TSMC founder Morris Chang

Morris Chang’s relationship with Nvidia began with a letter from Jensen Huang in 1997. Huang wrote directly to the TSMC founder after Nvidia received no meaningful response from TSMC’s San Jose sales office. Nvidia had developed promising chips and urgently needed a manufacturing partner.

Nvidia was a small company facing possible bankruptcy when Huang contacted Chang. Founded approximately four years earlier, it had only 50–60 employees. TSMC, by contrast, employed several thousand people and had already exceeded $1 billion in annual revenue in 1995.

Chang was irritated that TSMC’s sales office had ignored Nvidia. He had instructed his salespeople never to neglect prospective customers merely because they appeared small. Huang’s letter therefore raised both his curiosity and concerns about his organization.

Chang personally called Huang during a trip to California the following week. Huang answered the telephone amid a noisy discussion with employees. Upon realizing who was calling, he immediately demanded silence and announced that Morris Chang was on the phone.

Their first meeting convinced Chang that Huang was unusually articulate, optimistic, and candid. Huang openly acknowledged Nvidia’s financial difficulties but predicted that its next successful chip would rescue the company and make Nvidia a major TSMC customer.

Huang’s promise represented an enormous commercial ambition. Chang estimated that becoming a major TSMC customer would require Nvidia to generate at least $50 million in annual manufacturing revenue for the foundry.

Nvidia’s gaming chip fulfilled Huang’s prediction. Chang could not remember the chip’s precise designation but confirmed that it was commercially successful, helped Nvidia avoid bankruptcy, and transformed its relationship with TSMC.

Nvidia became one of TSMC’s five largest customers within approximately two or three years. The early success established a manufacturing partnership in which Nvidia designed chips and TSMC fabricated them, a relationship that continued through the enormous expansion of both companies.

The Nvidia partnership encountered a serious crisis at TSMC’s 40-nanometer manufacturing node in 2009. Production delays, manufacturing yields, and quality disagreements harmed customers financially and operationally. Nvidia was among the largest users of the process and suffered particularly significant damage.

The 40-nanometer yield problem affected multiple TSMC customers, not only Nvidia. Yield refers to the proportion of manufactured chips that work successfully. TSMC needed to solve the underlying process difficulties for its broader business, regardless of its separate dispute with Huang.

TSMC’s leadership initially refused to compensate Nvidia for its quality complaints. The CEO at the time relied on the quality director’s argument that TSMC was not responsible. Consequently, Nvidia had received no financial settlement despite its losses.

Chang had previously surrendered the CEO position to a potential successor while retaining the chairmanship. He explained that Taiwan’s corporate governance structure often leaves the chairman as the most powerful person even when someone else serves as CEO.

Chang returned as TSMC CEO in 2009 because several serious problems were accumulating. The Nvidia dispute was important, but difficulties involving profitability, management decisions, and employee layoffs also contributed to his intervention.

TSMC’s gross margins were deteriorating because manufacturing costs were not falling as quickly as prices. The company had committed to customer price reductions but had failed to achieve matching production-cost improvements, progressively compressing its margins.

The immediate trigger for Chang’s return was the dismissal of approximately 600–700 employees. During the financial crisis, the incumbent CEO terminated employees who had received the lowest performance evaluations, treating the dismissals as performance management rather than conventional layoffs.

Chang had specifically instructed the CEO to seek board approval before conducting layoffs. He had offered to convene a special board meeting and intended to oppose the proposal. The CEO avoided that process by classifying the dismissals as punishment for poor performance.

Chang opposed using performance ratings to select people for dismissal because evaluations are inherently subjective. Hundreds of different supervisors had assigned the ratings used to determine which employees lost their jobs. He believed employees would not regard that system as fair or credible.

TSMC traditionally used probation and transfers rather than dismissals for poor performance. Employees with weak evaluations could receive six months of probation, after which many returned to their existing jobs and others moved into positions better suited to their abilities. Chang said outright termination was extremely rare under his leadership.

Chang had challenged performance-based layoffs decades earlier at Texas Instruments. During an early-1970s management discussion, he opposed the prevailing recommendation to dismiss the lowest-rated employees, arguing that the method lacked credibility because ratings depended on individual supervisors.

Chang argued that temporary layoffs frequently make poor financial sense in semiconductor manufacturing. Severance could cost approximately six months of compensation, while recruiting and training replacements could require another six months. If employees would be needed again within a year, dismissing them could be more expensive than retaining them.

Chang believed semiconductor companies needed to retain skilled workers through economic downturns. Advancing technology and Moore’s law continually created new requirements for experienced personnel. He regarded indiscriminate recession-driven layoffs as a short-term management reaction that could damage longer-term capabilities.

The 2008 financial crisis had significantly reduced TSMC’s business. Chang understood why an inexperienced CEO might instinctively reduce headcount, but he believed the semiconductor industry’s particular economics made that response inappropriate.

Dismissed TSMC employees protested outside Chang’s home while he was still chairman. More than 100 demonstrators appeared during one protest, prompting authorities to deploy approximately 50–60 police officers to maintain order. The demonstration disrupted neighborhood access.

A second protest brought approximately 25 employees to a park near Chang’s home overnight. The situation greatly distressed his wife, Sophie, who repeatedly checked on the demonstrators during the night.

Sophie Chang personally brought breakfast to the protesting workers the following morning. Around 6 a.m., she visited a neighborhood market with a bodyguard and purchased enough Chinese breakfast food for roughly 25–30 people. The protesters thanked her and abandoned a planned march to Taiwan’s presidential residence that day.

The employee protests reinforced Chang’s decision to retake executive control. After returning, he offered the dismissed employees an opportunity to rejoin the company, reversing the earlier approach to personnel management.

Chang did not dismiss or financially punish the CEO responsible for the layoffs. He retained the executive’s organizational grade, salary, and bonus, and repeatedly told him he remained a possible successor. Instead, Chang reassigned him as president of new businesses.

The reassigned executive was Rick Tsai, who later became CEO and vice chairman of MediaTek. His new responsibilities at TSMC centered on solar cells and LED businesses, both then regarded as promising opportunities.

TSMC’s solar-cell and LED diversification efforts ultimately disappointed. Chang attributed the solar business’s difficulties largely to aggressive Chinese subsidies that drove prices extremely low and helped Chinese manufacturers dominate the market.

Chang attributed the LED business’s failure partly to concentrated patent ownership. He said a small number of companies controlled important LED patents and were unwilling to provide the access needed by new competitors. He also considered the LED market smaller than the solar market.

Chang never expected solar or LEDs to replace TSMC’s core integrated-circuit business. He believed semiconductor manufacturing would continue growing, while the adjacent businesses represented additional opportunities. Rick Tsai eventually left after the new-business assignment failed to produce the hoped-for results.

Chang made personal calls to TSMC’s major customers immediately after resuming the CEO role. Nvidia and Qualcomm were among those contacted. Huang remained friendly but firmly reminded Chang that the 40-nanometer manufacturing and delivery problems remained unresolved.

Chang promised Huang that the Nvidia dispute would receive immediate attention. He requested approximately two weeks to investigate before returning with a response. Resolving the matter became one of his principal priorities.

Chang devoted nearly half his working time during his first four or five weeks back to the Nvidia problem. He interviewed salespeople and employees involved in the dispute, reviewed the financial damage, and examined the downstream demands Nvidia was receiving from its own customers.

Chang separated the manufacturing solution from the financial settlement. Improving 40-nanometer yields remained essential to TSMC and all customers, while compensating Nvidia required determining a fair amount for the losses it had already suffered.

Chang ultimately authorized a settlement offer exceeding $100 million. He said the figure followed weeks of investigation and reflected his judgment about what was fair to both companies.

Chang delivered the settlement proposal during a private dinner at Jensen Huang’s home in Silicon Valley. Their families had an established habit of informal meals together, typically involving pizza and salad, often prepared in part by Huang’s wife, Lori.

Chang deliberately separated the personal dinner from the business negotiation. He proposed eating from approximately 6:30 p.m. until 8 p.m., then moving to Huang’s study to discuss the dispute. The arrangement preserved their personal relationship while allowing a serious commercial discussion.

The settlement was presented as a final offer with a 48-hour deadline. Chang told Huang there would be no bargaining or argument over the amount. If Nvidia rejected the proposal, the companies would proceed to arbitration, which Huang had already suggested to TSMC’s previous CEO.

TSMC’s previous CEO had offered Nvidia effectively zero compensation. Chang’s offer therefore represented a fundamental change in the company’s response, replacing denial of liability with a substantial proposed settlement.

Huang accepted the settlement within the two-day window. Chang regarded the resolution as important enough to include in his autobiography. The companies subsequently conducted many billions of dollars in additional business together.

The 40-nanometer manufacturing problems were strategically important beyond Nvidia. Chang emphasized that TSMC needed to master the process before advancing successfully to 28 nanometers, its next major technological opportunity.

Chang’s experience at Texas Instruments shaped TSMC’s research-funding policy. While leading TI’s worldwide semiconductor business, he considered its R&D budget of 4.8% of revenue inadequate and repeatedly requested an increase to 5.5%. His requests were denied.

Chang wanted TSMC’s researchers to stop negotiating their budgets annually. Before his intervention, research spending generally ran around 6–7% of revenue, with annual discussions between the R&D director and CEO over the precise amount.

Chang established R&D spending at 8% of revenue, regardless of recessions. The policy gave the research organization predictable resources and reduced the risk that management would interrupt important technical programs during economic downturns.

TSMC’s R&D leadership regarded the fixed funding commitment as one of Chang’s most valuable decisions. Researchers could develop longer-term plans, hire personnel, and allocate resources without repeatedly fearing budget reductions.

TSMC’s researchers identified 28 nanometers as a technological “sweet spot.” They provided Chang with technical reasons why the process generation could be particularly important and argued that the company should accelerate development.

Chang combined technical forecasts with strategic market analysis before committing to 28 nanometers. He relied on R&D specialists, market forecasters, and the business-development organization rather than making the decision solely on engineering considerations.

TSMC dramatically increased capital expenditure to pursue the 28-nanometer opportunity. After spending roughly $2–2.5 billion annually on fabrication facilities for much of the preceding decade, the company raised spending to nearly $6 billion in 2010.

The 28-nanometer investment encountered significant opposition from TSMC directors. Research spending at 8% of revenue had attracted little resistance, but roughly tripling capital expenditure raised serious questions about concentrating financial resources on one technological generation.

TSMC’s board had an unusually large independent and international component. Chang said more than three-quarters of directors were independent, with many based in the United States or Britain. Directors normally received meeting agendas two weeks in advance.

Chang cultivated informal discussions with directors before formal board meetings. He invited independent directors to dinner the preceding evening, allowing questions and disagreements to be addressed privately without becoming part of the formal record.

Directors were concerned enough about the 28-nanometer spending proposal to contact TSMC’s general counsel before the meeting. The general counsel also served as board secretary, and the directors requested a direct conversation with Chang.

Chang secured support by asserting his responsibility for operating the company. After presenting the technical, market, and strategic reasoning, he acknowledged the directors’ objections but insisted that operational accountability ultimately rested with him. They accepted his decision.

The 28-nanometer investment coincided with the rapid expansion of smartphones. The timing created enormous demand and helped establish TSMC’s leadership among independent foundries.

Chang did not personally foresee the full scale of the smartphone opportunity. He acknowledged that the market-development team may have had better visibility, but his decision also reflected technical assessments and advice from several sources.

Chang later identified 7 nanometers as another particularly important process generation. He described both 28 and 7 nanometers as opportunities requiring exceptional investment and timely execution.

Chang regarded sales and strategic business development as different functions. Salespeople handled customer relationships and immediate commercial requirements, whereas business development was responsible for understanding future markets and identifying the customers TSMC needed to win.

Before formalizing business development, Chang personally performed much of TSMC’s strategic customer work. He frequently traveled to meet existing major customers and prospective future customers, including companies such as Nvidia and Broadcom. He said these visits were among the activities he enjoyed most.

TSMC had previously debated whether to organize operations by individual divisions or shared functions. In 1996, president Don Brooks proposed a divisional structure resembling the organization of Texas Instruments.

Chang opposed dividing TSMC by fabrication plant because customers often used several generations of manufacturing technology. A single customer could require advanced processors and older-node components, making factory-based divisions commercially inefficient.

TSMC hired McKinsey to evaluate its organization after Brooks continued advocating decentralization. Following approximately two months of work and what Chang estimated as a multimillion-dollar consulting engagement, McKinsey supported the functional structure Chang preferred.

McKinsey used Boeing as an example of a large company organized around shared functions. Chang argued that dividing TSMC by fabrication plant would resemble dividing an aircraft manufacturer by individual aircraft models rather than recognizing shared customers and capabilities.

Rick Tsai nevertheless divided TSMC’s operations into advanced and mainstream technology groups during his 2005–2009 tenure. Chang permitted the experiment because he believed a chairman should generally allow the CEO to make and learn from manageable mistakes.

Chang made a distinction between tolerating management mistakes and allowing existential damage. He believed intervention was justified when an executive’s decisions threatened the company, but otherwise favored granting operational independence.

Mark Liu headed TSMC’s advanced-technology operations, while C.C. Wei headed mainstream technology. Around 2009, the advanced group employed approximately 10,000 people, while mainstream operations employed roughly 7,000–8,000.

Mainstream manufacturing remained strategically important even as TSMC developed newer processes. Older plants produced components that did not need leading-edge technology, including automotive electronics and camera sensors. The same customers frequently purchased both advanced and mature-node products.

Each operating division had its own relatively small business-development team. The teams numbered roughly 30–40 people each, leaving TSMC with fragmented commercial planning despite shared customers.

Chang decided to recombine operations and establish a unified strategic business-development organization. He believed the company had ample manufacturing talent but needed stronger executives capable of understanding future customers and markets.

Chang first offered the combined business-development leadership role to Mark Liu. He intended the assignment to give Liu substantial marketing experience and develop him into a more rounded potential chief executive.

Mark Liu declined because the new position involved far fewer employees. He contrasted his existing responsibility for approximately 10,000 people with a proposed organization of only 60–70, treating the reduction in headcount as a step backward.

Chang believed the business-development role mattered more than its small headcount suggested. He compared the situation to Henry Kissinger’s influence as national security adviser despite supervising fewer employees than the secretary of state.

C.C. Wei accepted the business-development assignment after Liu declined. Chang believed Wei accepted enthusiastically. The decision gave Wei direct exposure to TSMC’s future customers and markets; Wei subsequently became TSMC’s chairman and CEO.

Apple’s relationship with TSMC began through an unexpected family-connected dinner. After a board meeting in Taipei, Chang returned home to learn that Terry Gou, founder of Foxconn and a relative of his wife Sophie, planned to bring an Apple executive to dinner.

Terry Gou’s family relationship gave him a direct personal route to Chang. Gou and Sophie were described as cousins, while Gou’s company, Hon Hai/Foxconn, was already a major Apple manufacturing supplier. The relationship helped create an unusually informal first meeting between senior executives.

Chang had been interested in attracting Apple but believed the company could not be approached through ordinary sales tactics. He regarded Apple as exceptionally secretive and believed suppliers generally had to wait for Apple to initiate contact.

Gou brought Jeff Williams, a senior Apple operations executive, to Chang’s home. Chang recognized that Gou would not arrange such an introduction unless Williams represented a serious business opportunity.

Williams moved almost immediately into a proposal for TSMC to manufacture Apple’s semiconductor wafers. Chang estimated that Williams conducted approximately 80% of the substantive conversation at dinner.

Williams suggested Apple could provide business yielding TSMC a 40% gross margin. Chang privately regarded that as unattractive because TSMC’s margin was already approximately 45%, and he was pursuing a long-term objective of reaching 50%.

Chang chose not to negotiate the margin during the first dinner. He believed the proposed 40% margin was intended generously from Apple’s perspective, but considered an immediate pricing argument premature.

TSMC did not achieve Chang’s 50% gross-margin target before his retirement. He said the company came close but exceeded the threshold later, benefiting from technology leadership and market conditions associated with the COVID period.

Apple’s process requirement surprised Chang: it wanted 20-nanometer chips rather than 28-nanometer chips. TSMC was preparing to manufacture at 28 nanometers and expected to advance naturally toward 16 nanometers.

Chang regarded 20 nanometers as a costly technological detour. It represented an intermediate step between 28 and 16 nanometers, and TSMC’s R&D organization did not yet have sufficient resources to develop two major process generations simultaneously.

Apple had not been included in TSMC’s original 28-nanometer investment forecast. Chang emphasized that the enormous spending decision had already been made on the strength of other market opportunities. Apple emerged as an additional, unexpected customer rather than the justification for that earlier investment.

Serving Apple required TSMC to consider billions of dollars in further investment. The company needed additional capacity for 20-nanometer manufacturing even after making its exceptional commitment to 28 nanometers.

Chang relied on a longstanding relationship with Goldman Sachs to evaluate financing alternatives. TSMC had developed ties with the investment bank early in its history, including through its American depositary receipt listing. Chang had also served on Goldman Sachs’ board.

TSMC’s US-listed American depositary receipts traded separately from its ordinary Taiwanese shares. Chang said the ADRs carried a premium of approximately 20% at the time of the interview.

TSMC restricted conversion of ordinary shares into ADRs. Such conversions required board approval, which was not easily obtained. The arrangement prevented shareholders from freely exploiting the price difference between the markets.

TSMC examined dividend cuts, new equity issuance, and borrowing to finance Apple’s manufacturing requirements. New shares could have been issued in Taiwan or through the US market, while corporate bonds offered another source of capital.

Chang decided against issuing additional equity or reducing dividends. After consulting Goldman Sachs and reviewing the numbers, TSMC chose to borrow rather than dilute shareholders or disrupt its existing dividend policy.

TSMC agreed initially to manufacture only half the volume Apple requested. Chang considered full participation too financially risky given the enormous upfront investment required. The decision limited exposure while still establishing Apple as a major customer.

Apple’s purchasing employees initially considered TSMC’s half-volume offer unreasonable. C.C. Wei communicated the proposal and received a response amounting to “You must be crazy.” He referred the issue back to Chang rather than arguing with the purchasing team.

Chang personally explained TSMC’s financial limits to Jeff Williams in Cupertino. He described the decision as the result of prudent financial planning and emphasized that TSMC would need to issue corporate bonds even to serve half the requested demand.

Williams proposed eliminating TSMC’s dividend to create investment capacity. He believed shareholders would understand the sacrifice, but Chang rejected the suggestion.

Dividend dependence among shareholders was central to Chang’s refusal. His analysis indicated that roughly one-third of TSMC shareholders were strongly interested in receiving dividends. He expected eliminating payments to provoke substantial selling pressure and a sharp decline in the share price.

Williams accepted TSMC’s financing decision without a prolonged confrontation. Once Chang rejected the dividend proposal, Williams allowed the matter to rest, and the companies settled on the initial manufacturing commitment.

TSMC had developed customer-deposit requirements years earlier to manage capacity risk. During its first 10–15 years, manufacturing capacity was frequently scarce, while customers had incentives to request more wafers than they would ultimately need.

Chang instructed TSMC salespeople to warn customers that deposits could be confiscated if they failed to take reserved wafers. The policy attempted to prevent customers from transferring all the risk of speculative demand forecasts to the manufacturer.

TSMC never actually confiscated the deposits, according to Chang. The threatened penalty provided negotiating leverage, but the company ultimately allowed customers to use their deposits later rather than forfeiting them.

The dot-com downturn tested TSMC’s deposit policy. Chang dated the semiconductor decline from the first quarter of 2001 to a recovery around the third quarter of 2003.

TSMC had expanded capacity before the downturn, including by acquiring companies with existing fabrication facilities. Some customers who had reserved production capacity no longer needed their scheduled wafers once demand contracted.

TSMC allowed those customers to postpone their purchases rather than lose deposits. Chang said the customers eventually used the manufacturing credits they had purchased, preserving relationships while avoiding the reputational cost of confiscation.

Intel briefly threatened to interrupt Apple’s emerging relationship with TSMC in February 2011. Williams called Chang and requested that discussions pause for two months because Intel’s senior leadership had approached Tim Cook about supplying Apple’s chips.

Intel already had an established relationship with Apple through its Mac computers. The Mac product line used Intel processors at the time, giving Intel an existing commercial connection even though the discussion concerned Apple’s mobile chips.

Chang believed Intel was no longer as intimidating in 2011 as it had been during the 1990s. He considered TSMC approximately equal to Intel in relevant technology, stronger in manufacturing execution, and superior in customer relationships.

Chang initially assumed Intel’s approach might involve an exceptionally influential industry figure. He later learned that the relevant senior executive was Intel CEO Paul Otellini, rather than a figure such as the retired Andy Grove.

Chang contacted Apple again approximately one month into the requested pause. He arranged to visit Silicon Valley and asked Williams for an update on Intel’s competing proposal.

Williams unexpectedly arranged for Tim Cook himself to meet Chang. Chang remarked that it was rare for a business contact to refer a visiting executive upward to the contact’s own superior rather than delegate the meeting to a subordinate.

Cook hosted Chang for lunch at Apple and reassured him about Intel. They collected food from Apple’s cafeteria and brought it back to Cook’s office.

Cook’s explanation was that Intel did not understand how to operate as a foundry. Chang regarded the statement as decisive reassurance that Apple's consideration of Intel did not seriously threaten TSMC’s opportunity.

Chang interpreted Cook’s concern primarily as one of customer service and trust. He believed Intel had historically behaved as a dominant supplier, while a contract manufacturer needed to respond courteously even to demanding, unusual, or unreasonable customer requests.

Chang rejected the hosts’ suggestion that Cook was chiefly referring to Intel’s conflict of interest as both chip designer and manufacturer. Although TSMC’s refusal to compete with customers was important, Chang believed Cook was specifically criticizing Intel’s ability and willingness to accommodate customers.

Chang said many Taiwanese PC manufacturers disliked dealing with Intel. Based on conversations with Intel customers, he believed they wanted an alternative supplier and resented Intel’s historically superior attitude toward them.

TSMC and Apple eventually negotiated an acceptable 20-nanometer manufacturing price. Williams visited TSMC personally, explained Apple’s component-cost targets, and negotiated against TSMC’s own calculations of manufacturing costs and acceptable margins.

Chang identified manufacturing cost and customer acceptance as the central determinants of semiconductor pricing. Yield affected production economics, but the practical negotiation involved establishing a price that covered the supplier’s requirements while satisfying the buyer’s commercial limits.

The executives celebrated the pricing agreement with dinner in Taipei. Chang selected an expensive restaurant, prompting Williams to joke that an unfavorable deal might instead have resulted in dinner at McDonald’s.

The interview highlighted the different margins captured across semiconductor businesses. The hosts estimated TSMC’s gross margin at approximately 55–57%, while some customers earned gross margins of 70–80%. Both chip designers and manufacturers contributed substantial value.

Chang said pricing required finding a middle ground between immediate returns and long-term relationships. He regarded the balance as a matter of executive judgment rather than a universal formula, noting that commodity suppliers often lacked meaningful pricing power.

Apple’s 20-nanometer order caused TSMC to delay development of its 16-nanometer process. Chang identified this as the most serious technological cost of accepting Apple’s business, because TSMC could not then run the two development programs in parallel.

Samsung advanced ahead of TSMC at 16 nanometers because it had not needed to pursue Apple’s 20-nanometer program. Its freedom to concentrate on the next process generation gave it a temporary advantage.

Chang was shocked when Apple placed its initial 16-nanometer orders with Samsung. TSMC had already committed enormous capital to Apple and expected much of its 20-nanometer manufacturing equipment to support future production at 16 nanometers.

TSMC had expected approximately 80–90% of its relevant manufacturing equipment to be convertible to 16-nanometer production. Losing Apple’s next-generation orders could therefore have undermined the financial logic of the earlier investment.

Chang immediately confronted Williams by email about Apple’s Samsung orders. He explained that TSMC had invested heavily in manufacturing equipment in anticipation of serving Apple’s subsequent process generations.

Williams responded promptly and traveled to explain Apple’s decision. He assured Chang that Apple would purchase its 16-nanometer requirements from TSMC once TSMC’s technology became ready.

TSMC completed its 16-nanometer development approximately six months later and recovered most of Apple’s requirements. Chang confirmed that Williams had followed through on the commitment, limiting the long-term commercial damage of Samsung’s temporary lead.

The Apple episode illustrated the financial risk of allowing a major customer to redirect a manufacturer’s technology roadmap. TSMC secured valuable business but also accepted additional capital requirements, delayed its planned process development, and temporarily exposed subsequent orders to a competitor.

Samsung’s position as an Apple competitor complicated its foundry relationship with Apple. The hosts contrasted Samsung’s combined manufacturing and product businesses with TSMC’s dedicated manufacturing model, which did not place TSMC in direct competition with its chip-design customers.

Chang described TSMC’s position in the semiconductor supply chain as providing unusual visibility into competitors’ difficulties. By observing where customers moved their manufacturing orders, he could sometimes identify a supplier’s financial or technological problems before they became publicly obvious.

TSMC spent much of the 1990s trying to win Qualcomm as a major customer. Chang considered Qualcomm a technologically sophisticated company and therefore an especially attractive prospective foundry client.

Qualcomm initially gave TSMC very little business despite repeated executive meetings. Chang regularly met Qualcomm’s senior operations or procurement executive, who remained polite but continued relying primarily on another manufacturer.

Chang understood that IBM was Qualcomm’s principal foundry during that period. He gathered this information through industry relationships rather than receiving a direct explanation from Qualcomm.

Qualcomm began transferring substantial manufacturing business to TSMC around 1997–1998. Chang said the volume grew rapidly, indicating an important change in Qualcomm’s manufacturing relationships.

Chang interpreted Qualcomm’s shift as evidence that IBM’s semiconductor business was in difficulty. He believed IBM depended significantly on Qualcomm and a small number of other fabless customers to support its foundry operations.

IBM subsequently approached TSMC about jointly developing 130-nanometer technology around 1999. The request confirmed Chang’s expectation that IBM would seek a strategic response to the loss of business.

Chang rejected IBM’s joint-development proposal immediately. He believed IBM expected to dominate partnerships and would require TSMC engineers to work within IBM’s organization, making TSMC dependent on jointly controlled technology.

Preserving independent process-development capabilities was central to Chang’s refusal. He feared TSMC would lose control over its technical roadmap and become entangled in organizational and cultural disputes with IBM.

IBM reacted angrily to TSMC’s rejection, according to Chang. He believed IBM executives still regarded their company as technologically and institutionally superior to a smaller Taiwanese competitor.

IBM approached United Microelectronics Corporation after TSMC declined. UMC accepted the arrangement, which Chang said it later seriously regretted. By 1999, UMC was already smaller than TSMC rather than an equal-sized competitor.

Chang’s understanding of the semiconductor learning curve originated at Texas Instruments. He emphasized that he did not invent the concept but helped refine and apply it to semiconductor manufacturing.

The learning curve describes declining unit costs as cumulative production experience increases. Chang explained that repeated manufacturing creates efficiencies, while warning that the simple observation understates its strategic and operational implications.

Bruce Henderson, founder of Boston Consulting Group, introduced the experience-curve approach to Texas Instruments around 1970. He contacted senior TI executive Mark Shepherd and proposed applying BCG’s research to semiconductor manufacturing.

Henderson and Bill Bain made a presentation to TI management in Dallas. Shepherd invited Chang and other executives, and TI subsequently agreed to work with BCG on understanding the economics of semiconductor production.

Shepherd assigned Chang as TI’s principal collaborator with Bill Bain. Bain received an office near Chang and spent approximately three days each week in Dallas, studying TI’s manufacturing costs, prices, and semiconductor product families.

Chang and Bain worked closely for approximately two years. Bain required access to detailed company information, while Chang participated in discussions about the conclusions emerging from the analysis.

Chang said the collaboration gave him a lasting framework for business decisions. He continued using learning-curve reasoning decades later, especially when evaluating production volumes, costs, pricing, and competitive positions.

Bill Bain privately informed Chang that he planned to leave BCG and establish his own consulting firm. Chang said he was among the first people outside BCG to learn of the decision, despite Henderson’s high regard for Bain.

Bain described starting his own firm as a personal “world imperative.” Chang remembered the unusual phrase, which reflected Bain’s determination to pursue an independent venture.

The hosts emphasized that the learning curve rewards manufacturers capable of reaching high production volumes early. Spreading fixed costs across more units and gaining manufacturing experience can produce lower costs, attracting additional customers and creating a reinforcing advantage.

Learning-curve pricing may justify aggressive early price reductions. The hosts explained that a manufacturer can price products closer to expected future costs, potentially accepting weak or negative early margins to build demand and achieve competitive scale.

The strategy depends on estimating the mature market before manufacturing volume becomes large. A foundry must forecast how much demand a process generation will eventually attract, then invest and price accordingly.

Learning-curve economics connect technology planning with financial strategy. Pricing, plant investment, debt financing, equity issuance, and the timing of production expansion must all support the objective of reaching sufficient volume without jeopardizing the business.

Apple illustrated the limits of pursuing maximum volume at any cost. Accepting all of Apple’s requested production could have accelerated TSMC’s progress down the cost curve, but would also have exposed the company to far greater financial risk. Chang chose half the volume instead.

TSMC must forecast demand for its customers’ products rather than only demand for manufacturing services. The hosts noted that producing for Apple ultimately requires estimating future iPhone sales, while manufacturing for Nvidia requires judging the size and growth of markets such as artificial intelligence.

Forecasting errors can undermine an entire process generation’s profitability. The hosts argued that even a 5–10% miss in demand projections could significantly reduce margins and cash flow, weakening a foundry’s ability to finance its next technology generation.

Successful execution creates a cumulative financial advantage for the largest foundry. Greater volume supports lower costs and higher cash generation, enabling larger subsequent investments in both research and fabrication capacity.

Chang founded TSMC in 1987 at approximately 56 years old. Its subsequent rise made it one of the world’s most valuable companies and a central supplier to the global technology industry.

Chang said TSMC ultimately exceeded his original expectations, but not by an order of magnitude. He rejected the suggestion that its success had seemed highly improbable to him at the time of founding.

Chang denied that TSMC was ever intended to stop permanently after constructing two fabrication plants. He acknowledged that such a limit appeared in a very early plan, but said his knowledge of the learning curve made a lasting two-fab business implausible.

TSMC’s original business model focused exclusively on manufacturing chips designed by other companies. The company deliberately avoided competing with its customers in chip design or branded semiconductor products.

The pure-play foundry concept appeared in Chang’s original presentations to the Taiwanese government and investors. During their visit to TSMC’s museum, the hosts saw early presentation materials explicitly identifying the company as a dedicated manufacturing operation.

The hosts believed the founding strategy partly reflected Taiwan’s existing capabilities. Taiwan could develop advanced manufacturing expertise but lacked the same established chip-design ecosystem as companies such as Intel, making specialized fabrication a more realistic initial opportunity.

TSMC initially depended heavily on conventional integrated semiconductor companies rather than fabless designers. Customers such as established chip manufacturers outsourced less strategic production or used TSMC as an additional manufacturing source.

The hosts characterized much of TSMC’s early business as low-value or secondary work. Although unattractive compared with advanced proprietary products, the orders provided revenue, manufacturing experience, and the utilization needed to develop the company’s capabilities.

Chang maintained that he had anticipated the emergence of fabless semiconductor companies. The hosts noted that he had sometimes acknowledged favorable timing in earlier discussions, but in this interview he strongly emphasized that the business model was foreseeable.

An encounter with Gordon Campbell helped Chang recognize the economic potential of fabless chip design. Campbell approached Chang during his final period at General Instrument, seeking approximately $50 million to establish a semiconductor company.

Campbell initially wanted General Instrument to finance a conventional semiconductor startup. When Chang requested a written business plan, Campbell agreed to produce one within approximately three weeks.

Campbell subsequently reduced his required investment from $50 million to $5 million. He told Chang that he no longer needed the larger funding because he would not build a semiconductor fabrication plant.

Campbell’s revised plan demonstrated that chip designers could start companies with far less capital by outsourcing manufacturing. Chang recognized that separating design from fabrication could create a new class of semiconductor customers.

Another early fabless startup approached General Instrument for manufacturing capacity. The startup already existed without its own fabrication facilities and wanted General Instrument, which had unused plant capacity, to manufacture its wafers.

That early outsourcing relationship exposed a basic tension in the foundry business. The chip designer wanted the plant operated according to its own requirements, while General Instrument’s manufacturing manager wanted to control the facility. Both parties depended on the arrangement but had different operational priorities.

Chang recognized that a foundry would need to satisfy many customers with one manufacturing system. Individual customers might request different practices, but a successful foundry needed standardized processes flexible enough to serve them collectively.

The same difficulty created the foundry model’s economic advantage. By aggregating demand from independent customers, a manufacturer could operate at volumes that individual chip designers could not economically support on their own.

The hosts concluded that TSMC’s refusal to compete with customers became one of its most important competitive advantages. It could serve Apple, Nvidia, Qualcomm, AMD, and other chip designers without simultaneously trying to sell rival chips or products.

TSMC’s business differs fundamentally from vertically integrated manufacturers such as Intel and Samsung. It concentrates on manufacturing technology and production services rather than combining chip design, fabrication, and customer-facing semiconductor products within one organization.

The hosts believed customer neutrality became increasingly valuable as chip design separated from chip fabrication. A single specialized manufacturing company could aggregate orders from multiple competing chip designers without requiring those designers to trust a direct rival.

The rise of Arm’s processor architecture helped make this industry structure possible. The hosts argued that Arm enabled a large ecosystem of independent chip designers to compete in phones, computers, servers, and eventually AI hardware.

The hosts contrasted Arm’s ecosystem with a hypothetical continuation of Intel’s x86 dominance. If Intel had retained control of the dominant computing architecture and associated production, the market might have remained more vertically integrated, leaving less room for a large independent foundry.

AMD was identified as an important exception in the x86 market. It designs x86 processors while outsourcing manufacturing to TSMC, demonstrating that x86 designs can participate in a fabless model even though Intel historically dominated the architecture.

The hosts considered Apple’s adoption of Arm-based silicon particularly important. They argued that without a viable alternative architecture, Apple might have remained much more dependent on Intel, although they acknowledged uncertainty over whether x86 technology could have succeeded in mobile devices.

Several complementary semiconductor companies emerged within a remarkably short historical period. TSMC, Arm, Synopsys, Cadence, and ASML were all founded during the mid-to-late 1980s, creating specialized businesses that later became central to the industry.

The semiconductor value chain became divided among architecture providers, designers, software suppliers, equipment manufacturers, and foundries. Arm supplied architectures, Nvidia and Apple designed chips, Cadence and Synopsys supplied electronic design tools, ASML supplied manufacturing equipment, and TSMC manufactured the resulting semiconductor designs.

The hosts regarded the separation of these functions as one of the defining developments in the modern semiconductor industry. Each specialist could serve multiple companies rather than requiring every chipmaker to develop an entire vertically integrated operation.

The hosts visited TSMC’s headquarters and manufacturing ecosystem in Hsinchu Science Park during their Taiwan trip. The visit made the geographic concentration of semiconductor expertise and infrastructure more apparent than it had been during their earlier research.

Hsinchu Science Park contains far more than TSMC’s own factories. The hosts observed nearby facilities or offices associated with Arm, Synopsys, Cadence, Qualcomm, MediaTek, and other companies participating in the semiconductor supply chain.

The close proximity of semiconductor partners enables frequent technical coordination. Chip-design software companies, manufacturing engineers, designers, and customers need detailed information about future process technologies to ensure designs can be manufactured successfully.

The hosts believed the physical proximity of these companies contributed to faster collaboration. Rather than treating the relationships as distant commercial arrangements, they saw an ecosystem in which engineers and executives could interact directly across neighboring facilities.

Two universities within the science-park ecosystem supplied specialized technical talent. The hosts emphasized the importance of nearby institutions training engineers and doctoral researchers who could enter the surrounding semiconductor businesses.

Taiwan’s semiconductor infrastructure extends beyond the original Hsinchu Science Park. The hosts referred to approximately three major science-park ecosystems in Taiwan, including substantial manufacturing operations in the south.

Flights into Taiwan appeared heavily populated by semiconductor and technology workers. The hosts noticed travelers associated with companies including Google, Amazon, Arm, and Marvell, reinforcing their impression that Taiwan served as a central gathering point for the global chip industry.

The hosts argued that TSMC’s Taiwanese ecosystem could not be reproduced quickly elsewhere. It involved decades of accumulated manufacturing experience, universities, suppliers, designers, equipment specialists, and relationships that extended beyond the buildings themselves.

They distinguished overseas factory construction from recreating the entire semiconductor cluster. Facilities in places such as Arizona could serve important customers and government objectives, but the hosts doubted they could immediately achieve the same advantages as Taiwan’s concentrated infrastructure.

Government-backed industrial development played a central role in establishing Taiwan’s semiconductor cluster. The hosts regarded Hsinchu Science Park as an exceptionally successful example of targeted government support for a particular industry.

The hosts saw a major 2-nanometer fabrication facility under development during their visit. They described the complex as extraordinarily large, with one section apparently operational or approaching production while several additional phases remained incomplete.

The hosts expected 2-nanometer production to expand during the second half of 2025. They speculated that future Apple chips and Nvidia processors would help drive demand, while acknowledging that their observations did not establish specific customer orders.

Individual modern fabrication facilities occupied areas comparable to multiple football fields. The hosts emphasized that the enormous physical size and complexity of the buildings made the financial scale of TSMC’s investments more tangible.

TSMC historically retained older manufacturing capabilities rather than systematically replacing every factory with newer processes. The hosts contrasted this with Intel’s tendency to upgrade and repurpose existing facilities for advanced production.

The original government research infrastructure contributed to TSMC’s early factory base. The hosts described Fab 1 as originating with Taiwan’s Industrial Technology Research Institute, where Chang had worked before founding TSMC.

TSMC’s Fabs 2 and 3 were described as early company-specific factories dating from the late 1980s. According to the hosts, these facilities remained in operation decades later, illustrating the continuing usefulness of mature process technologies.

Older semiconductor processes retain commercial demand long after newer technologies become available. Industrial machinery, automobiles, and electronic products may need compatible replacement chips decades after their original manufacture.

Long equipment lifetimes create demand for exact replacement components. The hosts cited products manufactured 10, 20, or 30 years earlier whose maintenance requirements can depend on semiconductor technologies no longer considered advanced.

Mature-node manufacturing also serves new products that do not require maximum transistor density. CMOS camera sensors were identified as an example of semiconductor components that may benefit from specialized processes without requiring the newest two-nanometer technology.

Older, fully depreciated fabrication equipment can generate attractive margins. Once the original capital investment has been depreciated, continued production may require primarily operating expenses and maintenance capital, improving the economics of manufacturing mature-node products.

The hosts did not present TSMC’s older-fab strategy as universally superior to Intel’s approach. Intel historically emphasized concentrating assets on newer technologies, while TSMC maintained a broader range of process generations to serve customers with different requirements.

TSMC’s financial advantage increases when it can retain both advanced and mature-node customers. Older manufacturing facilities provide ongoing revenue while newer factories capture demand from the latest processors, expanding the volume available to the company.

The hosts argued that semiconductor manufacturing has strong winner-take-most characteristics. Large volumes reduce costs, improve available cash, and support the expensive investments required to maintain technological leadership.

The cost of building advanced fabrication facilities creates an increasingly high barrier to entry. The hosts cited approximately $20 billion for a modern fab and speculated that future plants might eventually cost $40 billion, $80 billion, or even $100 billion.

Increasing capital requirements could further concentrate advanced chip manufacturing. Very few companies would be able to finance facilities costing tens of billions of dollars while also funding the research needed to make those factories technologically competitive.

TSMC’s competitive advantage depends on both capital expenditure and research spending. Building a factory is insufficient without developing the semiconductor process technology that makes its equipment commercially useful.

The hosts highlighted TSMC’s unusually large and sustained reinvestment. They observed that its capital expenditure has often approached the scale of its annual net income, while R&D represents another substantial ongoing financial commitment.

Scale enables TSMC to invest more than smaller competitors. Higher manufacturing volume can support greater cash generation, allowing TSMC to spend more on future process technologies and capacity, reinforcing the advantage that produced its market position.

Advanced packaging has become an additional source of TSMC’s technological advantage. The hosts cited CoWoS, TSMC’s packaging technology used for AI chips, as an important capability beyond conventional transistor manufacturing.

Proprietary packaging can make it harder for customers to switch foundries. A chip designed around TSMC’s particular manufacturing and packaging technologies may require significant redesign or adaptation before another manufacturer can produce it.

TSMC also employs specialized packaging technologies for mobile processors. The hosts distinguished these solutions from CoWoS while identifying them as another area where the company’s technical capabilities strengthen customer relationships.

The hosts expected technological leaders to become more difficult to displace unless they made major mistakes. They argued that market share, manufacturing experience, capital availability, research capabilities, and proprietary packaging could reinforce one another over successive process generations.

Moore’s law was identified as the industry's central long-term economic force. The hosts connected the continuing demand for computing power to decades of semiconductor growth, beginning with the invention of the integrated circuit and Chang’s early career at Texas Instruments.

The hosts distinguished the original technical formulation of Moore’s law from its broader economic effect. They acknowledged that traditional transistor-scaling improvements face increasingly difficult physical limits, even as demand for computing power continues expanding.

Advanced packaging, interconnects, architecture, and software increasingly contribute to computing improvements. The hosts argued that innovation now extends beyond the narrower transistor-scaling mechanisms associated with the original form of Moore’s law.

The hosts emphasized that demand for computing provides continuing incentives for semiconductor innovation. Even when improvements become technically difficult, substantial commercial rewards encourage companies to develop new methods of increasing performance.

The global semiconductor market expanded enormously during TSMC’s existence. The hosts cited growth from approximately $26 billion in 1987 to $527 billion in 2024, illustrating the industrial expansion accompanying TSMC’s rise.

TSMC benefited from both overall semiconductor-market growth and changes in industry organization. The hosts argued that increasing demand for chips combined with the separation of design and manufacturing created the conditions for a trillion-dollar-scale independent foundry.

The hosts considered Chang’s early commitment to manufacturing specialization increasingly consequential in retrospect. What initially appeared to be a modest business serving other companies’ manufacturing requirements eventually became a central position in the global technology economy.

The overall explanation for TSMC’s success combined early recognition of fabless design, manufacturing specialization, customer neutrality, disciplined investment, and scale. No single decision guaranteed its outcome; the company repeatedly had to execute on process technology, secure customer demand, manage financing, and preserve commercial trust.

Chang’s personal management decisions illustrated the tension between financial discipline and long-term investment. He opposed layoffs that could destroy skilled capacity, protected dividends when financing Apple, limited Apple’s initial order to manage risk, and nevertheless committed aggressively to technology generations he believed would define future demand.

The Nvidia and Apple negotiations illustrated the importance of direct relationships between senior executives. Chang personally resolved a nine-figure Nvidia dispute with Huang and handled manufacturing commitments, pricing, and competitive uncertainty with Apple’s top leadership.

The interview challenged the idea that semiconductor manufacturing is merely a commodity service. TSMC’s role required technological invention, massive financial commitments, detailed knowledge of customer markets, management of manufacturing risk, and the ability to coordinate an increasingly complex global supply chain.

The hosts’ principal conclusion was that TSMC’s apparent inevitability becomes clearer when viewed through learning-curve economics. The company that consolidates manufacturing volume can lower costs, attract additional business, and reinvest more heavily in the next generation—provided it maintains customer trust, accurate forecasting, and disciplined execution.