TSMC: The Inventor of the Pure-Play Foundry

In one sentence: A "contract manufacturer" that runs a 40% operating margin — TSMC invented the pure-play foundry model and turned what everyone agreed was a commodity business into "the clearest example of process power I could ever imagine." Every chip in every iPhone, soon every chip in every Mac, the in-house silicon of NVIDIA, AMD, Qualcomm, Amazon, Microsoft, and Facebook, and nearly all the chips in cars, smart-home devices, and fighter jets come off its lines; 50%+ of the foundry market, 95%+ of the profit, 90% of the leading-edge node. Acquired uses it to argue one thesis: a man of 56, twice pushed out, whom the entire industry told there was "no market," designed a company custom-fit to Taiwan's endowments — and it grew into the ninth most valuable company on earth with a moat that is "everything but bulletproof." They recorded this (September 2021) in the middle of the global chip shortage, with Ford unable to build F-150s for lack of chips — "semiconductors run everything, and they run semiconductors." "Semiconductors are the new oil, David."

The Company on One Page

Founder Profile: Morris Chang

A backdrop of war: three flights and three wars before the age of 18 (the Second Sino-Japanese War, World War II, the Chinese Civil War). On December 8, 1941, three hours after Pearl Harbor, Japan invaded Hong Kong — David dwells on this detail that "everybody knows Pearl Harbor, but people don't often talk about Hong Kong." Ben's counterpoint: America's stability was itself a factor of production for innovation — if what you want to do with your life is innovate, "you could count on the fact that it's likely that 10 years from now, whatever economic structure or political structures exist will continue to exist."

Arriving in America:

"My reaction entering Harvard was sheer ecstasy, almost disbelief. What a country! The United States was at its peak in its moral leadership, its political leadership in terms of democracy, and it was the richest country in the world."

But under the ecstasy was a cold self-assessment:

"In the early 50s in the United States, there were Chinese laundrymen, Chinese restaurateurs, Chinese engineers, and Chinese professors. Those were the only respectable professions for Chinese. No lawyers, no accountants, no politicians."

Harvard mass-produces lawyers, finance people, and politicians — not engineers — so he transferred to MIT to study engineering. This was the first performance of a lifelong move: read the constraints, then pick the feasible optimum. It is structurally identical to the reasoning that, 35 years later, designed a pure-play foundry around Taiwan's endowments.

The $1 gap and the barroom night school: he failed his MIT PhD qualifying exams twice (two is the limit; his own dry joke: unfortunately the biggest impediment was that he failed the qualifying exam; fortunately they were kind enough to let him take it a second time — which he also failed). Job-hunting, Ford offered $479/month (his dream job), Sylvania's new Semiconductor Division offered $480; he asked Ford to beat it, Ford refused — and the automotive industry lost a mechanical engineer. For three years at Sylvania (Ipswich, Massachusetts), a man with zero electrical-engineering background studied Shockley's 1950 textbook Electrons and Holes in Semiconductors in his hotel room in the evenings, then went down to the hotel bar late at night — not to drink, but to buy drinks for an alcoholic senior engineer in exchange for answers. Of that colleague:

"Didn't solve all my problems, but he solved enough so that I could move ahead. He was my main teacher about electrical engineering."

The trigger to leave was a senior manager's talk at the plant: "We cannot make what we can sell and we cannot sell what we can make." He recognized he was on the wrong bus and got off — Ben: "That's a signal to move on if I ever heard one."

25 years at TI — two triumphs, two exits: joined TI in 1958, then the biggest semiconductor company in the world. Triumph one: IBM 7090 yield 0% → 20% (four months); IBM assumed luck at first, sent engineers to talk to him, and was won over by his ability to explain why it worked — which earned him his first management job (a germanium transistor development department with 20-plus engineers) and made TI president Pat Haggerty learn his name. Triumph two: learning-curve pricing (see Playbook). TI sponsored his Stanford PhD on full salary; he finished in two and a half years — when Stanford president John Hennessy later asked the secret in front of an audience, he answered, "I don't know. I'm focused. I didn't do much else." In 1972 he became VP over the whole semiconductor business, one step from CEO and the presumptive heir. Then the slide down: in 1978 Chairman/CEO Mark Sheppard, who believed a good manager could manage anything, moved him to rescue the consumer business —

"In the semiconductor business, it's just technology and costs; in consumer, technology helps, but it's also the appeal to consumers, which is a nebulous thing."

In 1983, after five years without a turnaround, he was demoted to "head of quality and people effectiveness" — "put out to pasture" — and resigned at 52 ("My career at TI, 30 years, done," his phrasing; by his 1958 start it was actually 25 years). David is indignant on his behalf: "This dude built your semiconductor business." Why did he never become TI's CEO? David identifies three explanations: a possible ethnic ceiling (no evidence, but no surprise if true), the failed consumer rotation, and disputed blame for TI semiconductor's own whiffs on MOS and the IBM PC. He then went to General Instrument in Manhattan as COO and found "a proto tech private equity firm" — acquiring semiconductor assets, cleaning them up, and reselling them, pure financial engineering at odds with his "research, build, develop technology, push the ball forward" values — and left within a year at 54. Two major setbacks in 18 months:

"After these two setbacks, at TI and GI, I did not think that my aspiration to be the CEO of a major US company was in the cards."

David's punchline: he was right — he never did become the CEO of a major US company.

A shadow parallel: Donald T. Valentine (later the founder of Sequoia) joined Sylvania's Semiconductor Division at almost the same moment as Morris — different location, different function, never overlapping. Valentine played water polo out in California and fell in love with it, went via Raytheon to Fairchild (after the Traitorous Eight left Shockley), then founded Sequoia; Morris ground it out in a Massachusetts hotel with Shockley's textbook. Sunny California against the ascetic East Coast — both routes led to the center of semiconductor history, one becoming the king of venture capital, the other the "Y Combinator of semiconductors."

Going to Taiwan in a near-retirement frame of mind: in 1985, when KT Li recruited him to run ITRI, every former colleague and his then-wife told him not to (the marriage may have broken up partly over it). He considered himself financially safe —

"By then I was financially pretty secure. I was not rich, but you also have to realize that the standards of wealth were much lower back in 1985... I could live according to the way I desire, which was actually pretty modest, for the rest of my life without having to earn a living or a salary."

— and treated it as a soft landing. On the ground it was immediate culture shock: ITRI staff were government "job-for-life" employees, and Taiwan was under (or had just exited) martial-law authoritarian rule; he forcibly rebuilt ITRI into a high-performing organization. Staff saw him as "a foreigner who suddenly became their boss": "They were scared of me." (David: they were right to be.) In the 1970s, while running TI semiconductor, he had visited Taiwan several times to discuss a TI-owned plant, meeting KT Li and other officials — the seeds planted long in advance. Note: he is from Ningbo, not Taiwan, so "returning to Taiwan" doesn't hold — Taiwan was a strange land to him at first; if anywhere was "home," it was more likely America (he was already a US citizen). "It's complicated."

Between the acts: at 74 (2005) he first retired to Rick Tsai, read literature, and credited his second wife with reinvigorating him; at 78 (2009) he came back. Three forces framed the return: the financial crisis, a handoff seen as failed, and the mobile-and-cloud waves forming as EUV first looked feasible. His read: the PC/x86/Intel paradigm was ending — "the index on all that's going to take over is TSMC." After locking down the Apple deal in 2013 he stepped down as CEO again, and in June 2018 retired completely at 86 ("presumably for real").

Personality slices: he holds Taiwan's highest civilian honor, the Order of Propitious Clouds with Special Grand Cordon (the highest of nine ranks; David: "He's like a knight of Taiwan"); asked how he finally quit smoking (around age 94): "I never stopped. I still smoke." — and he'd "delved into the data" to claim pipe smokers outlive non-smokers, that smoking harms the lungs but helps the mind (Ben: you can surely find data both ways — "this gives you a sense of who Morris is"). To prep, both hosts watched and listened to every publicly released piece of Morris footage. David's verdict: "Every founding story is unique, but I think this might be the most unique founding story we've had on Acquired so far" — you cannot script it: 56, twice pushed out, marriage falling apart, everyone against him, and then he went to Taiwan. The lesson: a founder's chips can be 30 years of full-industry pattern recognition rather than youth and fire; one of the most important companies in history was founded by a man then seen as washed up.

The Playbook

Each entry: origin story → insight → effect.

**1. Yield is learned, and yield is the crux of semiconductor manufacturing**

• Story: his very first project at TI in 1958 was the IBM 7090 — IBM, breaking with its own self-manufacture habit, farmed out the chips for its first transistorized mainframe to TI as a second source (David: "almost like a contract manufacturer of chips or like a foundry business"). IBM's own line ran ~10% yield; TI's ran 0%. Morris systematically tuned it as a mechanical engineer and hit 20% in about four months.
• Insight: chip fabrication is a chemical-and-mechanical process, yields are necessarily low early in a new process, and yield climbs a learning curve with cumulative volume — whoever runs at max capacity first learns first. Yield isn't a gift; it's a function of cumulative learning.
• Effect: both the start of his management career and, 30 years early, the first rehearsal of the foundry business model — the origin point the episode keeps returning to.

2. Learning-curve pricing: price is a weapon to accelerate learning, not a tool to recover cost

• Story: after 1967, Morris and BCG inverted the industry's pricing convention. The convention priced new products high to recoup huge fab investment; he did the opposite —

"Start low and then continually automatically reduce the price every quarter even when the market did not demand it... A lot of people thought we were being foolish. Why would you reduce the price when you didn't have to? But we did it because we believed in it, and indeed our market share just kept expanding."

• Insight: high prices suppress demand → the line runs below capacity → learning is slow → yields climb slowly → unit costs won't fall, a vicious cycle. Ben's added frame: the industry was passing its exact cost structure through to customers and missing the operating leverage — it should have eaten the upfront fixed cost itself and earned outsized margin on every chip off the line.
• Effect: TI's IC business became the biggest and most profitable in the world, beating Fairchild, National, and the newborn Intel. This is the second cornerstone of the later TSMC model.

3. Design the company for the constraints, not for the ambition

• Story: ordered by KT Li to produce a business plan in three days in 1987, Morris took inventory of Taiwan's endowments — the episode's central quote:

"We had no strength in research and development... no strength in circuit design, IC product design... little strength in sales and marketing, and we had almost no strength in intellectual property. The only possible strength in Taiwan that we had, and even that was just a potential one, not an obvious one, was semiconductor manufacturing, wafer manufacturing. And so what kind of company would you create to fit that strength and avoid all the other weaknesses? The answer was a pure-play foundry."

• Insight: strategy = matching endowments. Structurally identical to his 1950s choice to transfer to MIT after reading the ceiling on Chinese career paths — this is Morris's one consistent mode of thought.
• Effect: the pure-play foundry is born. The whole industry treated it as a joke (AMD's Jerry Sanders: "Real men have fabs"); today it is the entire skeleton of the ninth most valuable company in the world.

**4. You see more clearly from the outside: he hoped for, rather than foresaw, the fabless era**

• Story: the raw material of the insight had accumulated for two decades —

"When I was at TI and General Instruments, I saw a lot of integrated circuit designers wanting to leave and set up their own business, but the one thing or the biggest thing that stopped them from leaving those companies was they couldn't raise enough money to form their own company."

The same fact sat in front of everyone, but only he, now standing in Taiwan, read it as opportunity. His intellectual honesty is just as rare:

"What very few people saw, and I can't tell you that I saw was the rise of the fabless industry, I only hoped for it. But I had better reasons for hoping for it than the people at Intel, TI, and Motorola because I was now standing outside."

• Insight: the same fact looks to the incumbent like a moat (people who can't build fabs can't take your business) and to the outsider like a dammed-up lake (a whole generation of founders held back by a capital barrier). Position determines vision.
• Effect: from 1989–93 on, Qualcomm, Broadcom, Marvell, and NVIDIA were all founded on the TSMC platform — the dam broke, and the water all flowed into TSMC's riverbed.

5. "If you build it, they will come" is a bad strategy — and it worked

• Story: the mainstream verdict (Intel, TI, Motorola) was that pure-play had "no market." David channels a Don Valentine-style interrogation: "Where's the market? Show me the market." The whole idea was a solution looking for a problem — and it broke the cardinal rule: "This is what Sequoia and Don Valentine hated. They would never invest in developing a market."
• Insight (Ben's cold water): if a thing works 10% of the time and fails 90%, one in ten times it works; outcome alone can't tell you whether it was a good strategy or a bad one that happened to hit. David's close: "This is the thing about startups. There are all these rules, but they can all be broken. There is no formula."
• Effect: TSMC made it. But the hosts refuse to retroactively make it inevitable — Ben: "unless the foundry model and the fabless model was inevitable." David: "Yeah, maybe somebody else would have done it. Maybe." Ben: "But they didn't, and Morris did."

6. Bet on customers with no fab in their DNA: from IDM dregs to structural growth

• Story: the early years, TSMC lived on the big guys' scraps. Morris:

"The IDMs would let us manufacture their wafers only when they didn't have capacity, or when they didn't want to manufacture the stuff themselves anymore... as soon as they got the capacity, they would stop giving us orders, so it wasn't a stable market."

• David's jab: the products the IDMs "didn't want to make anymore" were the money-losing ones — they were effectively transferring their losses to TSMC.
• Insight (Ben): IDMs have manufacturing DNA and will always pull the most profitable products back in-house; fabless companies have no manufacturing gene and will never take the orders back. The right move is to bet on the latter's growth, not to service the former's capacity swings.
• Effect: the customer base shifted from "the IDMs' capacity release valve" to "the birthplace of a fabless generation," and orders shifted from cyclical dregs to structural growth.

7. The TSMC flywheel: profit → CapEx → process leadership → customer success → more profit

• Story: David lays out the episode's core model: the fabless market grows (TSMC seeds and enables it) → TSMC revenue grows → profit (at 50% gross margin, 40% operating margin) is plowed into more advanced equipment and new fabs → process leadership → fabless customers get better performance → they address more markets and use cases → existing customers get bigger and new ones are born → revenue grows further.
• Insight: the flywheel turns painfully slowly at first ("It goes slow. Like any flywheel, it takes a lot of effort and a lot of time to start turning it") — years of dregs — but once turning, its power "even exceeds his wildest dreams." The 2020 proof: $48B revenue, $20B operating profit, $17B (85%) straight back into CapEx.
• Effect: a triple analogy — David: "He's like the Y Combinator of semiconductors." "It's like AWS for chip companies. Never would have been possible before." Ben invokes Tencent: as Tencent distributes to its portfolio through WeChat, TSMC provides the manufacturing — with TSMC, fabless founders need to raise only 10%–20% of the capital they'd otherwise need. The extreme sample is NVIDIA: founded 1993, raised only $20M total, never built a fab, now a $350–400B company (~2x Intel). Morris once seriously considered becoming a VC — and in the end achieved a VC's economics by another route: buying the index on the entire future of semiconductor innovation.

8. Moore's Law + Rock's Law ⇒ the leader necessarily becomes a monopoly

• Story: Moore's Law (transistor count doubling) is, financially, about the OpEx of production at scale; Moore's second law — aka Rock's Law, after Arthur Rock — says the cost of a fab doubles every four years, which is about upfront CapEx, now $15B–20B per fab. David checks it live: TSMC's annual CapEx goes $17B → $30B → $100B over three years — faster than "doubling every four years."
• Insight: combine the two laws — pile as much investment as possible into fixed cost and amortize it at global scale; the cost threshold rises exponentially ⇒ the number of players who can keep up falls exponentially ⇒ the most advanced, most profitable one becomes the monopoly. David: "Winner take all, there you go."
• Effect: the convergence of leading-edge players — 22 (early 2000s, ~150nm) → 14 (late 2000s) → 6 (mid-2010s) → 2 (2021, only TSMC and Samsung at 5nm); Samsung's 3nm has already slipped to 2024, and the next node may leave just one. Morris's metaphor:

"The semiconductor business is like a treadmill that speeds up all the time. If you can't keep up, you fall off."

9. Money can't buy the catch-up: assembly is manufacturing, chipmaking is alchemy

• Story: on the Bloomberg Odd Lots podcast, the reporter covering TSMC said of SMIC that "they can do that and they are doing that, but they wouldn't know what to do with it. It's not because they're dumb." Ben: even with ASML's machine in hand, only a tiny number of people in the world who have done history's most advanced manufacturing know how to do the next generation.
• Insight (Ben's dividing line): don't conflate "Apple outsourcing to Foxconn" with "Apple outsourcing to TSMC" — assembling an iPhone is manufacturing; making a brand-new next-generation chip is alchemy, and "the alchemy can only be done by alchemists." The marriage of scale economies and process power means the engineering-know-how learning curve is measured in decades — even if the US government spent a trillion dollars in a year it couldn't replicate TSMC in the near term.
• Effect: David's verdict:

"All of their IP, all of their people, all of their know-how, all of their relationships with ASML and the like, no amount of money can replicate it."

10. The Rosenthal Postulate: never make strategic decisions based on pricing

• Story: Ben reads Bloomberg verbatim — in the mid-2000s, as Apple prepared to release the iPhone, Steve Jobs approached then-Intel CEO Paul Otellini about supplying the chips (Intel already supplied Apple's Mac processors); Otellini judged it a lowball offer and declined; Apple went to Samsung, later designed its own chips (acquiring P.A. Semi), and finally handed production to TSMC. David throws both fists in the air, calling it "the biggest strategic error of all time." Ben stresses: Intel's main reason for refusing wasn't "we're an x86 company" but that the money looked too low.
• Insight: David proposes a "postulate" (unprovable like a geometric axiom but required by the system): "Never make strategic decisions based on economics" — narrowed by Ben to "Never make strategic decisions based on pricing." If the structure is right and you can share the upside, refusing on structure is understandable; refusing only because the number is too low is a cardinal sin. "Humans are so prone to cutting off their noses to spite their faces."
• Effect: three examples strung into one line — Andreessen missing Uber over a $300M valuation he thought too rich; Intel missing the iPhone over a lowball offer; Ford failing to hire Morris Chang over $1 a month. Three quotes, three era-defining misses.

11. The purest specimen of the innovator's dilemma: Intel is to EUV as Kodak is to the digital camera; indecision is paralysis

• Story: in 2012 Intel became the biggest early investor in EUV, committing more than $4B, yet was slower than rivals to adopt it and always doubted its feasibility (WSJ: "Eventually, Intel calculated that it was a sure bet to try and improve existing ways of handling lithography"). As of recording, not one shipping Intel chip was manufactured by Intel using EUV — and EUV is the absolute precondition for the next generation of chips. The recent whipsaw: former CEO Bob Swan prepared to outsource Intel's chips to TSMC (the surrender path) → leadership change → Pat Gelsinger fully reverses, aiming to turn Intel itself into an outside foundry.
• Insight: Ben's analogy — Kodak invented the digital camera in its own lab, knew it was the future, and refused to commercialize it. An incumbent cannot counter-position against itself; this is structural, not a failure of vision ("Intel was there. They invested in it, they saw it, and they're like ehh"). Ben: "Indecision is paralyzing. This company has spun its wheels one direction or the other and all it's done is make itself deeper in the mud." The flip side: even so, Intel still holds ~80% of the computer-processor market and more in servers — "it takes a long time to become irrelevant," the incumbent's afterglow is long, but the direction is set.
• Effect: Intel fell behind at 5nm and has been overtaken in market cap by both TSMC (~2.5x) and ASML. David twice repeats the same line: "That's why you need startups."

12. The inverse of the Bezos beer law: how to build a platform company

• Story: David cites Bezos on AWS — early-20th-century German breweries generated their own electricity, then all handed it to the grid; anything that doesn't make your "beer" taste better (finance, accounting, generating power), outsource it. This chestnut he wants to double-underline, but the episode adds a new face.
• Insight: turn it around — whenever you see lots of companies all doing something that doesn't make their beer taste better (like everyone building their own fab), that's a huge platform opportunity — and that is how you build a platform company. TSMC is to chipmaking as AWS is to compute infrastructure.
• Effect: Ben's extension:

"When Marc Andreessen says the software is eating the world, it's only eating the world because TSMC has made it so freaking cheap to manufacture silicon... It's the cost of compute asymptotically approaches zero because TSMC, TSMC, TSMC."

13. Don't compete with your customers; give the brand to the customer

• Story: back when a fabless/IDM wanted to rent another IDM's spare capacity, the host usually demanded the right to sell your product under their brand — TSMC never did this; contrast Samsung, which still both designs and manufactures and collides with its customers everywhere. Apple will announce "3-nanometer process" on a keynote slide and take all the credit.
• Insight (Ben): "TSMC is totally fine with that because their job is not to market. It's to empower their customers." David: deliberately riding under the radar is itself strategy — "they intentionally have no brand. The brand is the customers." It's also why the public and the press ignored them for so long.
• Effect: neutrality made TSMC the default choice for every fabless company; deep ecosystem interlock (the engineering of EDA, IP, and foundry mesh together) further raises switching costs.

14. Turning a commodity business into one of the biggest moats in history

• Story: months before recording in 2021, TSMC first announced it would stop cutting prices, then announced a ~20% price increase. David: nobody had done this since the pre-Morris era — and remember, Morris's own second big innovation was automatic quarterly price cuts.
• Insight: foundry is an agreed-upon commodity business, and TSMC turned it into a monopoly with pricing power — customers have nowhere else to go. David: "This has got to be one of the biggest moats of all time."
• Effect: the Ben/David positioning couplet — Ben: "In many ways, they're the Apple of semiconductors. They don't have all the market share, but they have all the most profitable market share." David: "They are the iPhone of semiconductors." — 50%+ of the foundry market, 95%+ of the profit.

Moat Analysis (the 7 Powers framework)

7 Powers is Hamilton Helmer's strategy framework (Ben calls it "the best business theory book," David adds "go read the book if you haven't," and Hamilton has been on the show): a Power is defined as a source of long-term differential profit margin versus competitors, and there are seven. Acquired runs every episode through this checklist; the table below is this episode's verdict.

Bear Case

1. The "no longer need the leading edge" argument (raised on Odd Lots): David finds it very weak — Tesla doesn't want the most advanced node? Apple, Google, Amazon will say "we're good"? No chance. Ben adds the mechanism: software will always fill the complexity ceiling of the most advanced hardware it can run on — developers at most test on two- or three-year-old equipment ("Apple is slowing down my computer" is an illusion: it's not throttling code, it's that all software assumes it's running on the most advanced processor on earth).
2. Paradigm shift: the only thing that could unseat TSMC is a total shift on the scale of "mobile to desktop" — e.g., quantum computing arriving with an entirely new, orders-of-magnitude-cheaper implementation, or an "AWS moment" that suddenly collapses the cost of technology access; small shifts (a better substrate than silicon) TSMC probably weathers fine ("I bet they'd be fine"). David: absent a paradigm shift, "this is TSMC's to lose."
3. China taking Taiwan (the core bear case): David's added phrasing — depending on your view, "annex" or "assert its always-claimed sovereignty." Ben's scenario: short of a genuine world war, TSMC's capacity would be redirected to Chinese customers — Huawei has been a longtime customer. Losing TSMC: "Imagine if we didn't have access to leading-edge semiconductors anymore. That's everything. What part of our lives do not run on semiconductors?" The present proof: Ford halting F-150s. David's thought experiment: can process power be airlifted? — evacuate the people, redirect the ASML equipment; does process power follow? "I don't know." Counter-reference: the Toyota Production System failed to replicate itself at NUMMI, its GM joint venture (today's Tesla plant in Fremont) — though David notes, "it wasn't under threat of war." The Boeing comparison: on the US list of strategic soft spots, who matters more? David ends it in one line: "Boeing needs semiconductors." Korea/Samsung is structurally the same — North Korea is right there (David has been to the DMZ: "It's like an amusement park, super, super weird"); maybe the strait is like the two Koreas: right there but "this isn't actually going to happen"? Both feel the risk has ratcheted up noticeably over the past year. Ben's close: this is globalization — everyone sharing resources and entangling everything is in everyone's interest, until someone decides otherwise, and then big problems. "Hopefully it just continues to be okay that TSMC is located on an island that is of disputed claim." David's dark-comedy exit: maybe the best solution is discovering aliens are real (echoing his earlier carve-out — the Harvard astrophysics chair's book arguing Oumuamua could be an alien spaceship), which would be the force that unites humanity and makes these conflicts look petty. "But he said, don't use that as an investment thesis."
4. Geographic concentration: the market prices in the risk that TSMC has barely diversified geographically. Ben reads the real motive for Arizona ($12B, not the most advanced node, part of the Biden administration's subsidy bill), the new Japan plant, and mainland-China operations as "basically getting free money" rather than hedging; Morris himself said publicly it makes no business sense to put the leading edge in those countries — the ecosystem is in Taiwan. Arizona's US value is indirect: getting Americans reacquainted with the craft in case reshoring is ever needed.

Bull Case: structurally nobody can catch them — "structurally, nobody can catch them. Something unforeseen has to change" (David); pricing power is already cashing in (the ~20% increase); on the demand side ML, crypto, 5G — "anything they tell you about the next wave of computing is something TSMC makes that enables it all." Ben's one dialectical reservation: "But something unforeseen will change because it always changes."

Grading (this episode's twist: pantheon ranking)

Ordinary letter grades are too boring — grading Taiwan's decision to own ~50% of the company at founding is an A+ with nothing to discuss. The better question: where does TSMC rank in the pantheon of the greatest technology companies of all time?

• Frame of reference: it's the equivalent of circa-2000 Intel, but the set of products it manufactures is 100x the scale Intel's was then (Ben's estimate). The attribution problem: how much credit to TSMC vs. ASML vs. the world's chip-design talent (including Apple's 600 silicon designers)? Hard to disentangle.
• David: we can safely say it surpasses Intel — a heavy statement (Intel = Silicon Valley, the Traitorous Eight, the embodiment of Moore's Law); "how they grew 30% last year at an already unimaginable scale, Intel's not doing that."
• Ben: in compounding, all the value shows up at the end — the value TSMC will create for the world over the next one, two, three years is probably more than the entire silicon industry before it combined; but on pure value creation, he won't yet put it above Facebook, Amazon, Apple, Microsoft, Google. David plays devil's advocate: none of those companies' current innovations happen without TSMC. (Cue: "But they didn't, and Morris did.")
• Moat cross-comparison: geopolitical risk aside, David thinks no company in the world has a stronger moat than TSMC — the FAANG+Microsoft moats have all been crossed (Microsoft fell and came back on a new strategy; Facebook and Google are still young), and "TSMC is impenetrable." The hosts' relay: "It's like bulletproof." / "It's everything but bulletproof. Sadly."
• Historical analogy: who once had the most similar dynamic? Standard Oil — structurally crowding out all competition, the rich getting richer; the best part is its descendants still exist ("we'll talk way more about this later"). Self-check, David: "Maybe we're exaggerating because we're so deep in it. We always go native on these episodes."
• Final verdict: unanimous — Top 10, not yet Top 5 ("I'll put a stake in the ground... maybe we need to revisit this"). Ben's rider: the most successful B2B hardware company ever, full stop; for "most successful B2B company ever," it probably only has to fight Microsoft. David raises it further: maybe across all industries too — "semiconductors run everything, and they run semiconductors." Ben's closer: "Semiconductors are the new oil, David."

Deep Cuts (TSMC itself)

• KT Li's three-day draft: KT Li (the "Father of Taiwan's economic miracle") first had Morris run ITRI ("you're running our Bell Labs"), then ordered directly: "I don't want somebody else in ITRI to do it. I want you, Morris Chang, to start a new semiconductor company here in Taiwan, and I want you to make it into a global leader." The deadline for the plan was first a week, then a day later moved up to Friday — three days. Morris: "It was like in the movie The Godfather. It was an offer I couldn't refuse." This was his third job in three years, and Taiwan was not a democracy — refusal was effectively impossible. David's take on the government's side: "You are our foot soldier. We are the mafia." Ben's reminder: KT Li most likely wanted an Intel-style IDM; the pure-play foundry was the latitude Morris seized while executing the order. David: "necessity is the mother of invention" — these three days created the now-ninth most valuable company in the world.
• The $0 valuation reveal: David quizzes Ben live — "Do you know what the pre-money valuation was on TSMC?" Ben: "I couldn't find that anywhere." David: "It was $0. Morris Chang got no equity. Zero." Ben: "Oh my God." The $220M raise: government 50% ($110M), Philips 28%, and the Premier of Taiwan personally strong-arming the remaining ~22% out of local business leaders. David: "This is so the opposite of Silicon Valley." / "There by the grace of the government." Morris later put essentially all his excess cash flow into buying TSMC shares and is now worth ~$3B. Ben's take: his career had gone badly enough that he was willing to take that deal.
• The funding roadshow shutout: the government required the other half to come from "the future biggest customer" (an Intel or TI) to align interests; Morris saw Intel, TI, and Motorola, and the answer was uniform: "Morris, we like you, but no." No engineer would refuse a meeting with Morris (Ben: "He almost ran TI"), but nobody wrote a check. The last-ditch deal was with Philips. Morris's acid return volley: "the first rung of the second raters in semiconductors" — but they were the only ones willing to pay.
• The Apple deal: closed in 2010 (Bloomberg): a $9B initial investment, 6,000 employees, an Apple-dedicated plant built in 11 months (with several more years before it actually produced chips). Apple's lead negotiator was Jeff Williams ("Tim Cook's Tim Cook"), the deal reportedly inked over dinner at someone's home, in a living-room conversation. Jeff Williams: "If we were to bet heavily on TSMC, there would be no backup plan." David: "That was a bet-the-farm deal and something only Morris could do." — even though this founder held not a single share except the ones he bought; Ben: "It really speaks to founder gravitas," and "$9 billion of manufacturing capacity just for a deal with one company — it paid off." The dual-source window: a good share of the A4/A5 was fabbed by Samsung until the iPhone 6 (2014, the first bigger screen and a sales explosion) went fully exclusive to TSMC.
• The Jobs feud, fact-checked: the hosts reconcile it live — "thermonuclear war" was aimed at Google/Android; the grudge against Samsung came from the design-copying lawsuits; Tim Cook later read the "toxic hell stew" line from a stage. David: "It doesn't get any better than that."
• Tracing the Apple-silicon bloodline: David: "The lineage of all of Apple silicon... was ARM and DEC to Intel to P.A. Semi to Apple." (DEC's high-performance StrongARM was acquired by Intel, renamed XScale, then shut down; the engineers who wouldn't work for Intel founded P.A. Semi, which Apple acquired to become the source of Apple silicon.) Ben: "So you can trace Apple silicon all the way back to Apple." — because ARM was itself a three-way JV Apple co-founded in 1987, an Inception-style loop.
• The numbers panel (at 2021 recording): from a ~$4B market cap at the 1994 Taiwan IPO to $550B, a 27-year IRR of 19.9% and revenue CAGR of 17.4%; $28B cash on the balance sheet; revenue up 31% from 2019 to 2020; 50%+ foundry market share, 95%+ of the profit; 90% of the leading edge (at 5nm, Samsung just 5%–10%, TSMC "going to 100%").
• Rick Tsai's second act: Morris's longtime lieutenant, took over as CEO in 2005 and was ousted in 2009 — TSMC suddenly held a press conference announcing the 78-year-old Morris's return, the release saying it "will not affect TSMC's fighting spirit and is likely to spur greater intensity." David likens it to a Michael Jordan comeback: "I don't even know what number he's wearing. It's not 45 because that was the second act. He's beyond Jordan at this point." Rick Tsai later became CEO of MediaTek.
• Jensen Huang's ticket in: the vast majority of NVIDIA's chips are made by TSMC, but Jensen has said it took him a while to get on Morris's radar. Around 2005 the GPU market was game consoles (NVIDIA's stock tracked Sony PlayStation/Xbox contracts); what made it a $350–400B company was cloud and AI, and every generation of chip came from the same foundry.
• The fate of Jerry Sanders's line: "Real men have fabs." — the mid-1980s industry consensus against pure-play, distilled into one sentence; yet in the 2000s AMD spun its fabs out into GlobalFoundries and went fabless itself. David: "So ironic." (And the aside: "Real men have fabs and real men use power.") At recording, GlobalFoundries was preparing to IPO — Ben's verdict: "It's not TSMC."
• The Gelsinger résumé easter egg (David checking live): Pat Gelsinger started his career at Intel → EMC → CEO of VMware; he was also the outside candidate to replace Ballmer as Microsoft's CEO. Ben: "No way," adding that he'd heard Gelsinger is revered inside Intel and expected to bring good change. "We'll see." Plus David's jab, on Ben describing Bob Swan "throwing in the towel": "Real men are sensitive. They talk about their feelings."
• The hosts' daydream skit: David riffs about using Acquired's "brand power / network economies" to compete with TSMC, with CS-degreed Ben running the machines. Ben: "I wouldn't know the first thing to do." — and the machines have 50-some on backorder, unbuyable for years. A joke that happens to act out the entire process-power argument.
• The interest-rate gag: Morris joked that part of why he could feel financially secure in 1985 was that interest rates were higher then — less principal got you to the same goal.
• The episode-selection easter egg: TSMC sat in the hosts' Google Doc of episode ideas for two and a half years; this episode, with ARM and Sequoia Part 1 (P.A. Semi), forms a semiconductor "mini-series" — Acquired did P.A. Semi back in the early days (an episode in the 20s, when the show really was about small acquisitions).
• The Remaster note (added in the ~2024–25 re-recording open): since the 2021 original, TSMC's market cap has doubled, $550B → over $1T; TSMC and Saudi Aramco are the only two trillion-dollar companies in the world not on the US West Coast — Ben: "It's crazy that the rest are located on the West Coast of the United States, but it really underscores what an extreme outlier TSMC is."

Errata (transcription / slips of the tongue, relayed per the notes): David saying MediaTek "spun out of TSMC" is a slip — MediaTek actually spun out of UMC; Ben saying the "iPhone 4" launched in 2009 is a slip — it was the iPhone 3GS (the iPhone 4 came in 2010); "150nm" is the hosts' rough estimate of the early-2000s leading edge ("let's call it").

Era & Industry Trivia (tangents worth keeping)

• The ASML/EUV deep dive (Ben's special thanks to John Bathgate and Britton Johns from The Knowledge Project podcast): regular (white) light has a 193nm wavelength; the physical limit of lenses plus immersion (water) is about 11nm; the 5nm node can only be done with EUV — a laser hits a molten tin droplet 50,000 times a second to create a plasma that emits extreme ultraviolet light, which is then bounced through about 20 (newly invented) mirrors to etch the wafer; the laser must hit the tin droplet with more precision than the calculations to send the Apollo missions to the moon; the only company in the world that can make the specialized EUV laser is a single German firm (unnamed in the show). One machine costs ~$200M (next-gen heading to $300M) and takes four 747s to ship; ASML makes only ~50 a year, and "TSMC has ordered out all of them for years" — buying up the upstream equipment capacity is itself part of the moat. Nikon simply gave up on EUV. Ben on ASML: "the sole-source provider of one thing in the value chain to mostly one company — and they're bigger than Intel now" (and Europe's most valuable company).
• The Dutch foreshadowing: Ben plants "remember Philips was a Dutch company" early ("Putting a pin in Dutch") and pays it off at the ASML section: ASML was founded in 1984 as a joint venture between Advanced Semiconductor Materials International and Philips — David reacts live: "Oh, wow. I did not know that. That's crazy." And Philips also put up 28% of TSMC in 1987: the king of equipment and the king of foundry share the same bloodline.
• The five-layer value chain (Ben and David's industry map): IP/architecture (ARM et al.; in the SoC era you can just buy USB functionality as off-the-shelf IP — IP companies aren't patent trolls but real module suppliers) → EDA (the two leaders Synopsys and Cadence, "Excel or Figma for chip designers") → fabless design (Apple, NVIDIA, Qualcomm, post-fab AMD, Cerebras, Tesla) → equipment (Lam Research, Applied Materials, Tokyo Electron, ASML) → foundry (TSMC, Samsung, the catching-up SMIC). Ben's hypothesis: the speed of Moore's Law makes it impossible for any one company to be best at every stage at once, forcing the industry from vertical integration to horizontal, best-of-class specialization.
• Five silicon atoms per nanometer: "5nm" originally meant the transistor gate length and is now a marketing term; the hard constraint is that only about five silicon atoms fit in a nanometer between transistors — silicon can't be subdivided further, so future gains must come from a new substrate or from multi-core, architecture, and other "make the chip do more, faster" paths. David: this has always been true, will keep being true, and keeps getting more expensive — and TSMC is the only company that can keep up with the leading edge. Aside: silicon is the second most abundant mineable element on earth ("silicon is sand"), while germanium is expensive and rare — the 1960s pivot to silicon hinged on a breakthrough in doping.
• A sketch of the pre-transistor era: at the turn of the '50s, when Morris switched fields, Fairchild didn't yet exist and Shockley Semiconductor was maybe just getting going; it was still the age of vacuum tubes and ENIAC, the world of Bell Labs; the transistor had just appeared, the integrated circuit wasn't yet invented, and the material was expensive, rare germanium, not silicon. From a "germanium transistor development department" to 5nm — one man's career happens to span the entire history of the industry.
• TI's origins in oil: Texas Instruments built seismic instruments for oil exploration in the 1930s — David: "They were the TSMC, the technology provider to oil companies." (The technology-supplier role recurring across history, a distant rhyme with the closing "semiconductors are the new oil.") Another myth-buster: "Silicon Valley is all marketing." — the biggest semiconductor company of the day was in Dallas; TI was the "Google plus Facebook" of that era (Ben: just without the world paying attention).
• 1987 = the semiconductor Annus Mirabilis: TSMC and ARM were founded the same year. ARM's instruction set was then dismissed as useless: "PCs are always plugged in, so what do we need a low-power chip for? This thing's pathetic." — twenty years later, the mobile era's entire lineup of main chips (Qualcomm, Broadcom, MediaTek, Apple) is ARM + fabless, natively all TSMC customers, with Intel/x86 shut out of the whole paradigm.
• Kilby on the historic place of the IC:

"What we didn't realize then was that the integrated circuit would reduce the cost of electronic functions by a factor of a million to one, nothing had ever done that for anything before."

In 1958, Kilby (TI) and Noyce (Fairchild) separately co-invented the IC, and Kilby later won the Nobel Prize for it; Noyce went on to found Intel with Gordon Moore. That was the very year Morris joined TI — David's triple deadpan: "Coincidence? Yes, totally a coincidence."

• TI's culture disease and "the off-ramp of history": TI's culture was "company man" — rotate, stay, obey, the opposite of Silicon Valley's Traitorous-Eight leave-and-found culture. The result: top MOS talent either suffocated or left (LJ Sevin left to found Mostek, later the early VC firm Sevin Rosen Ventures; Motorola poached MOS engineers in bulk), and TI, with the best engineers, missed the MOS transition; in 1980 IBM's secret Boca Raton team put out a secret RFP for the IBM PC's microprocessor — in Morris's day, TI, with its IBM history, would have been the natural candidate, but the Intel 8088 won, x86 was standardized, and the whole next generation of computing passed TI by. Ben: "On the highway of history, TI accidentally took the off-ramp here." The lesson: a stock of technology is not a durable Power; talent-retention mechanisms and culture are.
• "Made in Taiwan" and Barbie dolls: 1980s American everyday goods carried this label (Barbie, toys, clothes), the imprint of Taiwan's low-end manufacturing era; the industry's dismissal of foundry used the same comparison — "this isn't manufacturing Barbie dolls here. This is real technology. You need to control it soup to nuts." Three-plus decades later, the "control it soup to nuts" IDMs line up to send their most advanced chips into Taiwan's foundries.
• The RCA tech transfer and UMC: in the 1970s Taiwan's government negotiated a semiconductor-line technology license from RCA (one of the few things the government got right before Morris arrived; it had earlier tried to negotiate with TI). By the time it landed, the technology was 2.5 generations behind the leaders and could only do ultra-low-end work; ITRI used it to incubate UMC (a design+foundry hybrid, a contrast to TSMC's pure-play); UMC later spun off its design business as MediaTek, a ~$50B company today. David's outcome-based read: "The government did pretty good." — the "half-right" specimen of industrial policy: the tech transfer, incubation, funding, and picking Morris were right; the brutal parts were the $0 valuation, the founder's zero equity, and the Premier strong-arming the raise.
• Old fabs don't close: after a leading fab makes leading-edge chips for a while, a new generation comes online but the old line isn't shut down — it converts to chips that don't need cutting-edge performance (automotive, IoT). The long-tail economics of mature-node capacity is also the industry backdrop to the 2021 shortage, where "what's short isn't 5nm but automotive chips."
• The interplay of VC and semiconductors (Ben): the origin logic of venture capital was built for semiconductors — to finance the model of "sink huge fixed cost into a foundry, profit on 50%-margin chips"; it happened to fit software (80%–90% margins) just as well or better, but software lacks the moat defensibility of "piling CapEx into manufacturing capability."
• Taiwan's gross-margin baseline: in the mid-1980s the average Taiwanese company ran 4%–5% gross margins — pure commodity territory, everyone price-warring, nobody able to build enterprise value (by contrast, Amazon's ~40% gross margin got it hammered as "low-margin"; David: "I can't even imagine running a company with that level of gross margins"). TSMC grew a 40%-operating-margin monopoly out of the very bottom of the global value chain.
• The two acts of the cloud: the cloud was first a golden age for Intel ("the best thing that ever happened for Intel," x86 CPU servers); as AI/machine learning rose, the center of gravity shifted to GPUs (parallel compute / matrix multiplication / vector math) and specialized chips (Tesla Dojo) — NVIDIA became a $350–400B company on cloud, not the PlayStation. Ben's correction: most cloud workloads today still run on CPUs (some non-tech company running Outlook on Office 365 doesn't need 5nm) — GPUs are the "forward-looking" part.

Cross-domain Notes

This episode has a real and important intersection with the PH domain: Taiwan chip geopolitics is a core entity in the PH domain's fortress-america, technate, and US-China rivalry narratives. This purely-business analysis, recorded in September 2021 with no contact with the PH discourse network, independently converges on several of its points — independent evidence from the business-domain lens, its value lying precisely in "different roads, same destination":

• "Semiconductors are the new oil" and single-point dependence: pushing 7 Powers to its limit, the hosts land on a conclusion structurally identical to the PH domain's geopolitical judgment — the entirety of the world's technological progress rests on one island of disputed sovereignty ("what part of our lives do not run on semiconductors?"), with Ford halting F-150s as proof of single-point risk transmitting into the real economy. If "semiconductors are the new oil" holds, then Taiwan is to chips what the Persian Gulf is to oil — exactly the structural reason the PH domain places Taiwan at the focus of US-China conflict, here derived independently from a business analysis.
• An early record of tech containment against China: at 2021 recording they already note that the Netherlands (reportedly under US pressure) didn't renew its trade agreement with China and that ASML's most advanced equipment isn't sold to SMIC; the Odd Lots reporter's judgment ("they can spend the money and are, but they wouldn't know what to do with it, and not because they're dumb") plus Ben's know-how argument provide an independent business-side read on "does tech containment work": the chokepoint isn't the equipment, it's the decades-long learning curve.
• A prelude to Fortress America / reshoring: the US talking about re-onshoring, the Biden administration subsidizing the Arizona plant ($12B, not the most advanced node), Tesla using Samsung capacity in Austin; Ben reads the diversification motive as "free money" rather than hedging, and Arizona's real value as "getting Americans reacquainted with the craft in case reshoring is ever needed" — mutually corroborating with fortress-america's reading of "pre-positioning the fortress." The show also notes "TSMC seems to have picked the side of the US" (the result of multiple administrations' pressure over five-plus years).
• The hardware precondition of the Technate: the technate narrative centers on an engineer-governed, self-sufficient North American technical body; this episode supplies its hardest constraint from the business side — cutting-edge manufacturing know-how belongs to "alchemists," cannot be airlifted, cannot be rushed (the NUMMI counter-example, the "can process power be airlifted?" thought experiment). Any self-sufficient technical body must solve this link first, and as of 2021 it's in Taiwan.
• The dollar and the gun: Ben's line — "The only way it could be more valuable is if the company had an army. It's like people talk about the US dollar is backed up by the full faith of the US government, which implies guns." — shares an intuition with the PH domain's mechanism analysis of the dollar system (pages like petrodollar) and can be recorded as an independent business-domain version.

Pages Worth Creating

• Entities: Morris Chang（张忠谋） (founder page: three wartime flights, 25 years at TI with two exits, a second resignation at 54, founding a company at 56 with zero equity, retiring at 74 and returning at 78)
• Episodes: Morris Chang 访谈：张忠谋亲述 TSMC (Acquired's dedicated Morris Chang episode, complementary to this one), NVIDIA 之一：GPU 公司（1993-2006） (NVIDIA series Part 1: the $20M raise, zero self-built fab, the "born on TSMC" fabless archetype)
• Concepts: 7 Powers 护城河框架 (Hamilton Helmer's framework, the standard analytical tool across the Acquired series; this episode contributes the textbook case for process power and the "anti-power" reverse case for cornered resource)

Source · acquired