Morris Chang and TSMC: The $100 Million Nvidia Settlement, Apple’s Secret Deal and the Making of a Semiconductor Giant

In 2009, Morris Chang arrived at Jensen Huang’s Silicon Valley home for dinner. They ate pizza and salad, as they had many times before. At 8 p.m., Chang suggested moving to Huang’s study. He had an offer to make: more than $100 million to settle a manufacturing dispute between TSMC and Nvidia.

There would be no negotiation. Huang had 48 hours to accept or take the dispute to arbitration. He accepted.

The meeting captured something important about the company Chang had built. Taiwan Semiconductor Manufacturing Company, better known as TSMC, did not design Nvidia’s chips, build Apple’s iPhones, or sell processors under its own name. It manufactured the semiconductors behind other companies’ products. That required extraordinary technical ability, billions of dollars in investment, and relationships capable of surviving expensive failures.

In a January 27, 2025 interview with Ben Gilbert and David Rosenthal of Acquired, Chang, then 93, revisited the decisions that turned a Taiwanese manufacturing venture into one of the world's most important technology businesses. Jensen Huang helped arrange the interview, which drew on Chang’s recently published Chinese-language memoirs.

His account reveals how Nvidia became a customer, why he returned as CEO after employee layoffs, how Apple nearly disrupted TSMC’s technology roadmap, and why Intel, IBM, and Samsung struggled to match different parts of its business.

The financial records and subsequent developments add another dimension. Some of Chang’s recollections are supported by contemporary filings; others remain personal accounts. And the economic forces he described in 2025 have since pushed TSMC to investment levels far beyond those that once alarmed its directors.

Nvidia: The Small Customer That Became Indispensable

Jensen Huang's Letter to Morris Chang in 1997

In 1997, Nvidia was a four-year-old graphics-chip company with approximately 50–60 employees and serious financial problems. TSMC was already considerably larger, employing thousands and having surpassed $1 billion in annual revenue in 1995.

Nvidia needed someone to manufacture its promising new chips. Its attempts to contact TSMC's San Jose office had produced little response, so Jensen Huang wrote directly to Morris Chang.

Chang was curious but irritated. He had repeatedly instructed salespeople not to neglect prospective customers simply because they were small. Huang's letter suggested that his organization had done exactly that.

The following week, Chang traveled to California and called Huang without warning. Huang answered amid a noisy discussion with employees. As soon as he realized who was calling, he demanded silence.

They arranged a meeting.

Huang impressed Chang with his confidence and candor. Nvidia was in financial difficulty, but Huang believed its next graphics chip would save the company and eventually make it one of TSMC’s largest customers.

Chang considered that a bold prediction. A major customer would need to generate approximately $50 million annually in manufacturing business for TSMC.

Within two or three years, Nvidia had become one of TSMC’s five largest customers.

What Nvidia’s Early Financial Records Reveal

The successful chip discussed in the interview was associated with Nvidia’s 1997 turnaround. The hosts suggested the RIVA 128, although Chang could not remember the precise designation.

A more detailed account appears in Nvidia’s January 1999 IPO prospectus.

The prospectus confirms that Nvidia commercially launched the RIVA 128 in August 1997. Its growing sales transformed the company's finances after disappointing earlier products. Nvidia reported a gross profit of approximately $7.8 million in 1997, compared with about $874,000 in 1996.

However, the manufacturing story was more complicated than a single chip rescuing Nvidia through TSMC alone.

Nvidia initially worked with SGS-Thomson, later known as STMicroelectronics, on the RIVA family. Its subsequent products, including the RIVA 128ZX and RIVA TNT, involved TSMC.

By 1998, Nvidia had become heavily dependent on TSMC, but the relationship was already experiencing difficulties.

The prospectus disclosed problems achieving volume production of the RIVA 128ZX and RIVA TNT at TSMC. It warned investors about manufacturing yields, delivery reliability, limited capacity commitments, and the difficulty of finding replacement suppliers.

That distinction matters. Chang's account establishes the importance of Huang's personal intervention and the relationship that followed. Nvidia's contemporary filing establishes that the partnership also faced manufacturing difficulties long before the much larger dispute in 2009.

The document also demonstrates a structural vulnerability of fabless companies: outsourcing removes the enormous expense of owning factories, but leaves chip designers dependent on manufacturers for production quality, capacity, and delivery.

The 40-Nanometer Failure in 2009

By 2009, Nvidia was an important customer of TSMC’s 40-nanometer manufacturing technology.

A process node describes a generation of semiconductor manufacturing. Moving to newer nodes generally enables improvements in transistor density, power consumption, or performance, although the nanometer designation does not directly measure a single physical feature.

TSMC’s 40-nanometer process was experiencing yield and quality problems. Yield means the proportion of manufactured chips that function correctly. Poor yields increase costs because a larger share of production cannot be sold as intended.

The difficulties affected several customers, including Nvidia, Qualcomm, and others. Nvidia was particularly exposed because of its substantial use of the process.

Chang explained that the dispute contained two related but separate problems. TSMC needed to improve manufacturing yields for everyone. It also needed to address the financial losses Nvidia claimed it had suffered.

The previous CEO had relied on TSMC’s quality director, who maintained that TSMC was not at fault. No compensation had been offered to Nvidia.

After returning as CEO in 2009, Chang telephoned major customers. Huang remained friendly but made clear that the manufacturing dispute had not been resolved.

Chang promised to investigate and respond.

During his first four or five weeks back, he estimated that nearly half his working time went into the Nvidia problem. He consulted salespeople, manufacturing specialists, and others familiar with the dispute. He also investigated the demands that Nvidia’s own customers were making of it.

Eventually, he arrived at a settlement figure.

A $100 Million Offer, a Private Dinner and a 48-Hour Deadline

Chang contacted Huang and announced that he would visit his home during an upcoming Silicon Valley trip.

The two men had an established personal relationship, including informal dinners with their families. Huang’s wife, Lori, often prepared the salad.

Chang proposed that they eat first and discuss business afterward. Huang immediately wanted to know when the business discussion would take place.

The agreed schedule was dinner around 6:30 p.m., followed by negotiations in Huang’s study at 8 p.m.

Chang presented an offer exceeding $100 million.

He recalled telling Huang:

"We are not going to argue. We're not going to bargain."

The proposal would remain available for 48 hours. Otherwise, TSMC would proceed to arbitration, which Huang had previously proposed to the former CEO.

Chang said he did not want arbitration, but believed the figure he had calculated was fair to both companies.

Huang accepted within the deadline.

The exact settlement amount was not disclosed in the interview, and the private negotiation has not been independently established through a public settlement agreement. The figure above $100 million therefore rests on Chang's account.

Still, its importance extends beyond the payment. A company that denied responsibility had moved to compensate a major customer, resolving a conflict that threatened years of cooperation.

TSMC and Nvidia subsequently conducted billions of dollars in additional business together. The manufacturing relationship continued into the artificial-intelligence era, when Nvidia’s processors became a major driver of advanced semiconductor demand.

The 2009 failure also foreshadowed another TSMC challenge: mastering 40 nanometers was a necessary step toward 28 nanometers, where Chang intended to make a far larger bet.

The 2009 Leadership Crisis: Why Morris Chang Took Back the CEO Job

Rick Tsai's Layoffs and the Dispute Over Performance Reviews

The Nvidia dispute was only one of several problems confronting TSMC in 2009.

Chang had previously handed the CEO position to Rick Tsai, a longtime executive whom he considered a possible successor, while retaining the chairmanship. In Taiwan, Chang explained, the chairman often remained the most powerful corporate figure even when another person managed daily operations.

The global financial crisis of 2008 sharply reduced semiconductor demand. TSMC’s revenue suffered, and its gross margins were under pressure because manufacturing costs were not falling as quickly as previously agreed customer prices.

Tsai responded partly by reducing employment.

According to Chang, approximately 600–700 employees with the lowest performance ratings were dismissed. Tsai did not characterize the decision as a conventional layoff. Instead, the dismissals were treated as consequences of poor performance.

Chang strongly objected.

He had instructed Tsai to take any proposed layoffs to the board, where Chang intended to oppose them. Treating the dismissals as performance management avoided that process.

The deeper disagreement concerned whether performance evaluations were a fair basis for deciding who should lose their jobs.

Ratings were assigned by individual supervisors. Chang argued that assessments by hundreds of different managers could not provide an objective comparison of employees across an organization.

He preferred placing struggling employees on six months of probation, followed by reassignment when appropriate. Under his leadership, outright dismissal for poor performance had been uncommon.

This was not a new position. Chang recalled challenging performance-based layoffs at Texas Instruments during the early 1970s, when other senior executives favored dismissing the lowest-rated employees.

His objection was both ethical and economic.

He calculated that severance might cost approximately six months of compensation, while preparing replacement employees could require another six months. If business recovered quickly enough to require those workers again, the financial benefits of dismissal could disappear.

Semiconductor production also depended on specialized experience. Retaining trained employees could protect capabilities that were difficult to rebuild during a recovery.

What Contemporary Reporting Reveals About the Layoffs

Independent reporting confirms that TSMC's workforce reductions generated a significant labor dispute.

On March 25, 2009, the Taipei Times reported protests by former employees. Workers alleged that some colleagues had received poor performance assessments after taking maternity, personal, or sick leave.

Some also accused managers of pressuring them to sign voluntary-resignation agreements, complicating access to unemployment benefits.

Those were allegations made by workers, not independently established findings of wrongdoing. TSMC told the newspaper that it was working with former employees to address unemployment-benefit claims.

On April 6, TSMC publicly acknowledged that more than 200 former employees had received documentation confirming that their employment ended because of deteriorating business conditions. It also announced additional severance arrangements.

The accounts differed on the total number of affected workers, partly because reports included different periods and categories of departures. Chang's retrospective figure was approximately 600–700. Contemporary accounts sometimes suggested higher totals.

More importantly, the public record establishes the chronology of TSMC's reversal.

On May 20, 2009, while still chairman rather than CEO, Chang announced that dismissed employees would be welcomed back beginning in June. He acknowledged that the company had misapplied its employee appraisal system.

The May 21 Taipei Times report also documented TSMC's improving business conditions. April sales had increased substantially from March, making reinstatement financially more feasible.

Chang did not formally replace Tsai as CEO until June.

There is another revealing difference between Chang's 2025 recollection and the contemporary record.

In the interview, he identified the layoffs as a central reason for reclaiming executive control. But when the leadership change was announced in June 2009, he publicly rejected speculation that it was related to the dismissals, according to the June 15 Taipei Times report.

Both statements are part of the historical record. The later account provides Chang's retrospective explanation; the earlier account shows what he was prepared to acknowledge publicly at the time.

Sophie Chang and the Employees Outside Their Home

The dispute became personal when dismissed workers began demonstrating outside Chang's residence.

He recalled one protest involving more than 100 people, prompting the deployment of approximately 50–60 police officers. The demonstrators disrupted access to the neighborhood.

Another group of approximately 25 protesters remained overnight in a small park near the house.

Chang's wife, Sophie, was distressed by the situation and repeatedly checked on the protesters during the night.

At approximately 6 a.m., she went to a neighborhood market with a bodyguard and purchased breakfast for roughly 25–30 people. She brought Chinese bread and other breakfast food to the demonstrators.

The workers thanked her and reportedly abandoned a planned march to Taiwan's presidential residence that day.

The incident did not resolve the underlying employment dispute. But it illustrated how a corporate decision had affected individual families and reached the founder's own doorstep.

Rick Tsai's Reassignment and the Failed Green-Energy Businesses

Chang did not dismiss Tsai or reduce his compensation.

He retained Tsai's organizational grade, salary, and bonus, and continued describing him as a potential successor. Tsai instead became president of new businesses, responsible for developing TSMC's solar-cell and light-emitting-diode operations.

Both industries appeared promising in 2009.

TSMC subsequently invested in solar research and solid-state lighting facilities. Its 2010 shareholder letter confirms construction activity involving thin-film solar and lighting operations.

Neither business achieved the hoped-for success.

Chang attributed the solar venture's difficulties to aggressive Chinese government subsidies, which supported low-priced production and helped Chinese manufacturers dominate the market. He blamed part of the LED venture's failure on important patents being concentrated among a small number of companies unwilling to grant sufficient access.

These were Chang's explanations of the failed ventures, rather than complete independent assessments of their economics.

He maintained that solar and LEDs were never intended to replace TSMC's core integrated-circuit manufacturing business.

Tsai eventually left the company and became CEO and vice chairman of MediaTek.

The $6 Billion Bet That Established TSMC's Advanced Manufacturing Position

Why Chang Protected Research Spending During Recessions

Chang's semiconductor career at Texas Instruments had convinced him that consistent research funding was essential.

When he led TI's worldwide semiconductor business, research spending represented approximately 4.8% of revenue. He wanted to increase the figure to 5.5%, but his requests were repeatedly rejected.

At TSMC, he established a different approach.

Previously, research budgets had generally fluctuated around 6–7% of revenue, with annual discussions between the CEO and research director.

Chang replaced that recurring negotiation with a commitment to allocate 8% of revenue to research, including during downturns.

The policy gave researchers more predictable resources. They could recruit engineers, plan projects, and pursue difficult technologies without repeatedly fearing abrupt cuts.

Chang recalled TSMC's R&D leadership describing the decision as one of the most valuable contributions he had made to the organization.

That funding policy became particularly important as the company approached 28-nanometer manufacturing.

The 28-Nanometer Sweet Spot

TSMC's researchers believed that 28 nanometers represented an unusually attractive technological opportunity.

They described it to Chang as a "sweet spot," borrowing an expression from tennis: the part of a racket that produces the best result when the ball strikes it.

The engineers presented technical reasons for accelerating development. Chang also considered market forecasts and strategic assessments from the company's business-development organization.

He was not simply accepting an engineer's prediction. He was combining expectations about technology with forecasts of how much manufacturing capacity customers would eventually require.

The conclusion was that TSMC should invest far more heavily.

For much of the preceding decade, the company had spent roughly $2–2.5 billion annually on fabrication facilities.

In 2010, that figure increased dramatically.

TSMC's 2010 annual shareholder letter confirms capital expenditure of $5.94 billion.

The same document records a regular R&D budget of approximately $940 million, up around 40%, plus $355 million in R&D-related capital expenditure. The investments supported work on 28-, 20-, and 14-nanometer technologies, alongside exploratory research on smaller nodes.

The report provides independent contemporary evidence of the scale of Chang's commitment.

Directors Challenged the Spending Plan

Research funding at 8% of revenue had attracted relatively little opposition. Tripling capital expenditure was another matter.

Chang recalled that more than three-quarters of TSMC's directors were independent, with many based in the United States and Britain.

Directors normally received board agendas approximately two weeks before meetings. Chang also invited independent directors to dinner the previous evening, allowing informal conversations that were not part of the formal meeting record.

The 28-nanometer proposal alarmed some directors enough that they contacted TSMC's general counsel, who also served as board secretary, before the scheduled discussion.

They wanted to challenge Chang directly.

He explained the research team's conclusions, market forecasts, and strategic case for expanding capacity.

The directors remained unconvinced.

Chang ultimately asserted that he was responsible for operating the company and needed authority to proceed.

They accepted his decision.

This was not a case of Chang identifying an obvious market opportunity that everyone else had missed. It was a major capital-allocation decision made under uncertainty, with substantial objections from people responsible for overseeing the company.

Smartphones Created the Demand

The investment coincided with rapid growth in smartphones.

Mobile processors required increasingly sophisticated manufacturing, and chip designers needed a foundry capable of producing them at scale.

TSMC's 28-nanometer capabilities helped it secure a much stronger competitive position.

Chang acknowledged that he had not personally foreseen the full scale of the smartphone market. He hoped that his business-development team had understood the opportunity more clearly, but the investment decision also depended on technical assessments and advice from other sources.

The company’s 2010 report recorded development progress at 28 nanometers and rapidly expanding customer engagements. It also reported that TSMC's revenue grew 48% in US dollars that year, compared with 43% for the overall foundry market.

Chang later identified 7 nanometers as another particularly important technological generation.

Both required a similar judgment: invest heavily before demand becomes certain, because waiting until customers urgently need capacity may leave insufficient time to construct facilities and refine manufacturing processes.

Apple: The Dinner That Changed TSMC's Business

Terry Gou Brings Jeff Williams to Chang's Home

Apple entered TSMC's story through an unusually personal connection.

After a board meeting in Taipei, Chang returned home to learn that Terry Gou planned to visit for dinner.

Gou, founder of Hon Hai Precision Industry, better known internationally as Foxconn, was related to Chang's wife Sophie. Chang described them as cousins.

Foxconn was already a major manufacturer of Apple products.

Gou intended to bring an Apple executive.

Chang immediately recognized the significance. Apple was exceptionally secretive, and he believed that suppliers generally could not obtain its business through ordinary sales approaches. They needed to be ready when Apple chose to make contact.

The visitor was Jeff Williams, a senior Apple operations executive.

Williams began discussing business almost immediately. Chang estimated that Williams accounted for approximately 80% of their substantive conversation.

The proposal was straightforward: Apple wanted TSMC to manufacture semiconductor wafers for its products.

Williams suggested an arrangement that would allow TSMC a gross margin of approximately 40%.

Chang was privately disappointed.

TSMC was already achieving roughly 45%, and he had set a long-term objective of reaching 50%. From Apple's perspective, the proposed margin may have seemed generous. From Chang's perspective, it would represent a step backward.

He decided not to argue about pricing during their first meeting.

There was a more difficult issue.

Apple Wanted 20 Nanometers, Not 28

TSMC was developing its 28-nanometer technology and expected to advance toward 16 nanometers.

Apple wanted an intermediate 20-nanometer process.

Chang regarded the request as a technological detour.

The company would have to redirect research resources toward an additional process generation. At that time, TSMC could not develop two major nodes simultaneously with the same ease that it later achieved.

There was also a financial complication.

The enormous 28-nanometer investment had been planned without including Apple as a customer. Its projected demand came from other market opportunities.

Apple's business was an unexpected addition, but serving it required billions of dollars in further spending.

The company faced a choice between pursuing one of the world's most valuable customers and preserving the financial resources needed for its existing technological plans.

Chang decided to pursue Apple, but not at any price.

Why Goldman Sachs Became Involved

Chang had maintained a longstanding relationship with Goldman Sachs, partly because he expected TSMC might eventually require sophisticated financial advice.

Goldman had been involved in the company's American depositary receipt arrangements, and Chang had served on the investment bank's board.

American depositary receipts, or ADRs, enable US investors to trade interests in foreign companies through American markets.

TSMC's Taiwanese shares and US-traded depositary securities operated in different markets. Chang noted that the ADRs carried a premium of approximately 20% around the interview period.

He also explained that converting ordinary Taiwanese shares into ADRs required board approval, limiting opportunities to exploit the price difference through arbitrage.

When Apple requested additional production capacity, TSMC considered several financing possibilities: reducing dividends, issuing new shares in Taiwan or the United States, or borrowing money.

Chang rejected dividend cuts and equity issuance.

After reviewing the numbers and consulting Goldman Sachs, the company decided to borrow.

That decision protected existing shareholders from dilution and preserved dividend payments, but it also added financial obligations.

Apple Requested More Capacity Than Chang Would Provide

Apple wanted TSMC to manufacture a very large volume of chips.

Chang accepted only half the requested quantity.

The reason was financial exposure. Manufacturing all the wafers Apple wanted would require a much larger investment, based on Apple's forecast of future demand.

If that demand failed to materialize, TSMC would be left with expensive capacity and insufficient revenue to support it.

C.C. Wei, then involved in TSMC's business development, communicated the proposal to Apple's purchasing employees.

According to Chang, their response was:

"You must be crazy."

Wei brought the objection back to Chang rather than trying to settle it himself.

Chang traveled to Cupertino to meet Williams personally.

He explained that TSMC had completed a prudent financial assessment and could responsibly serve only half the requested volume. Even that commitment would require issuing corporate bonds worth billions of dollars.

Williams suggested eliminating TSMC's dividend to free additional resources.

Chang refused.

His financial analysis indicated that approximately one-third of TSMC's shareholders attached substantial importance to dividend income. He feared that cancellation would provoke heavy selling and a significant decline in the stock price.

Williams did not prolong the disagreement.

Apple accepted the initial manufacturing commitment.

The agreement illustrates the difference between receiving an enormous customer order and accepting a financially responsible one. More business could improve manufacturing economies of scale, but excessive borrowing or underused capacity could threaten the entire company.

Intel Nearly Interrupted Apple's Deal With TSMC

Paul Otellini Approaches Tim Cook

In February 2011, Williams contacted Chang with unexpected news.

Intel's leadership had approached Tim Cook about manufacturing Apple's chips. Apple wanted to pause its discussions with TSMC for approximately two months while considering the proposal.

Intel was already an established Apple supplier because Mac computers used Intel processors.

In the 1990s, Chang had regarded Intel as a formidable technological force. Its manufacturing capabilities and market influence made it difficult for smaller competitors to challenge.

By 2011, his assessment had changed.

He believed TSMC had reached approximate technological parity with Intel in relevant areas, enjoyed stronger manufacturing execution, and had better relationships with foundry customers.

He was concerned about Apple's discussions but did not consider Intel unbeatable.

After approximately one month, Chang contacted Williams to arrange a visit.

Williams told him that Tim Cook would meet him instead.

Chang found that unusual. In his experience, executives commonly delegated meetings to subordinates. Williams had instead arranged a meeting with someone more senior.

What Tim Cook Told Morris Chang

Cook invited Chang to lunch at Apple. They collected food from the company cafeteria and returned to Cook's office.

Chang asked about Intel.

According to his recollection, Cook explained that Intel simply did not know how to operate as a foundry.

Chang regarded the response as sufficient reassurance.

The episode later attracted attention because it suggested Intel had an opportunity to manufacture Apple's mobile processors but failed to secure the work. A January 2025 report by Tom's Hardware documented Chang's account.

Cook's private explanation is known here through Chang's recollection, not an independently released Apple record.

The Difference Between a Chipmaker and a Foundry

The Acquired hosts suggested that Cook might have been referring to Intel's structural conflict: it designed and sold processors while also potentially manufacturing chips for competing designers.

Chang disagreed with that interpretation.

Although customer neutrality was important, he believed Cook was referring primarily to Intel's attitude toward customers.

Intel had historically been a dominant supplier. Chang recalled hearing dissatisfaction from Taiwanese PC manufacturers that relied on Intel but disliked its approach to customer relationships.

A dedicated foundry needed different habits.

Customers could be demanding, insist on unusual requirements, and request changes that complicated production planning. The manufacturer's job was to accommodate their needs where practical while maintaining technical and financial discipline.

Chang believed TSMC understood that relationship better.

The distinction mattered because technical superiority alone did not guarantee business. A company could possess advanced factories yet fail to attract customers if those customers doubted its willingness to serve them.

Apple’s 20-Nanometer Decision Gives Samsung a Temporary Lead

The Hidden Cost of Winning Apple

Apple's business created a major technological trade-off.

TSMC devoted research resources to developing the requested 20-nanometer process. That delayed progress toward 16 nanometers, which Chang had regarded as the more natural successor to 28.

Samsung faced a different development path. Without the same Apple-driven detour, it was able to advance more quickly on the next process generation.

Samsung reached the relevant manufacturing capability before TSMC.

Apple then placed its initial 16-nanometer orders with Samsung.

Chang was shocked.

TSMC had already committed extraordinary capital to Apple's manufacturing requirements. Its financial planning assumed that much of the equipment used for 20-nanometer production could support later 16-nanometer manufacturing.

Chang estimated that approximately 80–90% of the relevant equipment could be converted.

If Apple moved its next-generation business permanently to Samsung, TSMC could lose a major part of the future production required to justify its earlier investment.

Jeff Williams Promises Future Orders

Chang contacted Williams directly and explained why the decision was so troubling.

Williams replied and traveled to meet him.

He assured Chang that Apple would buy its 16-nanometer requirements from TSMC when TSMC's technology was ready.

Approximately six months later, TSMC completed development and recovered most of Apple's business.

Chang credited Williams with honoring the commitment.

The incident revealed a risk that was not immediately visible when TSMC first negotiated with Apple.

A customer could demand a special manufacturing process, require substantial investment, and then shift future orders to a rival whose technology advanced more quickly.

Even strong relationships could not eliminate competitive pressure. Apple needed chips when its products were ready, not when TSMC would have preferred to deliver them.

Negotiating the Price of Apple's Chips

Pricing remained another central issue.

Williams visited TSMC and explained Apple's component-cost targets. TSMC presented its manufacturing costs and acceptable margins.

Yield, equipment utilization, process complexity, and production volume all influenced the economics, but the final price still needed to satisfy both organizations.

Chang regarded this as an exercise in commercial judgment rather than a universal mathematical formula.

Charging the highest possible price could damage a valuable long-term relationship. Charging too little could prevent the foundry from recovering its enormous investment.

The two sides eventually agreed.

They celebrated at an expensive Taipei restaurant, prompting Williams to joke that an unfavorable agreement might have resulted in dinner at McDonald's instead.

Chang was still pursuing TSMC's long-term 50% gross-margin objective at the time and had not reached it before retiring. He attributed the company's subsequent margin expansion partly to growing technological leadership and favorable market conditions around the pandemic period.

By 2025, TSMC's annual report recorded a gross margin of 59.9%, compared with 56.1% in 2024.

That was substantially above the 40% margin Williams had suggested during their first dinner.

How TSMC Managed Customers, Deposits and Production Risk

Why Customers Had to Reserve Capacity With Deposits

Apple was not the first customer to request more manufacturing capacity than TSMC could comfortably provide.

During the company's first 10–15 years, production capacity was frequently scarce.

Customers had incentives to reserve large quantities of wafers. If demand turned out to be lower than expected, the customer might cancel orders, leaving the manufacturer with idle equipment.

The customer faced relatively little risk from making an optimistic forecast. TSMC faced the financial consequences of constructing the factory.

Chang addressed the problem through deposits.

He instructed salespeople to warn customers that reserved capacity required advance payments and that deposits could be forfeited if customers failed to take their wafers.

The policy changed the negotiation. Customers now had a financial reason to forecast their needs more carefully.

TSMC Never Actually Confiscated the Deposits

Despite the warning, Chang maintained that TSMC never confiscated customer deposits.

The policy was tested during the dot-com downturn, when semiconductor demand weakened sharply.

Chang dated the decline from the first quarter of 2001 through recovery around the third quarter of 2003.

TSMC had expanded manufacturing capacity, including through acquisitions of companies with existing fabrication plants. Customers had reserved production in anticipation of continued growth.

When demand collapsed, some no longer needed their scheduled wafers.

Instead of confiscating their money, TSMC allowed them to postpone orders and use their deposits later.

Chang said the customers eventually used the credits.

The strategy combined strict contractual expectations with flexibility when circumstances changed. Deposits discouraged speculative reservations, while allowing delays protected customer relationships.

It was another example of the financial judgment required in a business where manufacturers invest years before knowing the exact demand for their output.

Why Morris Chang Reorganized TSMC's Leadership

The Disagreement With Don Brooks and McKinsey

Chang believed TSMC should organize its business around shared functions rather than divide operations into largely independent manufacturing units.

The question arose as early as 1996, when company president Don Brooks proposed a divisional structure resembling organizations Chang had encountered at Texas Instruments.

Chang resisted.

The same customers frequently needed chips manufactured using several technological generations. A company might purchase leading-edge processors alongside older-node components for cameras, communications, or other functions.

Dividing TSMC by fabrication facility could therefore fragment relationships with customers whose business crossed factory boundaries.

Brooks continued advocating decentralization, so TSMC hired McKinsey to investigate.

After approximately two months and what Chang remembered as a multimillion-dollar consulting engagement, McKinsey supported a functional organization.

Its consultants used Boeing as an example. Aircraft manufacturers might divide operations into major businesses, but dividing the company by individual aircraft models would ignore shared engineering capabilities and customers.

Chang believed the same principle applied to TSMC's factories.

Rick Tsai's Advanced and Mainstream Divisions

During his 2005–2009 tenure, Rick Tsai nevertheless divided TSMC's manufacturing operations into advanced and mainstream technology groups.

Chang allowed the experiment.

He believed a chairman should normally permit the CEO to exercise independent judgment and make manageable mistakes. Intervention became necessary only when decisions threatened serious damage.

Mark Liu led advanced technology, overseeing approximately 10,000 employees. C.C. Wei led mainstream technology, responsible for roughly 7,000–8,000 people.

Both groups contained relatively small business-development teams, each with approximately 30–40 employees.

Chang considered that arrangement inefficient because customers used both advanced and mature-node manufacturing.

He eventually recombined operations and sought stronger centralized business development.

Why Mark Liu Declined and C.C. Wei Accepted

Chang first offered Mark Liu leadership of the unified business-development organization.

He believed Liu needed more commercial experience to become a well-rounded potential chief executive.

Liu declined.

The proposed organization contained approximately 60–70 employees, compared with the roughly 10,000 people under his existing responsibility. Liu regarded the change as a substantial reduction in organizational scope.

Chang disagreed with judging a position by headcount alone.

He compared the role's potential influence to that of Henry Kissinger as US national security adviser, a position that could wield considerable power without supervising a large department.

C.C. Wei accepted the assignment.

It gave Wei experience identifying future markets and customers rather than concentrating exclusively on manufacturing operations.

Wei subsequently rose through TSMC's leadership and became chairman and CEO.

The episode illustrated Chang's distinction between managing today's production and deciding which customers and technologies would matter years later.

Qualcomm, IBM and the Importance of Technological Independence

Qualcomm's Business Moves Away From IBM

Throughout much of the 1990s, TSMC tried to win Qualcomm as a major manufacturing customer.

Chang considered Qualcomm technologically sophisticated and commercially attractive.

He met senior Qualcomm executives repeatedly but initially secured little business. Qualcomm relied primarily on another manufacturing partner, which Chang understood to be IBM.

Around 1997–1998, that began to change.

Qualcomm transferred increasing manufacturing volume to TSMC.

Chang interpreted the movement as evidence that IBM's semiconductor operations were encountering difficulties.

He believed IBM relied heavily on Qualcomm and a relatively small number of other fabless customers. Losing significant business from one could undermine the economics of its manufacturing facilities.

The changing orders provided information about IBM's competitive position without IBM directly disclosing its problems.

IBM Proposes Joint Development of 130-Nanometer Technology

Around 1999, IBM approached TSMC about collaborating on 130-nanometer manufacturing technology.

Chang rejected the proposal immediately.

He believed IBM would expect to dominate the partnership, potentially placing TSMC engineers inside an IBM-controlled development organization.

The concern was not simply disagreement over ownership or prestige. Semiconductor process technology determined which chips a foundry could manufacture, how efficiently it could produce them, and when it could advance to the next generation.

Losing independent control over that technology could compromise TSMC's entire competitive strategy.

Chang also feared organizational and cultural conflicts between the companies.

According to his account, IBM reacted angrily and subsequently approached United Microelectronics Corporation, or UMC, another Taiwanese semiconductor manufacturer.

UMC accepted a partnership that Chang believed it later regretted.

The episode reinforced a principle that appeared throughout his interview: TSMC needed relationships with customers and suppliers, but could not allow another company to control its manufacturing roadmap.

The Economics That Made TSMC So Difficult to Challenge

Morris Chang, Bruce Henderson and Bill Bain

Chang's understanding of semiconductor competition was strongly influenced by his earlier experience at Texas Instruments.

Around 1970, Boston Consulting Group founder Bruce Henderson contacted TI executive Mark Shepherd about the experience curve, sometimes called the learning curve.

The principle is that unit costs tend to decline as organizations accumulate manufacturing experience.

Henderson and Bill Bain presented the concept to TI's management. Shepherd subsequently assigned Chang to work closely with Bain.

Bain received an office near Chang and spent approximately three days each week in Dallas. For roughly two years, they examined manufacturing costs, prices, production volumes, and semiconductor product families.

The collaboration shaped Chang's approach to business strategy.

He did not claim to have invented the learning curve. He credited Henderson and Bain with developing the framework and remembered its influence throughout his career.

Bain later privately informed Chang that he intended to leave BCG and establish his own consulting firm.

Chang recalled Bain describing the venture as a personal "world imperative," an expression that stayed with him.

Why Manufacturing Volume Changes Everything

Semiconductor factories require enormous upfront investment.

Equipment, buildings, and process-development programs must be financed before manufacturers can produce chips at commercial scale.

Once production increases, costs can decline through several mechanisms.

Fixed costs are distributed across more wafers. Engineers improve manufacturing procedures. Equipment utilization increases. Yields can improve as experience accumulates.

Lower costs make the foundry more competitive, potentially attracting further customers and increasing production volume again.

This creates a reinforcing cycle.

The Acquired hosts connected the learning curve to aggressive pricing. A manufacturer seeking long-term leadership might accept relatively low initial margins because it expects production costs to decline as volume increases.

Such a strategy can work only if the company's predictions about future demand are reasonably accurate.

A manufacturer must therefore estimate the ultimate market for a process generation before committing the capital required to serve it.

Why TSMC Must Forecast iPhone and AI Demand

TSMC's customers do not simply purchase a generic manufacturing service.

Apple requires chips because consumers buy iPhones, computers, and other devices. Nvidia needs manufacturing capacity because customers purchase graphics processors and AI computing systems.

TSMC must consequently forecast demand for products it does not design or sell.

Those forecasts guide plant construction, equipment purchases, debt financing, and the timing of process development.

The hosts suggested that even a 5–10% error in demand projections could materially damage profit margins and cash flow, particularly where investment commitments were enormous.

That was their analytical assessment, not a quantified finding from TSMC's financial statements.

The consequences could extend beyond one process generation. If a foundry generated less cash than expected, it could struggle to fund the next technological advance.

Why Chang Did Not Always Pursue Maximum Volume

The learning curve encourages manufacturers to seek high production volumes, but Chang's decisions showed that volume was not his only objective.

Accepting all Apple's requested capacity could have accelerated TSMC's movement down the cost curve.

It would also have exposed the company to considerably greater financial risk.

Chang chose half.

Similarly, he was willing to spend aggressively on 28 nanometers because he believed the technological and market evidence justified the investment.

The difference was his judgment about the risks attached to each opportunity.

Learning-curve advantages were valuable, but they did not justify betting the company's survival on every large customer request.

The Foundry Model: How TSMC Separated Chip Design From Manufacturing

Gordon Campbell's $50 Million Business Plan

Chang founded TSMC in 1987, but his understanding of the emerging fabless semiconductor industry developed earlier.

One important encounter involved entrepreneur Gordon Campbell, who visited Chang during his final period at General Instrument.

Campbell wanted approximately $50 million to establish a semiconductor company.

Chang requested a written business plan.

When Campbell returned, the proposed funding requirement had fallen from $50 million to $5 million.

The reason was simple: Campbell had decided not to build a semiconductor fabrication plant.

Instead, his company would concentrate on chip design and contract with someone else to manufacture the designs.

The change dramatically reduced the capital needed to enter the semiconductor industry.

For Chang, it demonstrated the commercial potential of separating design and production.

The Problem of Manufacturing Chips for Other Companies

Another early fabless startup approached General Instrument seeking access to unused factory capacity.

The arrangement produced an organizational conflict.

The chip designer wanted manufacturing procedures adapted to its requirements. General Instrument's factory manager wanted to retain control of production.

Neither position was entirely unreasonable.

Designers needed processes suited to their chips. Factory operators needed consistent procedures to maintain yields, quality, and efficiency.

Chang recognized that a dedicated foundry would have to resolve this conflict repeatedly.

It needed sufficiently standardized manufacturing systems to serve many customers while remaining responsive to different designs and technical requirements.

The difficulty also created the business opportunity. By combining demand from many independent designers, a specialized manufacturer could achieve production volumes beyond what individual startups could support.

Why TSMC Did Not Need Its Own Chips

TSMC's original strategy was to manufacture semiconductors designed by customers rather than produce competing branded chips.

The decision appeared in the company's earliest business presentations, which the Acquired hosts later examined at TSMC's museum.

Chang founded the business at approximately 56 years old.

An early business plan reportedly contemplated only two fabrication plants, but Chang rejected the suggestion that he intended TSMC to remain permanently limited to that scale.

His knowledge of manufacturing economics made such an outcome implausible.

The initial business was modest. Established semiconductor companies often outsourced secondary production or used TSMC as an additional manufacturing source.

These orders were less attractive than the advanced processors that later became central to the company's success, but they generated revenue and manufacturing experience.

Over time, the number and sophistication of fabless designers increased.

Nvidia, Qualcomm, AMD, Apple, and others demonstrated that a company could develop valuable semiconductor products without owning the most advanced factories.

TSMC supplied the industrial capability that made such specialization practical.

Arm, Intel and the Wider Semiconductor Ecosystem

The industry's reorganization involved far more than TSMC.

Arm developed processor architectures that companies could license and incorporate into their own designs.

Synopsys and Cadence provided electronic design automation software. ASML developed essential semiconductor manufacturing equipment. TSMC supplied fabrication services.

Several of these businesses originated or took shape during the 1980s, when the semiconductor industry was moving toward greater specialization.

The hosts argued that Arm's rise was particularly important because it helped create a wider population of independent chip designers.

Intel's x86 architecture historically supported a more concentrated market, although AMD later demonstrated that x86 processors could also be manufactured through a fabless model.

Apple's transition toward Arm-based processors further strengthened the independent manufacturing ecosystem.

The hosts considered alternative historical outcomes possible. Had Intel maintained stronger control over mobile computing, the semiconductor industry might have remained more vertically integrated.

That interpretation is a counterfactual argument rather than an established historical certainty.

What the record does demonstrate is that specialized architecture providers, software companies, equipment suppliers, and foundries became essential parts of a global production system.

TSMC did not need to replace those businesses. It needed to cooperate with them.

Taiwan's Hsinchu Science Park: More Than a Collection of Factories

How the Semiconductor Cluster Developed

During their Taiwan visit, the Acquired hosts toured TSMC's headquarters and manufacturing environment in Hsinchu Science Park.

They encountered an unusually concentrated collection of semiconductor companies.

Businesses associated with Arm, Synopsys, Cadence, Qualcomm, MediaTek, and other industry specialists operated within the wider ecosystem.

The concentration mattered because semiconductor design and manufacturing require continual technical coordination.

Design-software providers need information about manufacturing processes. Chip designers need to understand what foundries can produce. Equipment manufacturers need to develop machines capable of performing increasingly demanding operations.

Geographic proximity makes frequent collaboration easier.

Nearby universities supplied engineering graduates and advanced researchers. Government-backed industrial development helped create the infrastructure in which those companies could grow.

The hosts described Hsinchu as an exceptionally successful example of building an industrial cluster through coordinated public and private investment.

They also observed semiconductor-industry workers traveling to Taiwan from international technology companies, reinforcing the country's role as a meeting point for a global supply chain.

Why Arizona Cannot Immediately Reproduce Hsinchu

The hosts questioned whether other countries could quickly replicate the advantages accumulated in Taiwan.

A modern fabrication facility can be constructed overseas, but its surrounding ecosystem is not contained within the building.

Factories require skilled engineers, specialist suppliers, technical support, research institutions, experienced managers, and relationships developed over many years.

TSMC's Taiwanese manufacturing infrastructure also extends beyond Hsinchu into other science parks, including major facilities in southern Taiwan.

During the visit, the hosts saw an enormous 2-nanometer manufacturing complex under development.

They expected the technology to expand during 2025, speculating about future demand from companies such as Apple and Nvidia.

Subsequent company records provide a firmer account of what happened.

TSMC's 2025 annual report states that 2-nanometer technology entered high-volume manufacturing in the fourth quarter of 2025.

It also confirms that the company's first Arizona facility entered high-volume production in the fourth quarter of 2024, while manufacturing expansion continued in Arizona, Japan, Germany, and Taiwan.

The international expansion demonstrates that TSMC can operate beyond Taiwan. It does not establish that every overseas site has reproduced the full economics and supplier density of Hsinchu.

Mature Factories Remain Valuable

TSMC's advantage also extends beyond its newest process technologies.

The hosts visited or discussed early facilities associated with the company's founding, including Fab 1, which originated in Taiwan's Industrial Technology Research Institute.

Fabs 2 and 3 were described as early TSMC manufacturing facilities dating from the late 1980s.

Rather than replacing all older factories with advanced production lines, TSMC historically retained significant mature-node manufacturing capacity.

Older technologies still serve valuable markets.

Automobiles require control chips that do not necessarily benefit from leading-edge processes. Industrial equipment may need compatible replacement components decades after the original machinery was manufactured.

Camera sensors, microcontrollers, power-management devices, and communications components can also use specialized mature processes.

A factory whose original investment has been substantially depreciated may remain economically attractive because continuing production requires comparatively less capital.

TSMC's 2010 annual report illustrates the breadth of these technologies. Alongside 28-nanometer development, it described older processes supporting automotive engine controls, image sensors, embedded memory, microcontrollers, and power electronics.

The hosts contrasted this strategy with Intel's historical tendency to upgrade facilities toward newer generations.

Neither approach was presented as universally correct. TSMC benefited from serving both advanced and established technologies, often for the same customers.

Advanced Packaging: The Next Source of Manufacturing Advantage

Why Manufacturing a Chip Is No Longer the Whole Job

As conventional transistor scaling becomes increasingly difficult, semiconductor performance depends on more than shrinking individual features.

Advanced packaging connects multiple chips or components into a larger system.

This is especially important for artificial-intelligence processors, which require high-speed communication between computing chips and high-bandwidth memory.

TSMC developed technologies including CoWoS, or chip-on-wafer-on-substrate, to support these requirements.

The hosts identified CoWoS as an important competitive advantage. They also discussed specialized packaging for mobile processors.

These technologies can increase customer dependence on TSMC.

Moving a design to another manufacturer may require adapting not only the fabrication process but also the packaging system and its electrical connections.

The company's 2025 report identifies CoWoS, InFO, SoIC, and other technologies as parts of its expanding advanced-packaging portfolio.

The October 2026 GlobalFoundries Agreement

A subsequent development shows how important advanced packaging has become.

On October 8, 2026, GlobalFoundries announced a multiyear $2 billion manufacturing agreement with TSMC.

Under the arrangement, GlobalFoundries will manufacture silicon interposers at its Malta, New York facility for TSMC's CoWoS ecosystem.

Interposers are components that provide dense electrical connections between processors, memory, and other chips within advanced packages.

The agreement is intended to expand manufacturing capacity for AI and high-performance computing applications, with initial production expected to ramp during the first half of 2028.

The relationship is notable because GlobalFoundries is itself a semiconductor manufacturer.

It demonstrates how the industry's specialization continues evolving. A leading foundry can now rely on another foundry for particular components supporting its own advanced-packaging services.

The competitive advantage lies increasingly in coordinating the entire manufacturing system, not simply operating the newest fabrication process.

What TSMC's 2026 Financial Results Show About Chang's Strategy

From $6 Billion in Capital Spending to $60–64 Billion

The financial scale of TSMC's business has changed dramatically since Chang's 2010 investment.

The $5.94 billion capital expenditure that concerned directors then is small compared with the company's current requirements.

Following its third-quarter results on October 15, 2026, TSMC indicated planned annual capital expenditure of $60–64 billion, according to Associated Press reporting.

That represents approximately ten times the amount spent in 2010, without adjusting for inflation.

The increase reflects the growing cost of advanced fabrication, larger production requirements, and investment in advanced packaging.

The Acquired hosts had estimated that a modern advanced fab might cost approximately $20 billion and speculated that future facilities could eventually cost $40 billion, $80 billion, or even $100 billion.

Those larger amounts were projections, not confirmed construction costs.

Nevertheless, TSMC's actual capital expenditure shows why manufacturing leadership is increasingly difficult for smaller competitors to challenge.

The financial resources required are enormous even before accounting for ongoing research and process development.

Revenue, Profit Margins and Customer Scale

TSMC's audited 2025 results provide a picture of the business that emerged from Chang's model.

According to its 2025 annual report, the company recorded:

These figures demonstrate that TSMC serves a much broader market than any single major customer.

Its advanced manufacturing revenue is substantial, but hundreds of customers also use an extensive range of specialty and mature technologies.

The company's financial performance strengthened further in 2026.

Its official third-quarter results reported revenue of $46.72 billion and a gross margin of 66.4%.

For the fourth quarter of 2026, TSMC projected revenue between $52.3 billion and $54.3 billion.

Those figures also place the Apple negotiations of 2010–2011 in perspective.

Chang once rejected a proposed 40% gross margin while trying to reach 50%. By 2026, the company was reporting margins well above that original goal.

This does not mean every product or customer generates the same margin. Consolidated gross margin reflects the company's combined business, including its technology mix, pricing, costs, and utilization.

The Planned $265 Billion US Expansion

TSMC's international investment plans have also expanded.

In March 2025, the company announced that its planned US investment would increase from $65 billion to $165 billion, adding proposed advanced manufacturing plants, packaging facilities, and research capabilities.

On July 16, 2026, another $100 billion expansion was announced, bringing planned US investment to $265 billion.

The US National Institute of Standards and Technology announcement states that the plan includes additional advanced semiconductor manufacturing and packaging facilities in Arizona.

TSMC chairman and CEO C.C. Wei described the investment as a response to strong multiyear demand from major American customers.

The figure represents planned investment rather than expenditure already completed.

The scale nevertheless illustrates how far the industry has moved from Chang's early concern about funding just half of Apple's requested manufacturing capacity.

The company now needs to coordinate enormous investments across multiple countries while retaining the technical strengths and operational discipline associated with its Taiwanese facilities.

Moore's Law, Computing Demand and the Limits of TSMC's Advantage

The Industry Grew More Than Twentyfold

The semiconductor industry provided an unusually favorable environment for TSMC's growth.

The Acquired hosts cited an increase in global semiconductor revenue from approximately $26 billion in 1987, when TSMC was founded, to $527 billion in 2024.

The precise market total depends on the industry definition and reporting series, but the central development is clear: semiconductor demand expanded enormously during TSMC's existence.

Computers became more powerful, mobile devices became widespread, and internet infrastructure expanded. More recently, artificial intelligence created substantial demand for advanced processors and memory systems.

Moore's law helped shape the industry's development by describing the historical growth in transistor density.

However, its original technical formulation should not be confused with the broader demand for computing power.

The hosts acknowledged that conventional transistor scaling faces physical and economic constraints.

They also argued that innovations in processor architecture, advanced packaging, interconnects, and software can continue improving computing capability even when shrinking transistors becomes more difficult.

Growing demand creates incentives to pursue those alternatives.

Why Manufacturing Scale Creates a Competitive Cycle

TSMC's position reflects the interaction of several advantages.

Greater manufacturing volume can improve costs and yields. Stronger financial performance supports research and factory construction. Better technology attracts additional customers, who provide the volume required for the next investment cycle.

The effect can become self-reinforcing.

The hosts described advanced semiconductor manufacturing as a market with winner-take-most characteristics, where the leading company's scale can become increasingly difficult to overcome.

That argument is supported by TSMC's enormous capital expenditure and broad customer base, although it does not establish that the company is immune to competition.

Samsung and Intel continue pursuing semiconductor manufacturing capabilities. Customers can change suppliers when rivals develop better technology, as Apple's temporary move to Samsung demonstrated.

A foundry's advantage also depends on decisions that cannot be solved solely by spending money: which technologies to develop, how much capacity to construct, what prices to charge, and which commercial risks to accept.

The Business Morris Chang Actually Built

Chang rejected the suggestion that TSMC's eventual success had been entirely unexpected to him.

He believed the company could become substantial from the beginning, although its eventual scale exceeded his original expectations.

His early experience with fabless semiconductor designers, his understanding of learning curves, and his commitment to manufacturing specialization all shaped that belief.

Yet the detailed history makes clear that growth was not automatic.

Nvidia nearly went bankrupt before becoming a major customer. Manufacturing failures threatened its relationship with TSMC. The financial crisis produced controversial layoffs and a leadership reversal. Directors questioned the 28-nanometer investment. Apple demanded additional technology and capacity, and Samsung temporarily captured its subsequent orders.

Several decisions involved accepting short-term costs or turning down apparently attractive business to protect the company's long-term position.

Chang compensated Nvidia rather than allowing the conflict to proceed unresolved. He opposed dismissing workers who might soon be needed again. He committed heavily to research and manufacturing technology while refusing to accept all Apple's requested capacity. He protected TSMC's independence when IBM proposed a technological partnership.

Those decisions were not identical, but they followed a recognizable commercial approach: invest aggressively where the long-term opportunity justified it, preserve the capabilities needed to compete, and avoid commitments that could compromise the business.

By October 2026, TSMC was operating at a scale far beyond the original two-factory plan. Its customers spanned hundreds of companies, its manufacturing technologies covered multiple generations, and its expansion plans involved hundreds of billions of dollars.

The business had become much more than a manufacturer taking orders from chip designers.

It was a company required to understand its customers' future markets, develop technologies years in advance, finance extraordinarily expensive facilities, and coordinate a global network of specialized businesses.

That is the central explanation emerging from Chang's account: TSMC did not merely benefit from the rise of Nvidia, Apple, smartphones, and artificial intelligence. It built the manufacturing capabilities those industries needed before much of their eventual demand was certain.