicon team, were made possible by a factory that predated them.
The discipline here is not faith. It is structural reasoning. Chang knew from the IBM four-transistor project at TI that design and fabrication could be separated. He knew from the learning curve that a factory serving many customers would accumulate experience faster than one serving a single company's product line. And he knew design talent was cheaper to assemble than a fab. If those three things were true, the fabless industry was a matter of when, not whether.
The risk is being early for a long time. Chang covered it by taking whatever revenue the present offered, including IDM overflow, while keeping the company's structure aimed at the future. He did not refuse the old customers. He just did not build around them.
Tactic: Identify the cost barrier that is keeping a class of customers from existing, remove it, and fund the wait with whatever business the current market will give you.
Principle 2
Never compete with the people who feed you
TSMC's founding rule (no proprietary chips, ever) looks like a limitation and works as a moat. A fabless company handing its crown-jewel design to a manufacturer is making an act of trust, and that trust is impossible if the manufacturer also sells chips. Chang understood that his product was not wafers. It was neutrality.
The proof came when Apple briefly weighed Intel against TSMC. Intel had formidable technology and the most prestigious name in the industry. It lost the iPhone because, in Tim Cook's words as Chang recounts them, "Intel just does not know how to be a foundry." Decades of designing and selling its own products had shaped Intel's culture, its priorities, and in Chang's telling its attitude toward partners. A foundry has to be humble by design. An integrated manufacturer cannot fake humility on demand.
Chang applied the same rule to people. When he tried to recruit Jensen Huang as TSMC's chief executive in 2013, he noted that Huang's large Nvidia stake would have had to be sold to avoid bias toward one customer, and he made a point of saying he held no major stake in any TSMC customer himself. Neutrality was a governance principle as well as a business one.
⚖
Integrated vs. Pure-Play
Why neutrality beat technical prestige in the foundry business.
| Dimension | Integrated manufacturer (Intel model) | Pure-play foundry (TSMC model) |
|---|---|---|
| Primary customer | Its own product lines | Everyone else's product lines |
| Customer trust | Compromised by competition | Structural: no competing products |
| Learning-curve volume | Limited to own designs | Aggregated across hundreds of customers |
| Cultural posture | "Best in the room" | "Respond to every request courteously" |
| Outcome by 2017–2018 | Overtaken in market cap and leading-edge process | Most advanced manufacturer on earth |
Tactic: If your business depends on customers trusting you with their most valuable assets, give up every adjacent line of business that would make you their competitor, and say so publicly and permanently.
Principle 3
Price at the cost you will reach, not the cost you have
The experience-curve doctrine Chang absorbed at Texas Instruments, in a seminar from BCG's founder and two years working beside Bill Bain, holds that unit costs fall predictably with cumulative volume. Chang pushed its most aggressive implication: price products at the cost you expect to hit after the learning, not at today's cost. Accept early losses, win the volume, ride the curve down. At TI this became a policy of regular, aggressive price cuts that helped make the company the era's dominant chipmaker and was eventually copied across the industry.
The point is not cheapness. It is converting time into an advantage that competitors cannot match without making the same bet. A rival who prices at current cost is stuck at low volume and high cost. A rival who matches your forward price without your volume loses money. Either way, the company that commits first sets the terms.
Chang later carried the same reasoning to TSMC's scale. "I was a serious student of learning curve," he said, "and I would never stop at just two" fabs. Capacity is how you buy position on the curve.
Tactic: Model where your unit costs will be at ten times your current volume, price close to that number, and treat the early margin you give up as the price of getting there before anyone else.
Principle 4
Treat yield as the business
Chang's first assignment at Sylvania was germanium transistor yield. The observation that sent him to Asia was that TI's Japanese fab yielded twice what its Texas fab did. The crisis that brought him back as CEO in 2009 was a yield failure at 40 nanometers. The lesson he drew at 28 nanometers, where TSMC moved to a single set of design rules, was about yield. In his career yield is never a metric reported to management. It is the business itself, and the place where engineering discipline, labor quality, and customer trust show up as a number.
The managerial point is that yield problems cannot be solved from a distance. They live in the accumulated tacit knowledge of a specific workforce on a specific line. Chang rose at TI, by his own account, by "accomplishing something significant" in each job, which in his early years usually meant raising yields. He earned authority by fixing the most physical problem in the building.
Tactic: Find the single operational metric where your engineering, people, and customer outcomes converge, and make sure your most senior people have personally owned it.
Principle 5
Learning is local, so concentrate
Chang's most important late-career insight extends the learning curve: it has a geography. "It works only when you have a common location," he told MIT in 2023. "Learning is local." Knowledge that cuts costs lives in people who work near each other, under shared conditions, with low turnover. Spread them out and the curve flattens.
Chang had expensive evidence. The Camas, Washington, fab, built on the assumption that its costs would match Taiwan's, never closed the gap. Chips made there cost roughly 50 percent more, even after years of changing managers and engineers. Taiwan's advantages, as he listed them in 2021, were all about concentration: dedicated technicians from trade schools, local professional managers, a dense supplier network, and turnover of around 3 to 4 percent. These advantages compound, and they are hard to ship.
The flip side, which Chang stated plainly, is that concentration is a historical condition, not a permanent fact. The United States held these advantages in the 1950s and 1960s. India, Vietnam, or Indonesia might hold them later. Learning is local, and so is forgetting.
Tactic: Before distributing a high-skill operation across sites for cost or politics, estimate the learning-curve penalty of fragmentation, and if you must expand, seed the new site with people who carry the tacit knowledge in their heads.
Principle 6
Make R&D a constant, not a negotiation
When Chang retook the CEO's job in 2009, one of his first moves was to fix research spending at about 8 percent of revenue, permanently, in good years and bad. It ended the annual budget fight, which in a capital-intensive cyclical business tends to end the same way every downturn: research gets cut, and three years later the company finds it has fallen a node behind.
A fixed ratio does two things. It lets researchers plan on the timescale their work actually requires, which in process technology means many years. And it removes a recurring political contest from the executive agenda, so leadership can spend its attention on capital allocation and customers. The cost is some flexibility. Chang judged that flexibility in R&D spending was mostly a polite word for underinvestment.
Tactic: Set your core R&D spend as a fixed share of revenue, take it out of the annual budget process entirely, and revisit the ratio only on a multi-year cadence.
Principle 7
Bet on the physics, not the forecast
The 28-nanometer decision is often told as visionary market timing: Chang tripled capital spending to about $6 billion, over his board's objections, just before the smartphone boom. The more accurate story is more useful. Chang had no strong view about where demand would come from, and TSMC's own forecasts were pointing at solar panels and LEDs. What he had was confidence in the technology: his researchers said 28 nanometers would be a long-lived "sweet spot," and he knew which technical hurdles any leader of the coming generations would have to clear.
He bet on the factory, not the application. If TSMC was the best place to make advanced chips, whatever the next big market turned out to be would come to TSMC. The choice of gate-last over gate-first at 28 nanometers, while the IBM-Samsung-Chartered alliance stumbled, was exactly the kind of bet that pays off regardless of which end market wins.
◆
The Three Big Bets
Chang's largest commitments, classified by what they actually depended on.
| Decision | What the bet depended on | Type |
|---|---|---|
| Common Wafer Fab (1985–87) | Separability of design and fabrication; learning-curve economics | Structural |
| 28nm capex tripled to ~$6B (2010) | Process "sweet spot"; gate-last technical choice | Physics |
| Apple at 40% margin, bond-financed (2010–14) | Anchor customer volume driving the learning curve | Scale |
| Market forecast in 2009 (solar, LED) | End-market prediction | Wrong |
Tactic: When you can't predict which market will win, invest in the capability every plausible winner will need, and judge the bet by technical certainty rather than demand forecasts.
Principle 8
Return the small customer's call personally
In 1997, with TSMC past a billion dollars in revenue, Chang personally called back the chief executive of a fifty-person graphics company whose letter his sales office had ignored. He was annoyed at his own salespeople: "we should never be negligent in talking to future customers, even if they're small." Nvidia became a major customer within two or three years and eventually one of the most valuable companies on earth.
What matters is how this works. A pure-play foundry's growth depends entirely on which of its customers grow, and that cannot be known in advance. The cost of a CEO's phone call is trivial. The cost of missing the next Nvidia is incalculable. Chang's personal involvement also signaled something to Huang that a sales rep could not: the founder took him seriously. That is why, when Chang called, Huang told his shouting colleagues to be quiet.
Tactic: Set up a mechanism that brings unsolicited inquiries from small but ambitious prospects to senior leadership, and personally answer a few every quarter.
Principle 9
Settle fast, settle generously, set the clock
When 40-nanometer yield problems hurt Nvidia in 2009, Chang did not litigate fault, though a credible industry voice later argued TSMC was not chiefly to blame. He calculated a settlement himself, more than $100 million, offered it over dinner, and gave Huang forty-eight hours to accept or go to arbitration. Huang accepted.
There are three separate moves here. Doing the math personally made the number credible. Making the offer generous made it hard to refuse. The deadline ended the dispute before it could spread through the relationship. Chang was not buying absolution for the past. He was buying the next two decades of Nvidia's business, and the price was a rounding error against that.
Tactic: When a key relationship is damaged, have the most senior person calculate a fair settlement, present it in person, and attach a short deadline that turns a grievance into a decision.
Principle 10
Don't lay off by performance review
Chang's view of downturns is one of his least intuitive. At TI he was the only executive in the room who objected to choosing layoffs by performance rating. "People would not respect us if we lay off by performance," he said. "It's very subjective." When TSMC's 2008–2009 performance-based layoffs of some 700 people led to protests, he returned as chief executive.
His reasoning is mostly arithmetic. Separation costs about half a year's pay, and training a replacement takes at least half a year. If you will need the people back within a year, and in an industry driven by Moore's Law you almost always will, a layoff pays twice for the same knowledge. Beyond the arithmetic is the learning curve: a workforce with 3 to 4 percent turnover is accumulating experience that a churning one never will.
Tactic: Before any reduction in force, calculate the combined separation and retraining cost against your expected rehiring horizon, and if headcount will come back within a year, cut other costs instead.
Principle 11
Trade margin for the anchor customer, then answer every request
To win Apple, Chang accepted a 40 percent gross margin, below what TSMC was already earning, agreed to a 20-nanometer detour that required dedicated capacity, and financed the build with corporate bonds. (He ignored Apple's suggestion that he eliminate the dividend.) He absorbed a two-month pause while Apple courted Intel, and the shock of Apple placing its first 16-nanometer orders with Samsung. The return was the largest, most demanding, most reputation-building customer in consumer electronics, a volume engine that drove TSMC down the learning curve faster than any rival could follow.
Winning the anchor was half the job. Keeping it was the other half: "We have learned to respond to every request. Some of them were crazy, some of them were irrational — but we responded to each request courteously." A thin margin is tolerable when the customer's volume, demands, and prestige make you better at everything else.
◷
The Apple Courtship
A deal won through patience, concession, and service.
2010
Terry Gou brings Apple's Jeff Williams to dinner at Chang's home in Taipei.
2010
Apple proposes a 20nm detour and a 40% margin; TSMC accepts and issues bonds.
2010s
Apple pauses talks for two months after Intel approaches Tim Cook.
2011
TSMC reported to begin trial production of Apple processors.
2014
Apple sources A8 from TSMC and becomes its most important customer.
2017
TSMC's market cap surpasses Intel's for the first time.
Tactic: Accept below-target margins from one customer whose volume and standards will raise your capability across the business, and staff to say yes to that customer's unreasonable requests.
Principle 12
Let expertise compound before you found
Chang founded TSMC at fifty-five, after twenty-five years at TI, a brief run at General Instrument, and a quiet exit from a company that had put him "out to pasture." He did not succeed despite his age. He succeeded because of what the years had given him: the experience-curve doctrine from the BCG seminar, the IBM foundry precedent, the Japanese yield gap, the contacts that made Philips sign, and the credibility that made a government give him a blank cheque. Nobody without three decades inside the industry could have known exactly which of its rules to break.
His own advice about why he made the jump points the same way. He took the lower-paying job in Taiwan because it "was so new, and appeared to be so challenging." He followed interest, not money, at a stage of life when interest had been informed by decades of hard knowledge. The habits that compounded were built early: the 168-hour budget, the 10 to 20 hours a week of learning, the self-taught solid-state physics after his PhD plans collapsed.
Tactic: Treat your mid-career years as tuition for a later venture: keep a running list of the industry's broken assumptions you have personally watched fail, and act when one of them meets an institution willing to fund the fix.
Free playbook
Get The Business Model Playbook
58 business models, one visual page each: how the money flows, the metrics that matter, and who runs it. Free when you join the Faster Than Normal email.
Free. No spam. Unsubscribe anytime.
Part IIIQuotes and Maxims
In their words
— Morris Chang, MIT Manufacturing@MIT lecture, October 24, 2023It works, the learning curve, the experience curve, it works only when you have a common location. Learning is local.
— Morris Chang, SEMI oral history, 2007You need to follow your interests, not where you think the big money is… As it turned out, when you don't chase money, money comes to you.
— Morris Chang, Acquired, January 2025We have learned to respond to every request. Some of them were crazy, some of them were irrational — but we responded to each request courteously.
— Morris Chang, Brookings Institution interview, April 2022I think it will be a very expensive exercise in futility.
— Morris Chang, on the TI layoff debate, Acquired, January 2025People would not respect us if we lay off by performance. It's very subjective.
Maxims
- Sell neutrality, not wafers. A manufacturer that never competes with its customers earns a trust no amount of technical prestige can buy.
- Price the future. Set prices at the cost the learning curve will deliver and let the volume pay you back.
- Learning is local. Tacit knowledge compounds where people work near each other and stay, and it does not travel well.
- Fix the R&D ratio. A permanent percentage of revenue ends the annual fight that research always loses in a downturn.
- Bet on the physics. When the market is unknowable, invest in what every possible winner will need.
- Return the small call. The founder's attention costs almost nothing, and the next giant is often writing to a sales office that hasn't answered.
- Buy the next twenty years. In a damaged relationship, a generous settlement with a hard deadline is cheaper than being right.
- Don't pay twice for knowledge. If you will rehire within a year, a layoff costs more than it saves.
- Let the anchor customer squeeze you. A thin margin from the right customer makes you better at everything else.
- Follow interest, informed by decades. The best late-career bets come from expertise deep enough to know which rules to break.
