TSMC Founder Morris Chang: The Interview

In one sentence: This is a rare personal interview, not a company-history narrative — 93-year-old Morris Chang, in his own Taipei office a week before the episode aired, telling two hosts (who got the meeting via a Jensen Huang referral and spent all of 48 hours in Taiwan) about his TI years, the founding of TSMC, the NVIDIA and Apple deals that rewrote semiconductor history, and the learning curve — the thinking tool he has used his whole life. The hosts prepped from the unpublished English translation of his memoir (volume two came out in late 2024 in traditional Chinese, a full 26 years after volume one; translated by Karina Bao with funding from Tyler Cowen and Emergent Ventures). Morris himself supplied the closing line that captures how rare this was: "It was my pleasure. Even though it's the first time in a long, long time that I have talked so long."

Interview Timeline

Organized by the chronology of events as Morris narrates them (not broadcast order):

The Morris Chang Portrait (core section)

The arc of a career

At TI he rose to run worldwide semiconductors — the company's largest business — but "at TI I was not the CEO, just one of the executives under the CEO." Two defeats there were later inverted into TSMC institutions: his R&D budget requests were denied every time, and he stood alone against performance-rating layoffs. In his final months at General Instrument he witnessed, back to back, Gordie Campbell (no fab, funding needs collapsing from $50M to $5M) and Atmel (fabless, asking GI to foundry its wafers) — effectively running a live prototype of the foundry business model before ever founding TSMC. He started TSMC in 1987 at 56 (Ben's figure, which Morris did not dispute), handed the CEO role to Rick Tsai in 2005 to serve as chairman only, and retook it in 2009 under the triple pressure of the 40nm crisis, a gross-margin alarm, and the layoff scandal. When he retired, gross margin still sat slightly below the 50% he had chased for years — the target was finally cleared after his time, by COVID and process leadership (55-57% today, per Ben).

How he thinks

• The flying-salesman CEO. "I was on the plane most of the time, building customers. That was my pleasure. I really liked it." Except for the four years out of the CEO seat, he personally maintained the top 15 customers and hunted for the next 15 (Ben's framing, which Morris endorsed).
• A "serious student" of the learning curve. "I absorbed a lot of learning curve stuff, which I used up to now. I found that highly fruitful as a thinking tool." His self-assessment is equally precise: not the inventor, not the initiator — "I think I had a role at TI... in refining it to the point where a semiconductor company can use it effectively. That's my role."
• The seed-planting philosophy. "I planted a lot of seeds when I ran TSMC. I knew that one of these days we will probably need top-level investment bank advice." The early ADR listing bound TSMC to Goldman Sachs; Morris himself served on Goldman's board. When the Apple deal demanded billions in financing, that seed paid out on the spot.
• Betting, but confident. Ben calls the Apple deal a bet-the-company move. Morris: "Yeah, I know, bet the company. But I didn't think I would lose." The hosts note Jensen has said the exact same words.
• Refusing the fluke narrative. Ben calls TSMC an "unlikely success of grand scale." Morris's four-word reframe sets the tone for the whole episode: "Unlikely, in your opinion." Then: "It did exceed my expectations. TSMC's size and importance exceeded my expectation, but not by an order of magnitude."
• Honest attribution. Did he foresee the smartphone wave? "No, I didn't." "Maybe the business development guy... had a more detailed visibility than I did." On fabless, though, he gives no ground: "No, I saw it coming." Ben notes that in earlier interviews Morris sometimes credited luck on timing, but here he goes on the record with foresight. When David quotes Don Valentine on founding Sequoia — "I had an advantage. I knew the future." — Morris allows exactly one sentence: "Well, at least I had a glimpse of it."
• A chairman's sense of proportion. "Sometimes you have to let the CEO make his own mistakes and learn from them. Of course, not if the whole company is going down the drain. You have to interfere then, but only then." That is precisely how he waved through Rick Tsai's 2005 re-split of operations.
• Precision recall at 93. When a host says revenue took four years to recover after the dot-com bust, he recomputes it on the spot, quarter by quarter: started Q1 2001, recovered Q3 2003 — three years. When Ben suggests he founded a pure-play foundry because TSMC could "never catch Intel," he snaps back: "I didn't say that we'll never catch Intel." Ben: "Fair enough. Look where we are in 2025."

The key decisions (in his own words)

1997: personally answering a small customer's letter. NVIDIA was four years old, 50-60 people, near bankruptcy; Jensen's letter went through the post office to Hsinchu because TSMC's San Jose office never responded. Morris's first reaction was curiosity plus irritation — at his own salespeople: "I had always told our salespeople that we should never be negligent in talking to future customers, even if the customer seems to be a very small one." He remembers Jensen's audacity — the chip "would not only save the company, it will also make NVIDIA a major customer of TSMC" (the bar then: at least $50M a year) — "That was actually quite a bold statement." And: "his prediction came true" — top five within 2-3 years. Was the chip the RIVA 128 (David asks)? Morris can't recall the number, doesn't think it was called RIVA — just "the games chip."

2009: the comeback and the $100M with a 48-hour fuse. The 40nm crisis was a compound failure of yield, quality, and delivery; NVIDIA, likely the node's biggest customer, bore the brunt — and the previous CEO, siding with his quality director's "TSMC is not at fault," had offered zero. Back in the seat, Morris spent nearly half of his first 4-5 weeks on it: mastering the facts, working the intelligence ("I also knew that NVIDIA customers were after them. They had demands on NVIDIA too. So I used all the intelligence I could get."), and after weeks of study "worked out a number... I thought it was fair to both sides." The choreography ran to the minute: an email inviting himself to Jensen's house for the old ritual of salad and pizza (Lori made the salad; whether the pizza was delivered or homemade, he can't remember). Jensen's reply: "When do we discuss business then?" Dinner at 6:30; at 8:00 sharp Morris checked his watch — "Jensen, why don't we go to your study?" — and put down the offer: over $100M, 48 hours, no negotiation, or we go to arbitration. Jensen accepted within two days. The inner version: you probably didn't want to arbitrate with your best customer? — "No, I didn't want to." His verdict on the tale: "I know. I liked it too. That's why I included the story in my autobiography."

Fixing R&D at 8% of revenue. The root was a TI wound: 4.8%, and "my request was denied every time I raised it" when he asked for 5.5%. TSMC was running at 6-7%, renegotiated annually between the R&D director and the CEO. "I wanted to set a percentage of revenue number, so we don't have to argue every year… I want to make him at ease." The number itself was almost offhand, but grounded: "I said, oh, let's pick 8%, regardless of whether there's a recession or not." The endorsement came from the then-R&D director (then the department's number two), who "over the past 10-15 years" has repeatedly called it the best thing TSMC ever did for R&D — "that is what propelled our R&D effort."

2010: tripling CapEx on 28nm. "Sweet spot" was R&D's word first — "like a tennis racket, you want to hit the ball with the sweet spot of the face" (Morris's aside: he played tennis 40 years ago). Three input streams fed the decision: the market forecasting group, business development (positioned as strategic marketing), and R&D's technical case — "He gave me a lot of technical reasons. So I decided I would believe him." CapEx jumped from the prior decade's $2-2.5B a year to nearly $6B. The independent directors — more than three-quarters of the board, mostly from the US and England — didn't even wait for the pre-board dinner; they phoned the general counsel (an American, doubling as board secretary): "We want to talk to the chairman. We don't like this idea at all." Morris took the call about a week before the meeting and closed with: "Well look. I heard you, but I am still the guy that's responsible for the operation of the company. You need to let me go ahead with this one." Plus the admission: "You really can't convince anybody on something like this." The memoir cites Julius Caesar: "There is a tide in the affairs of men which, taken at the flood, leads on to fortune." — "I decided this 28-nanometer was going to be our tide. Our next tide, anyway. There will be others." 7nm was the next one.

2010-11: the Apple deal, end to end. One November day in 2010, the board meeting broke around 6 and Morris got home about 6:30. Sophie — who rarely meets him at the door — was at the door: her second cousin (same grandparents) Terry Gou, chairman of Hon Hai/Foxconn, had called that afternoon to say he was bringing an Apple vice president to dinner at 8. Morris's instinct: "It must be someone special for TSMC." The guest was Jeff Williams — not just any VP, but Apple's COO. Sophie simply had the family's Chinese chef (who doesn't cook Western food) add a few dishes — "The food was not important." Jeff pitched almost from the moment he sat down: "We like you to foundry our wafers." He did about 80% of the talking that night. Two moments of contact: Jeff offered to "let" TSMC have a 40% gross margin — Morris thought "my goodness, we're already at 45%" and said nothing ("this dinner was not the time to discuss price"), while registering that Jeff meant it generously; and when Morris assumed Apple wanted the soon-to-ramp 28nm, Jeff said "No" — 20nm. A surprise and a disappointment: the natural evolution after 28 is 16; 20 is a half step, and "a half step is a detour" — and R&D then couldn't develop two nodes at once.

Then came the financial planning with Goldman Sachs: no dividend cut, no new shares, bonds only — and after seeing the numbers, Morris decided to take only half of Apple's stated demand. The new BD chief C.C. Wei relayed "half" to Apple's lower-level procurement and brought back the response verbatim: "You must be crazy." Morris flew to Cupertino to explain the "prudent" financial planning to Jeff in person. Jeff was very quiet and offered one suggestion — cancel the dividend, "your shareholders will understand." Morris refused: the research showed about a third of shareholders cared intensely about dividends, and "our stock is going to drop like hell." Even at half volume, TSMC still had to borrow billions. In mid-February 2011 Jeff called, briefly: talks must pause two months, because "the highest level of Intel" had approached Tim Cook. Morris first assumed some eminence like Andy Grove; it turned out to be merely the sitting CEO — "the Italian," Paul Otellini. He wasn't worried: running down Apple's supplier criteria, he judged TSMC roughly at parity with Intel on technology (his own assessment), stronger in manufacturing, and far ahead on customer trust. A month later he emailed Jeff — he'd be in Silicon Valley anyway, could he drop by? Jeff: come, but I'll be away — "I have asked Tim Cook to see you." Morris: "This freedom of delegating his boss to see a visitor was a privilege that I seldom had in my career." Tim Cook took him through the cafeteria line, trays carried back to Tim's office, and settled the matter in one line: "There's nothing to worry about because Intel just does not know how to be a foundry." — which Morris read as the third criterion, customer trust ("a very short, but a very satisfactory answer"). Jeff Williams later came to set the price himself: TSMC had done its homework on cost and acceptable price; Jeff simply named a number derived from Apple's component-cost targets. Deal done, they went out to a three-star Taipei restaurant, where Jeff quipped: "If you didn't like the pricing, we'd probably be going to a McDonald's."

~2013-14: "the real shock." The cost of taking 20nm arrived on schedule: 16nm slipped, and Samsung — traveling light after losing the 20nm business — skipped to 16 and got there first. Morris heard Apple had placed its first 16nm orders with Samsung: "I got a shock. It was a real shock." TSMC had invested on the order of $10B-plus over several years for Apple's 20nm ("tens of billions" in the telling), counting on 80-90% of the 20nm equipment converting to 16nm — "if Apple's 16 goes to Samsung, where does that leave us?" He rejects the framing that he'd been played ("I wouldn't say that"); he simply emailed Jeff Williams to confront it. The reply came instantly: "Don't worry, I'll be there." The next week Jeff was in Hsinchu with the promise: as soon as your 16 is ready, we buy all our demand from you. About half a year later TSMC's own 16nm was ready, and most of Apple's 16nm demand came to TSMC. Today all of Apple's chips are made by TSMC (Ben's addendum).

People: rotating successors, and the Kissinger analogy. After merging the two operations groups in 2009, Morris first offered the BD directorship to Mark Liu, who ran 10,000 people — the point being to round out his marketing gap and make him a complete executive. Mark refused: you want me to run a department of sixty or seventy people? "That was the end of that conversation." The analogy in Morris's head: "Kissinger probably had a couple of hundred people reporting to him, whereas the Secretary of State had thousands of people all over the world reporting to him. And who had more power? Kissinger." — those sixty-odd people were charged with finding all of the company's future business. C.C. Wei took the job — "I thought he accepted it even delightfully" — and is today TSMC's chairman and CEO. The handling of the previous CEO was equally unruffled: more than one conversation, same job grade, salary, and bonus, a new title as president of new businesses (solar and LED), and a face-to-face assurance that he remained a succession candidate.

Core Insights (each: story → insight → effect)

1. Never neglect a small customer — the first-principles discipline of the foundry business

• Story: The San Jose office ignored a near-bankrupt NVIDIA; Jensen resorted to a letter through the mail; Morris phoned personally and showed up the next day.
• Insight: A foundry's customers are design companies — today's 50-person company on the edge of bankruptcy can be tomorrow's top-five customer. Sales organizations naturally allocate attention by current revenue and will systematically miss future giants; only the top can set and enforce the rule.
• Effect: NVIDIA reached the top five within 2-3 years, a relationship now worth many billions — and this very interview was Jensen's referral repaid.

2. Layoffs by performance rating are not credible; in talent-dense industries layoffs are usually bad math

• Story: In the early 1970s TI debate, Morris stood alone: "Seven hundred worst-performing people in the company. Who gave the 700 people the bad ratings? Seven hundred supervisors. Very subjective." In 2008-09 TSMC's sitting CEO cut 600-700 people under the "worst performance review" label — deliberately routing around Morris's requirement that any layoff go before a special board meeting.
• Insight: "The separation expense is usually about half a year, and it takes at least half a year to train a person. So if you need the people back within a year, you shouldn't lay off." An inexperienced CEO's knee-jerk in a downturn is to save money by cutting people. TSMC's alternative: probation capped at six months, after which most return to their jobs or transfer — "we almost never actually fire people."
• Effect: The disguised layoffs triggered two rounds of protests at Morris's doorstep and directly precipitated his comeback; institutionalized probation-over-firing became TSMC's HR bedrock.

3. Don't litigate with your best customer: a unilateral fair price plus a deadline beats haggling

• Story: The 40nm dispute had dragged on over a year; the predecessor offered zero; Jensen had proposed arbitration. Morris spent weeks mastering facts and intelligence, computed a number he judged fair to both sides, and — after the home dinner — offered it in the study: over $100M, 48 hours, no counteroffers.
• Insight: The customer relationship is worth far more than any one-time payout; the "deadline plus no-negotiation" structure forecloses anchoring wars; physically separating dinner from dealmaking (6:30 for food, 8:00 sharp for the study) keeps the relationship and the business each in their own room.
• Effect: Jensen accepted within two days; the business since has run to "many, many, many billions" (Ben).

4. Fix R&D at 8% of revenue and abolish the annual budget fight

• Story: TI's 4.8% and the perpetually denied 5.5%; TSMC's 6-7% and yearly haggling.
• Insight: Technology development is a continuous, cycle-spanning process — interrupting it costs far more than the accounting savings; a fixed percentage frees the R&D chief from budget defense and licenses big ideas.
• Effect: The R&D director spent 10-15 years calling it "the best thing we ever did for R&D"; both the 28nm and 7nm sweet-spot bets grew out of it.

5. Big capital decisions run on conviction plus responsibility, not persuasion

• Story: The 2010 CapEx triple; independent directors revolting before the board dinner; Morris synthesizing three input streams (market forecasting, BD, R&D) into a judgment, then falling back on governance rather than data.
• Insight: "You really can't convince anybody on something like this." The future cannot be proven; all you can offer is "operational responsibility is mine — authorize me this once."
• Effect: 28nm collided with the smartphone era (which Morris candidly admits he did not foresee — "maybe the business development guy had a more detailed visibility than I did"), cementing process leadership; 7nm reran the same "taken at the flood" playbook.

6. Top customers cannot be sold to — they come to you; your job is to have spent years becoming worth coming to

• Story: Ben asks whether he strategized for months to win Apple. Morris corrects him: "Strategizing is probably too strong a word. I mean just thinking, also knowing that we just can't do anything about it." "Apple is a very close-mouthed company. If you try to offer your services, they would just tell you to go away. They will come to see you when they are ready."
• Insight: With unsellable customers, the only strategy is to make your technology, capacity, and reputation the obvious first choice for when they're ready — and to plant relationship seeds years ahead (Goldman Sachs, the ADR, the board seat) so that financing and governance channels exist when the once-in-a-lifetime order lands.
• Effect: Apple walked in through Terry Gou's dinner and pitched TSMC ("We like you to foundry our wafers"); the multi-billion-dollar financing came together fast with Goldman's help; the endgame is all of Apple's silicon at TSMC.

7. Strategy matched to endowment: the pure-play foundry that never competes with its customers

• Story: The original pitch decks (both versions preserved at the TSMC Museum of Innovation — one for the Taiwan government, one for investors) carry "dedicated pure play foundry" as a literal bullet point. 【Hosts' post-game】 David's endorsement: TSMC is the only leading-edge player that competes with its customers nowhere in the value chain, so customers entrust it with their most secret designs. Ben and David add the honest historical footnote: at founding, pure-play was less foresight than "what could we win?" — TSMC, and Taiwan generally, had no chip-design capability, so competing with customers wasn't even possible; pure-play-as-optimum carries an "accident of history" component. Ben quotes Steve Jobs: "you can connect the dots looking backwards."
• Insight: Strategy should grow out of endowment — "we don't compete with our customers" was both an honest acceptance of a capability constraint and, as it turned out, the deepest structural trust advantage in the industry. Contrast (David): Samsung is Apple's number-one competitor, so its orders were never stable.
• Effect: Demand aggregates one-way into TSMC; even Apple's 16nm detour through Samsung ended with everything back at TSMC.

8. Fabless was a foreseen wait

• Story: In his last months at General Instrument, Gordie Campbell's ask collapsed from $50M to $5M for one reason: "I'm not going to build a fab." Morris: "See? That was the start for me that there will be fabless companies." Then Atmel — fabless — came asking GI to foundry wafers, and GI's idle fab immediately produced the fight over whose way the fab would run.
• Insight: Make the trend call first, then accept the long wait — the first 5-7 years ran on "someone else's worst orders" from IDMs, building competency, capability, volume, and literal fabs. Ben relays what Morris once told Jensen: growth was slow in the early years "because we were waiting for the customers to emerge"; Ben estimates the fabless takeoff landed within 12 months of Morris's forecast. As for the legend that the original plan stopped at Fab 2: "That was only the very initial plan. We were never going to stop there... I was a serious student of learning curve and I would never stop at just two."
• Effect: When fabless became the mainstream, TSMC was already in position; the crappy-business years were the price of admission.

9. The learning curve is an operating system, not a slogan

• Story: Around 1970 Bruce Henderson brought Bill Bain to TI; Morris worked with Bain for about two years turning the experience curve into a tool a semiconductor company could actually run on. "Explaining the learning curve theory is simple. But one will be foolish if one just takes the simple explanation and thinks that that's all it is."
• Insight: 【Hosts' post-game】 It is a work-backwards-from-the-endgame choreography: the goal is to be the largest-volume player when the game ends → drop prices to endgame levels early, even at a loss, to squeeze out competitors and aggregate all demand → scale then feeds back into cost and next-generation R&D. It drives pricing, strategic finance, and when to borrow (David: "almost Costco-like in the ballet"). The risk is just as extreme — "if you're off by 5%-10%, that's going to tank your entire profitability for that node generation, which is going to tank your free cash flow, which is going to mean you can't play the game in the next turn." The manufacturer must bet on its customers' end markets (how many next-generation iPhones, how big AI gets) — the counterintuitive madness of the business.
• Effect: Morris has used it his entire life ("which I used up to now"); once internalized, the TSMC story flips from accident to inevitability 【hosts' post-game】.

10. A half step is a detour — costing out the node roadmap

• Story: Apple wanted 20nm, not 28 or 16; and one layer earlier: "Only after 40, if we do the 40 well, can we do the 28."
• Insight: Nodes form a dependency chain, and inserting a half-step node cannibalizes the next full step — R&D then couldn't run two nodes at once (later it could), so taking 20 necessarily delayed 16, opening Samsung's leapfrog window. The buffer must be computed in advance: 80-90% of 20nm equipment was expected to convert to 16nm.
• Effect: Apple's first 16nm orders going to Samsung — "the real shock" — was the cost side of that ledger; Apple honored its return promise within half a year. Taking Apple was still right; the cost was real, and priced in beforehand.

11. Customer trust is the third competitive dimension beyond technology and manufacturing; "co-development" is its opposite

• Story: Tim Cook's "Intel just does not know how to be a foundry," which Morris reads as the third of his three criteria (technology, manufacturing, customer trust). "Before Apple became our customer, I knew a lot of Intel's customers in Taiwan... None of them liked Intel. None of them. Intel always acted like they were the only guys." Whereas: "We had to respond to each request courteously, which we do. Intel has never done that." On IBM's 1999 co-development offer for 130nm: "IBM still consider themselves to be the senior partner in any partnership they established... When we do that, we'll lose our ability to develop our own process."
• Insight: The resentment a monopolist accumulates among its customers is the challenger's largest hidden asset; and process sovereignty is non-negotiable — "co-development" hollows out the weaker party's capability.
• Effect: "We declined without having to think about it at all." UMC took the IBM deal and regretted it badly within a few years; Intel's lobbying bought only a two-month pause, and Apple came back anyway.

12. The deposit system: say the harsh word, never do the harsh thing

• Story: Through the chronically capacity-short 1990s, customers would "commit with a sentence while you commit real capital." Morris instituted customer deposits and coined the threat: "Everybody believes in the word 'confiscate.' It was first used by me." The San Jose salespeople, finally allowed to talk tough to customers, erupted "in an uproar of happiness."
• Insight: Incentive alignment runs on credible deterrence, not actual execution — through the 2001-03 recession no deposit was ever seized; deliveries were deferred until every customer had used theirs up.
• Effect: Capital discipline preserved, customer relationships intact — "But of course, really we never confiscated any money."

Strategy & Industry Judgments

Morris's own judgments:

• The eternal tension of the foundry model (from the GI/Atmel prototype): "The difficulty was you have to satisfy a lot of customers, and everyone wanted the fab to be run his way. But you can only run fab one way, which will satisfy more or less all the customers. The advantage of course is you have a lot of customers."
• Functional organization is the only answer for a foundry: In 1996 Don Brooks proposed dividing by fab (TSMC then had 3-4 fabs); McKinsey's two-month, multi-million-dollar answer was functional-is-best, exhibit Boeing. Morris's commercial bedrock: "we have almost the same group of customers. How do you divide up the company...? Customers move from one fab to another, the same customers" — splitting by fab "is like putting the 707 and the 737 and 757 in different divisions." Rick Tsai re-split in 2005 (waved through under let-the-CEO-learn); merging the groups was among Morris's first acts in 2009. He also drew the sales/marketing line: "Sales effort is just the tactical side with the customers. Marketing is the strategic side to the outside world." — and named the department business development rather than marketing "because it has a good translation in Chinese."
• The intelligence dividend of pure play: "Sitting in Hsinchu, being in the foundry business, I actually see a lot of things before they actually happen." Exhibit: Qualcomm's abrupt 1997-98 order shift (they never even named their previous supplier; Morris knew from other channels it was IBM) → instant inference that IBM Semiconductor's core business had collapsed → prediction that IBM would come asking → the composed refusal in 1999.
• There is no formula for value split: pricing looks first at cost, second at what the customer accepts; TSMC's 55-57% gross margin against customers' 70-80% — who sorts out the middle? "I don't get the privilege of sorting it out. C.C. Wei has the pleasure and the duty." The principle is to find the middle ground, but: "Every CEO says, ah, I worry about the long range. But in truth not everyone does." Most commodity suppliers get no choice at all — technology leadership is what buys the option to negotiate.
• The gross-margin alarm: "prices falling faster than costs, gross margin percentage steadily declining — you never want to see that" (item two on the 2009 comeback problem list: price-reduction schedules promised to customers that manufacturing costs failed to track).
• Dating Intel's decline: In the 1990s, hearing Intel was competing with you meant "trembling with fear"; "In 2011 Intel was no longer a name that, when you hear it, you would stand up and bow." He also corrects a host on the 28nm era: TSMC led absolutely among foundries, with TI and a few others in the top tier — "but not including Intel"; in 2010 Intel could not be ignored.
• Why the new businesses died: "The solar business could have been quite good. But China ruined it. They subsidized the hell out of it, and now they control the solar cell industry — prices pressed extremely low, still low today." LED: a smaller market, "and the patents are controlled by a few companies who will not let up at all." When Ben calls it ironic that TSMC hunted for new growth while sitting on what became a $600B/year IC market, Morris pushes back: "Why is it so ironic? I knew the IC business would be great" — though yes, they genuinely believed in solar and LED at the time.
• Governance details: board agendas mailed two weeks ahead; independent directors over three-quarters of the board, mostly American and British; an off-the-record dinner with them the night before each meeting. The ADR trades at a 20% premium to the Taiwan shares; converting local shares to ADRs requires case-by-case board approval, rarely granted — deliberately closing the arbitrage.

【Hosts' post-game】 (kept strictly separate from Morris's own words):

• Natural-monopoly endgame: a new fab costs ~$20B today, headed for $40B, $80B, $100B — the set of players able to deploy $100B on "a building full of machines" only shrinks. "Of course this will be the end state of this industry, is to have a dominant player." Twin engines: the largest-volume player can outspend everyone on both CapEx and R&D, so the leader only extends the lead (absent big strategic mishaps or execution mistakes); proprietary packaging — CoWoS for AI chips plus a separate mobile packaging technology — raises customers' double-sourcing barriers further. TSMC's annual CapEx bars track net income almost one for one — "they basically spend all the money."
• The TSMC/Intel fab-philosophy fork: Intel keeps upgrading existing fabs to the leading edge, permanently surrendering old-node capacity; TSMC's Fab 2 and Fab 3 have run since the late 1980s — fully depreciated, "almost like free to run," printing high-margin dollars off the long tail: replacement chips (a part in a 10/20/30-year-old system or car must be replaced with the same old-node chip — the business TSMC was built on) and CMOS image sensors (Sony sensors don't need 2nm; they need 40nm-and-up, down to 1-micron nodes). Ben's stance: "not better or worse... but different" — though the market-cap test favors TSMC's path.
• ARM as the hidden premise of the trillion-dollar outcome: had Intel/x86 kept its dominion, the world would have stayed vertically integrated and there would have been no foundry window for fabless companies; because ARM conquered phones, then computers, servers, and AI chips, the horizontal world — independent architecture, manufacturing, EDA, and design companies — could exist. David: "if ARM hadn't existed, there would've been nowhere else for this vector of innovation to go." The timing coincidence: TSMC, ARM, Synopsys, Cadence, and ASML were all founded within a year or two of each other in the mid-to-late 1980s. (Ben's edge case: AMD designing x86 chips made by TSMC — real, but not the counterfactual's main line.)
• The Hsinchu ecosystem is physically non-relocatable: partners share the neighborhood (Cadence, Synopsys, ARM buildings adjacent), customers are across the street (Qualcomm, MediaTek), and two universities inside the park feed PhDs straight into the ecosystem; coordination happens by walking across the street. David: it's as if all of Silicon Valley were packed into one government-sponsored industrial park. This frames the Arizona question — the overseas fab is not leading-edge, not high-volume, and severed from the ecosystem; TSMC built it for customer and government reasons, not economics (Ben's extension). The industrial-policy verdict: "this has to be the most successful government-funded industrial program in world history" — not a land-grant-universities scattershot but a "rifle shot."
• Moore's Law undefeated — on the demand side: the global semiconductor market went from $26B (1987, TSMC's founding year) to $527B (2024). David: "Moore's Law is undefeated." Ben's rebuttal: the technical definition expired long ago — packaging, software gains, and proprietary interconnects are now being folded into "Moore's Law" to keep it alive. David's clarification: he means demand — compute appetite doubling every 18-24 months with no sign of slowing after 50-60 years. "Integrated circuits are the fabric of our world today."
• The 7nm watershed: at 28nm TSMC was one of a handful of leaders; only around 7nm did it break away entirely (Ben's framing, unchallenged by Morris).
• 2nm on the ground (Ben's observation and inference): Phase 1 — about a quarter of the building — operating, three more phases unbuilt, small-batch trial production underway, ramp planned for H2 2025; product guesses are the next iPhone chip and a great many NVIDIA GPUs — "nobody says so, but..."

Deep Cuts

• "Quiet. Morris Chang is calling me." — In 1997 Morris, unannounced, dialed the number on the letterhead; Jensen picked up mid-argument with his staff, heard who was calling, and shouted the line at the room. They met in person the next day.
• The letter went through the post office — the alliance that rewrote semiconductor history began with a CEO's handwritten plea mailed to Hsinchu, because the San Jose office wouldn't return calls.
• "When do we discuss business then?" — Morris proposed "just salad and pizza"; Jensen's reply email asked when the real business would happen. David: did he ask who's paying for the pizza? Morris: "He did not ask."
• Checking the watch — at 8:00 sharp it was Morris who looked at his watch and said "Jensen, why don't we go to your study?" — then put $100M+ on the table with a 48-hour fuse.
• Mrs. Chang's breakfast diplomacy — in the second protest, ~25 people camped overnight in a small park a block from the house. Sophie, awake all night and checking the window, took a bodyguard to the market at 6 a.m. and bought Chinese breakfast — fried bread, steamed buns, soy milk, enough for 25-30 — and handed it out in the park. The protesters thanked her and promised not to march on the presidential palace that day. "All of this contributed to my taking back the CEO job." Morris, to the hosts: "Chinese fried bread — I don't know whether you ever had it or not; probably not."
• "He gave him zero." — the previous CEO's total compensation to NVIDIA for 40nm, against Morris's $100M+/48-hours.
• The tennis racket sweet spot — Morris asks Ben if he plays tennis; Ben: badly. Morris: "Very good. Forty years ago I was like you."
• "You must be crazy." — Apple lower-level procurement's verbatim reaction to "we'll take half"; C.C. Wei offered no comment and carried it home unedited.
• The delegated boss and the cafeteria trays — Jeff Williams had his own boss take the meeting (Ben: people say "someone on my team will see you," never "my boss will see you"); the CEO of the world's most valuable company and the godfather of semiconductors carried cafeteria trays back to the office and talked over lunch.
• "Don't worry, I'll be there." — the instant reply to the 16nm confrontation email; Jeff Williams was in Hsinchu the following week.
• Gordie Campbell's business plan — "No, it's all in my head." Three weeks later, instead of a plan: "I'm sorry I didn't send you anything because I don't need you anymore" — $5M "I can gather up very easily." The two had dinner about two months before the recording (November 2024).
• "For heaven's sake" — a rare flash of heat, recalling Atmel and GI fighting over how the fab would run: "General Instrument owned the fab anyway, for heaven's sake."
• The Don Valentine couplet — Valentine: "I had an advantage. I knew the future." Morris: "Well, at least I had a glimpse of it." (And yes — "Yeah, I knew him.")
• Michael Porter's 700 pages — Porter has written three or four 700-page volumes; "I have all of them." Porter once served on TSMC's board; there's a story about him in the memoir the interview didn't have time for.
• The conference room across the hall — every board meeting used to happen in the room across the corridor from where the interview was taped: "have you seen my conference room?"

Era & Industry Trivia

• The learning curve's family tree: around 1970, BCG founder Bruce Henderson cold-called TI CEO Mark Shepherd — TI being then the largest semiconductor company — and brought Bill Bain to Dallas to present the experience curve. Bain then spent ~3 days a week at TI for about two years (Mondays to Dallas, Wednesday or Thursday nights back to Boston to report to Henderson), with Morris as designated counterpart. Ben's footnote: the founder of BCG, the founder of Bain, and Morris Chang jointly developed the theory at TI — "how crazy is it?"
• Bill Bain's farewell: after those two years — "You are the first one I tell this to outside the Boston Consulting Group. I am leaving Boston Consulting Group to start my own consulting company." Morris, puzzled (Henderson plainly prized him), got the answer "the [...] imperative" (a word is missing in the transcript; roughly, a personal imperative) — the first time Morris had ever heard the phrase.
• McKinsey's multi-million-dollar answer: the 1996 structure fight — two months and several million dollars of fees for one sentence, "functional is best"; pressed by Don Brooks for one large functional company, McKinsey answered on the spot: "Boeing." Brooks had planned one year in Taiwan and stayed 6-7, by the end "somewhat tired of running the company."
• Kissinger vs. the Secretary of State: the power analogy in Morris's head when pitching the BD job to Mark Liu — headcount is not power. Ben's tag: certainly not the Secretary of State whose name you've forgotten. (Editor's note in the extraction: that would be William Rogers.)
• The golden founding window of the 1980s 【hosts' post-game】: TSMC, ARM, Synopsys, Cadence, and ASML all founded within a year or two of one another.
• A memoir 26 years in the making: volume two arrived in late 2024, in traditional Chinese, unpublished in the West; Karina Bao is translating it with funding from Tyler Cowen and Emergent Ventures, and the hosts prepped from the unpublished manuscript. The interview itself was Acquired's Hail Mary through "friend of the show" Jensen Huang, who personally made it happen.
• Rick Tsai's afterlife: the ex-CEO assigned to solar/LED found those businesses unworkable, resigned, and is today vice chairman and CEO of MediaTek.
• The pitch decks in the museum: the TSMC Museum of Innovation preserves both original decks — one presented to the Taiwan government, one to investors — with "dedicated pure play foundry" as a written bullet point (seen first-hand by the hosts).
• Taiwan's three science parks: the Tainan park is larger than the original Hsinchu park; Hsinchu has two universities inside the park, their PhDs absorbed directly into the ecosystem 【hosts' post-game】.
• Backpack sociology (David): the flight to Taipei was full of Google, Amazon, ARM, and Marvell backpacks — "the semiconductor version of the SF-to-Silicon-Valley commuter shuttle," fabless companies on pilgrimage.

Cross-domain Notes (PH)

This episode is a rarity for the PH domain's chip-geopolitics thread: first-hand testimony from the principal himself. On why Taiwan became the single point of the semiconductor world, this is not analyst inference but a 93-year-old founder on the record, plus the hosts' on-site observations — evidentiary value above any secondhand account. Three genuine crossings:

• fortress-america / chip reshoring: The hosts' on-the-ground case — the Hsinchu ecosystem (adjacent partners, two in-park universities, walk-across-the-street coordination) is physically non-relocatable, and the Arizona fab exists for "customer and government reasons," not economics (note: Ben's extension, not Morris's words in this episode) — sets physical upper bounds on CHIPS-Act-style reshoring. Morris's own "a half step is a detour" and the learning curve's endgame logic explain why process leadership cannot be subsidized into existence mid-race. If the Fortress America narrative assumes the US can rebuild leading-edge manufacturing quickly, this episode is among the weightiest counter-evidence.
• technate / technological monopoly structures: The natural-monopoly endgame derived from the learning curve plus the CapEx/R&D twin engines ($20B → $100B fab-cost inflation shrinking the eligible-player set) is a live industrial case of compute-and-manufacturing power concentrating into very few nodes — usable as comparative material for the PH domain's technocratic-concentration analyses (this is the hosts' post-game framework; cite with that layering).
• China industrial policy: Morris's first-person verdict on solar — "China ruined it": massive subsidies, crushed prices, eventual control of the industry, and the death of TSMC's solar venture — is an independent, named, interested-party data point for the PH domain's China-overcapacity narratives.

No crossover with the PH domain's strong narratives (deliberate-collapse, Pax Judaica); none is forced.

Pages Worth Creating

• Entities: Morris Chang（张忠谋） (person page: the TI → GI → ITRI → TSMC arc and decision inventory — this episode is the best source draft), Jensen Huang（黄仁勋） (shared with the NVIDIA series)
• Episodes: TSMC：纯代工模式发明者 (Acquired's TSMC company-history episode, this interview's companion piece), NVIDIA 之一：GPU 公司（1993-2006） (first of Acquired's four NVIDIA episodes, including a Jensen interview — the Jensen-side counterpart testimony)
• Concepts: 学习曲线（Learning Curve） (experience curve: the BCG-TI lineage, endgame-backwards pricing, the natural-monopoly corollary)

Source · acquired