TSMC Founder Morris Chang

2025-01-27 - source - Read full transcript

Ben Gilbert (host)David Rosenthal (host)Morris Chang

Key insights

TSMC's decision to never compete with its customers by staying a pure-play foundry is the single biggest explanation for its dominance.

Chang's original pitch deck to the Taiwanese government and investors explicitly named 'dedicated pure play foundry' as the strategy. Because TSMC never designs, markets, or sells its own chips, it can serve every fabless and integrated design house without conflict - a structural advantage no integrated device manufacturer, including Intel, can match.

pure-play-foundry-strategy

Learning-curve (experience-curve) economics, learned at Texas Instruments from BCG's Bruce Henderson and Bill Bain, is the pricing logic behind TSMC's dominance.

The theory says the largest-volume player ends up with the lowest cost, so the winning strategy is to price ahead of your current cost curve (sometimes below cost early on) to win volume fastest, crowd out competitors, and reach scale economics before anyone else. Chang applied this deliberately at every TSMC node.

learning-curve-economics

The rise of the ARM architecture created the structural opening for the fabless/foundry industry split that TSMC needed to exist.

Had x86 remained dominant, there would have been no independent chip-design ecosystem for a pure-play foundry to serve - Intel would likely have stayed fully vertically integrated and made the leading-edge chips for phones itself. ARM, TSMC, Synopsys, Cadence, and ASML were all founded within a few years of each other in the mid-to-late 1980s, and the hosts argue this cluster of near-simultaneous founding is not a coincidence.

pure-play-foundry-strategy

Chang personally resolved a 2009-2010 quality/yield dispute with NVIDIA by negotiating a settlement over pizza and salad at Jensen Huang's house rather than escalating to arbitration.

After retaking the CEO role, Chang spent weeks building a fair settlement number (over $100 million), delivered it to Huang with a 48-hour take-it-or-leave-it deadline to avoid drawn-out arbitration, and Huang accepted within two days. Chang frames the personal relationship (built over many prior casual dinners) as what made a fast, non-adversarial resolution possible.

key-account-management

Chang refused to base layoffs on individual performance ratings, arguing subjective ratings destroy trust and rehiring costs (severance plus ~6 months retraining) usually exceed any savings within a year.

He first took this position at TI in the early 1970s and enforced it at TSMC: the company's practice was six-month probation rather than firing, and it almost never actually terminated employees even after probation. When a TSMC CEO circumvented this by using 'worst performance review' to lay off 600-700 people during the 2008 financial crisis, protestors gathered outside Chang's home, and the episode was a direct trigger for Chang retaking the CEO role.

leadership-philosophy

Fixing TSMC's R&D budget at a flat 8% of revenue, regardless of recession or business cycle, was one of the highest-leverage decisions Chang made as CEO.

Before this, the R&D director had to renegotiate the budget every year with the CEO; setting it as an automatic 8% removed that friction, gave R&D multi-year planning certainty, and (per TSMC's own longtime R&D leadership) was repeatedly cited as one of the best things done for the organization.

capital-allocation

TSMC's roughly threefold increase in capex around the 28nm node (from ~$2-2.5B to ~$6B per year) was a bet-the-company move that the board initially resisted.

R&D and the newly consolidated business-development group told Chang 28nm would be the industry 'sweet spot'; Chang backed the bet against board pushback, quoting Shakespeare's 'tide in the affairs of men.' The bet paid off as 28nm coincided with the smartphone boom, and TSMC repeated the same pattern at 7nm.

capital-allocation

TSMC deliberately committed to only half of Apple's initial 20nm demand and financed the buildout with new debt rather than cutting the dividend or issuing new stock.

Apple COO Jeff Williams suggested eliminating the dividend to fund the capacity; Chang refused because roughly a third of TSMC's shareholders were dividend-focused and cutting it would have tanked the stock. Taking half of Apple's stated demand, rather than the full amount, was a deliberate hedge against the risk that a single customer's own forecast could be wrong - which is exactly what nearly happened when Apple briefly placed its first 16nm orders with Samsung.

capital-allocation

Chang anticipated the rise of fabless semiconductor companies years before TSMC's founding, from direct encounters with early fabless founders.

While at General Instrument, Gordon Campbell told Chang he only needed $5 million (not the originally requested $50 million) because 'I'm not going to build a fab,' and a separate founder pitched what became Atmel on the same premise. These encounters, alongside Chang's later relationship with Don Valentine, convinced him fabless would become a real category well before TSMC's founding, though he says the timing (about within 12 months of his own estimate) still exceeded his expectations.

pure-play-foundry-strategy

TSMC declined IBM's offer to co-develop the 130nm process in 1999 because IBM's co-development structure would have made TSMC subordinate to IBM's own roadmap.

IBM, having just lost Qualcomm's business to TSMC, wanted TSMC to send engineers into an IBM-led co-development arrangement; Chang refused because IBM 'always considered themselves the senior partner' and TSMC would have lost the ability to develop its own independent process technology. IBM instead partnered with rival Taiwanese foundry UMC, which came to regret the arrangement.

pure-play-foundry-strategy

Apple briefly defected to Samsung for its first 16nm orders after TSMC's Apple-driven 20nm detour delayed TSMC's own 16nm roadmap.

TSMC's R&D lacked the resources to develop both the 20nm node Apple requested and the originally planned 16nm node simultaneously; Samsung, having lost the 20nm business entirely, skipped straight to 16nm and won Apple's first 16nm orders. Chang describes this as a genuine shock after TSMC had committed tens of billions of dollars expecting the equipment to convert to 16nm for Apple; the business reverted to TSMC once its own 16nm was ready about six months later.

key-account-management

Early in TSMC's history, customer capacity deposits (with the rhetorical threat to 'confiscate' them) were used to align customer demand commitments with the capital risk TSMC was taking on their behalf.

During capacity-constrained years, TSMC required deposits before building capacity for a customer, and salespeople were told to warn customers the deposit could be confiscated if they didn't take the wafers - a threat Chang says was never actually carried out, including through the multi-year dot-com downturn when several customers simply let demand slip rather than forfeiting deposits.

key-account-management

Books referenced

Morris Chang autobiography, Volume 2 - Morris Chang - Published a couple of months before this episode after a 26-year hiatus from volume one; only in traditional Chinese, the hosts prepped from an unpublished translation by Karina Bao (funded by Tyler Cowen and Emergent Ventures).
Competitive Strategy (and related works) - Michael Porter - Chang mentions owning all of Porter's ~700-page books on competitive strategy; Porter was later a TSMC board director.

Media referenced

TSMC (original episode) - podcast - Acquired's original TSMC episode four years earlier is what got Ben and David obsessed with semiconductors and led to this interview.
TSMC Remastered - podcast - Released the week before this episode; referenced as required listening.
NVIDIA episodes (four, including a Jensen Huang interview) - podcast - Cross-referenced as the rest of the semiconductor-industry arc Acquired has covered.
Qualcomm (live episode) - podcast - Referenced as a 'sleeper pick' covering Irwin Jacobs and how CDMA works.
ACQ2 episode with Rene Haas (ARM) - podcast - Most recent ACQ2 episode at time of release; Rene Haas helped the hosts prep questions for Chang.
ACQ2 episode with Sassine Ghazi (Synopsys) - podcast - Co-founder Aart de Geus and CEO Sassine Ghazi also helped prep questions for this interview.
Defunct Land - other - Ben's carve-out: a YouTube channel about the history of failed/closed theme parks, recommended as an Acquired-adjacent visual history series.
Asianometry - other - YouTube channel about semiconductors; its creator Jon hosted Ben and David for dinner in Taipei the night before the interview.
Culpium (Substack) - article - Tim Culpan's newsletter, cited as a source of prep topics for the interview.
Everything Everywhere All at Once - movie - David's carve-out, watched for the first time on the flight to Taipei.

Companies

TSMC - Subject of the episode; the pure-play semiconductor foundry Morris Chang founded in 1987.
NVIDIA - Became a TSMC customer in 1997 after Jensen Huang cold-emailed Chang directly; grew into one of TSMC's top five customers.
Apple - Became a TSMC customer starting around 2010-2011 via COO Jeff Williams and CEO Tim Cook; TSMC took on debt to fund a bet-the-company capacity buildout for Apple's chips.
Intel - TSMC's chief rival as an integrated device manufacturer; briefly courted by Apple for the iPhone business before Apple stuck with TSMC.
Qualcomm - Early fabless-industry customer; TSMC won its business away from IBM in the late 1990s.
IBM - Lost Qualcomm's business to TSMC, then asked TSMC to co-develop 130nm technology; TSMC declined and IBM took the offer to UMC instead.
Samsung - Skipped straight to the 16nm node after losing Apple's 20nm business, briefly winning Apple's first 16nm orders before TSMC's own 16nm was ready.
UMC - Rival Taiwanese foundry that accepted IBM's co-development offer that TSMC had declined, and later regretted it.
Texas Instruments - Chang's employer before TSMC; where he ran worldwide semiconductors and first learned learning-curve theory from BCG.
Foxconn (Hon Hai) - Terry Gou, Foxconn's founder and a relative of Chang's wife Sophie, brought Apple COO Jeff Williams to Chang's home for the dinner that opened the Apple relationship.
MediaTek - The TSMC executive Chang put in charge of the failed solar/LED new-businesses unit later became MediaTek's vice-chairman and CEO.
Goldman Sachs - Long-cultivated relationship (Chang sat on Goldman's board) that TSMC leaned on for advice when financing the Apple capacity buildout via debt.
Boston Consulting Group - Bruce Henderson brought learning-curve/experience-curve theory to TI in the early 1970s, where Chang absorbed it.
Bain & Company - Bill Bain worked at BCG under Bruce Henderson on the TI learning-curve engagement before leaving to found his own consulting firm.
ARM - Discussed as the architectural enabler of the fabless/foundry industry split; without ARM as an x86 alternative, the pure-play foundry model would have had no independent chip-design ecosystem to serve.
Synopsys - EDA company founded in the same mid-to-late-1980s wave as TSMC and ARM; part of the tightly coupled Hsinchu Science Park ecosystem.
Cadence - EDA company cited alongside Synopsys as part of the semiconductor ecosystem clustered in Hsinchu.
ASML - Lithography equipment maker founded in the same era as TSMC and ARM; part of the interlocking industry that emerged together.
General Instrument - Chang's employer immediately before TSMC; where he first encountered fabless-company founders (Gordon Campbell, and Atmel's founders) pitching to use GI's spare fab capacity.
Atmel - Early fabless company that asked General Instrument to manufacture its wafers, an early real-world instance of the fabless model Chang later built TSMC around.

Techniques and frameworks

Pure-play foundry model - TSMC's founding strategic choice to never design or sell its own chips, so it never competes with its customers - the core explanation the hosts give for TSMC's dominance.
Learning curve / experience curve pricing - BCG framework (Bruce Henderson, Bill Bain) that Chang applied at TI and TSMC: price ahead of current cost, sometimes at a loss, to win volume fastest and ride the cost curve down before competitors can.
Fixed R&D budget as % of revenue - Chang set TSMC's R&D spend at a flat 8% of revenue regardless of the business cycle, removing annual budget negotiation and letting the R&D org plan multi-year roadmaps.
Customer capacity deposits - TSMC required deposits from customers (with the threat, though rarely used, to 'confiscate' them) to align customer demand commitments with TSMC's own capacity investment risk during capacity-constrained decades.

Summary

This episode is a rare, direct interview with Morris Chang, the 93-year-old founder of TSMC, recorded in Taipei and secured through an introduction from Jensen Huang. Rather than a company-history retrospective, it is built around a small set of specific stories from Chang's newly published (Chinese-only) memoir volume two: how the TSMC-NVIDIA relationship began and nearly broke down, how TSMC decided to bet the company on the 28nm node, and how TSMC won and then briefly lost and re-won Apple's business. Hosts Ben Gilbert and David Rosenthal let Chang narrate at length and save their analysis for a "Playbook" segment after the interview ends.

The NVIDIA and Apple stories both illustrate the same pattern: Chang treating relationship management as a personal, high-touch discipline even at massive scale. He describes resolving a bitter 2009-2010 yield dispute with NVIDIA over a casual pizza-and-salad dinner at Jensen Huang's house, delivering a $100M+ settlement offer with a 48-hour deadline to force closure without arbitration. The Apple story is similar in tone but structurally riskier: TSMC took on billions in new debt (rather than cutting its dividend, which Chang refused despite Apple COO Jeff Williams suggesting it) to fund capacity for only half of Apple's stated 20nm demand - a hedge that nearly backfired when Apple's first 16nm orders briefly went to Samsung before reverting to TSMC.

A second major thread is capital allocation and pricing discipline, rooted in the learning-curve (experience-curve) theory Chang absorbed at Texas Instruments from Boston Consulting Group's Bruce Henderson and Bill Bain in the early 1970s. Chang describes fixing TSMC's R&D budget at a flat 8% of revenue to remove annual budget fights, and pushing through a controversial threefold capex increase around the 28nm node against board resistance, framing it (quoting Shakespeare) as "taking the tide at the flood." Both decisions are presented as deliberate applications of learning-curve logic: get to volume and scale fastest, even at short-term financial risk, because the market has natural-monopoly characteristics that reward whoever gets there first.

A third thread is TSMC's founding strategic bet on the pure-play foundry model itself - never designing or selling chips, so it never competes with its own customers. Chang recounts recognizing the coming fabless industry years before TSMC's founding, through direct encounters at General Instrument with early fabless founders (including the founder who told him "I'm not going to build a fab"), and recounts TSMC's refusal of IBM's 1999 co-development offer as a deliberate choice to preserve technological independence rather than become a "junior partner." The hosts' post-interview analysis ties this to the emergence of ARM as an alternative to x86 - without ARM, they argue, there would have been no independent fabless ecosystem for a pure-play foundry to serve, and ARM, TSMC, Synopsys, Cadence, and ASML being founded within a few years of each other in the mid-1980s is presented as more structural than coincidental.

Chang also discusses his approach to layoffs and organizational trust: he refused to base TSMC layoffs on individual performance ratings (which he calls too subjective to be credible), a stance that traces back to disagreements at TI in the early 1970s. A previous TSMC CEO's attempt to lay off 600-700 employees via bad performance reviews during the 2008 financial crisis triggered protests outside Chang's home and was a direct factor in Chang retaking the CEO role in 2009. The episode closes with the hosts' on-the-ground observations from visiting TSMC's Hsinchu Science Park headquarters - the physical concentration of TSMC, ARM, Synopsys, Cadence, and TSMC's customers within walking distance of each other, and two universities feeding graduates directly into the ecosystem - which they call possibly the most successful government-funded industrial policy initiative in history.

Notable Quotes

"The letter said... they were looking for a foundry. They had approached TSMC's San Jose office, but they really got no answer. Would I please contact Jensen." - Morris Chang, describing NVIDIA's first outreach to TSMC in 1997

"If in a year you have to hire the laid off people back, then you shouldn't lay off, because the layoff separation expense is usually about half a year, and it takes at least half a year to train a person." - Morris Chang

"I decided this 28-nanometer was going to be our tide... 'There is a tide in the affairs of men which, taken at the flood, leads on to fortune.'" - Morris Chang, quoting Shakespeare on the 28nm capex decision

"Intel just does not know how to be a foundry." - Tim Cook, as recounted by Morris Chang, explaining why he wasn't worried about Apple's flirtation with Intel as a chip supplier

"The insight is that the goal that you are playing for is to be the largest volume player at the end of the game... we accelerate the pricing to where we think it will get to at the end of the game." - David Rosenthal, summarizing learning-curve pricing strategy in the post-interview Playbook