How TSMC became the world's indispensable chip manufacturer

Executive overview

Every leading-edge chip in every iPhone, Nvidia GPU, and hyperscaler data centre is made by one company — TSMC — operating on an island whose sovereignty is disputed by China. Morris Chang founded TSMC at 56 with no equity, on a government mandate, in a country with no semiconductor heritage, and built it into the most profitable contract manufacturer in history.

The core insight: instead of competing to design chips, TSMC created the infrastructure that let everyone else compete — turning itself into the AWS of semiconductors and enabling the entire fabless industry.

Pure-play foundry manufacturing, executed at scale and sustained over decades, compounds into an impenetrable moat.

Morris Chang: becoming a semiconductor engineer

• Born 1931 in Ningbo, China; fled war three times before age 18 — the Second Sino-Japanese War, WWII, and the Chinese Civil War
• Arrived in the US in 1949 via Harvard (one year), then MIT to study mechanical engineering; completed both undergrad and master's in three years
• Failed his PhD qualifying exams at Stanford twice; had to go get a job as a junior engineer
• Joined Sylvania's semiconductor division in Ipswich, Massachusetts over Ford Motor Company — by a margin of $1 in salary
• Self-taught electrical engineering by studying Shockley's textbook in his hotel room each evening, then buying drinks for a senior colleague at the hotel bar each night to ask questions
• This was 1954: transistors existed but integrated circuits didn't; silicon wasn't yet standard; germanium was still the dominant substrate

Morris Chang at Texas Instruments

• Joined TI in 1958 — the same year Jack Kilby invented the integrated circuit there; TI was "Google plus Facebook" of its era in semiconductors
• First assignment: fix a failing IBM transistor line with near-zero yield; within four months raised TI's yield to 20%, double IBM's own first-party line
• That achievement made him the first manager at TI whose name the president knew; he built a 20-person germanium transistor development department
• TI sponsored his Stanford PhD (paying full salary); he finished in two and a half years
• By 1967 made general manager of a semiconductor division; invented learning-curve pricing with BCG: price low from day one to drive volume, accelerate yield improvement, and grab market share — made TI's IC business the largest and most profitable in the world
• Promoted to VP running the entire semiconductor business in 1972; widely expected to become CEO
• Passed over for CEO (likely due to racial discrimination); moved to VP of consumer products in 1978 — a mismatch that didn't play to his strengths
• Demoted to "head of quality and people effectiveness" in 1983; resigned shortly after, ending a 25-year career

TI's critical missed opportunity

• In 1980, IBM issued a secret RFP for the processor in what became the IBM PC, developed by a skunk-works group in Boca Raton
• TI — which had the IBM relationship going back to the 1950s — should have been the obvious choice
• Instead, Intel won with the 8088; the x86 architecture then standardised across all IBM clones; TI took the off-ramp from the next generation of computing
• Morris was already in consumer products and unable to influence the outcome

Transition: General Instrument and the decision to go to Taiwan

• After TI, joined General Instrument as president/COO — a financially-engineered conglomerate with no technology culture; left within a year
• Was weighing two options: become a VC, or take an operating role; chose neither conventional path
• Recruited by Taiwanese minister KT Lee to run ITRI (the government's Bell Labs equivalent) in 1985; colleagues, advisors, and his wife all told him not to go
• Arrived to find a government workforce in a non-democracy, scared of their new foreign boss; only meaningful existing asset was an outdated RCA technology-transfer license (two-plus generations behind the leading edge), which had been spun into UMC — also the ancestor of MediaTek

The founding of TSMC

• After one year at ITRI, Lee told Morris to start a semiconductor company and make it a world leader — "an offer I couldn't refuse"
• Morris had less than a week to produce a business plan; applied the question of what kind of company fits Taiwan's one strength (manufacturing) and avoids all weaknesses (design, IP, marketing, sales)
• Answer: a pure-play foundry — contract manufacturing only, never competing with customers
• This was a radical idea; Jerry Sanders of AMD had just said "real men have fabs," and every serious semiconductor company made its own chips
• Raised $220 million at a $0 pre-money valuation — Chang received no founder equity; the government owned 50%, Philips 28%, with the Taiwanese premier personally strong-arming the remaining business leaders to invest the rest
• Intel and TI both declined to be strategic investors or anchor customers; Philips was the only willing partner
• TSMC officially founded in 1987 — the same year ARM was established as a JV between Apple, Acorn, and VLSI Logic
• Early market: the "dregs" from IDMs — excess capacity they didn't need, and chips they were losing money on and wanted to offload

The fabless model and the flywheel

• The short-term market was unstable: IDMs gave TSMC work when they had no capacity, then took it back; chips outsourced to TSMC were often the ones losing money
• Morris's long-term thesis: IC designers at TI and GI wanted to leave and start companies but couldn't raise the capital needed to build fabs; TSMC removed that barrier
• This made TSMC effectively the world's largest semiconductor-focused venture enabler — not investing in startups, but building the manufacturing platform they would depend on
• Within a few years: Qualcomm, Broadcom, Marvell, Nvidia (founded 1993, raised only $20 million, never opened a fab) all built on TSMC
• The flywheel: more fabless customers → higher TSMC revenue → higher capex from 50% gross / 40% operating margins → more advanced process nodes → customers can address larger markets → more fabless customers
• In the early 2000s, 22 companies were at the leading edge of process technology; by the mid-2010s, six; by 2021, only TSMC and Samsung at 5 nm, with TSMC holding 90%+ share of leading-edge volume

The semiconductor value chain

• IP: ARM licenses instruction-set architectures; dozens of companies sell off-the-shelf IP blocks (USB, Bluetooth, memory controllers) that designers drop into chips without designing from scratch
• EDA: Synopsys and Cadence provide the software that chip designers use to create layouts — the Figma for silicon design
• Fabless companies: Apple, Nvidia, Qualcomm, AMD (post-spin-out of Global Foundries), and thousands of startups
• Equipment manufacturers: ASML (Netherlands) is the sole global maker of EUV lithography machines — ~50 per year, $200 million each, taking four 747s to ship, with ASML crews on-site to assemble and operate them
• Foundries: TSMC at the apex; ASML, founded in 1984 as a Philips JV, is now larger than Intel by market cap

Extreme ultraviolet lithography

• At 5 nm process nodes, the transistor features are smaller than the wavelength of white light (193 nm), making conventional optical lithography impossible
• EUV solution: drop molten tin into a vacuum chamber; hit it with a precision laser to create a plasma that emits extreme ultraviolet light; repeat 50,000 times per second
• Precision required is more exacting than the calculations used to navigate the Apollo missions — performed continuously at 50,000 pulses per second
• The wavelength is so short it is absorbed by all conventional mirrors; ASML had to invent new mirror types; a single German company makes the only laser capable of driving the process
• Nikon once competed with ASML on lithography; it was too hard and they gave up
• Intel invested $4 billion in EUV as early as 2012 but then decided to delay adoption in favour of improving existing DUV techniques — a textbook case of the innovator's dilemma; as of the original recording no Intel-manufactured chip used EUV

The Apple deal and Morris's third act

• Morris retired at 74 in 2005 and handed TSMC to his longtime lieutenant Rick Tsai
• Returned as CEO at 78 in summer 2009, announcing "golden opportunities ahead" — mobile and cloud
• The iPhone (2007) and Android (2008) created a computing paradigm shift: the new leading silicon companies were all fabless ARM designers (Qualcomm, Apple, MediaTek, Broadcom) — exactly TSMC's customers
• In the mid-2000s, Steve Jobs approached Intel CEO Paul Otellini about supplying iPhone chips; Intel considered the offer a lowball and passed; Apple initially used Samsung, then designed its own chips on the PA Semi acquisition (lineage: DEC StrongARM → Intel XScale → PA Semi → Apple Silicon)
• 2010–2012: TSMC committed $9 billion and 6,000 employees to build a dedicated fab for Apple in 11 months; Jeff Williams described it as a no-backup-plan bet for both companies
• iPhone 6 (2014) was the first fully TSMC-only product; it was the best-selling iPhone to that point
• Morris stepped down as CEO in 2013, remained chairman; fully retired at 86 in June 2018

Competitive position and powers

• Scale economies: the capex leader can buy more ASML machines and maintain a technology curve no competitor can afford to match; TSMC announced $100 billion capex over three years in 2021
• Process power: 40 years of manufacturing knowledge, yield expertise, equipment integration, and process recipe accumulation cannot be replicated by spending money; China's SMIC faces this problem more than the equipment access problem
• Switching costs: customers are so deeply integrated into TSMC's process design kits, IP libraries, and design-for-manufacturing rules that porting a chip to another foundry takes years
• Counter-positioning: IDMs could never have launched a pure-play foundry without cannibalising their own integrated margin model; the business model required an outsider
• Market position: 50%+ of all foundry revenue, 90%+ of leading-edge (5 nm) revenue, ~95% of foundry profit
• Moore's second law (Rock's law): fab costs double every four years — structurally forces consolidation to the most capitalised player

Geopolitical risk

• Taiwan's disputed sovereignty — China claims it as its own territory — is the single largest structural risk to TSMC's continuity
• If China annexed Taiwan and controlled TSMC, the West would lose access to leading-edge chip production; Ford already couldn't make F-150s during the 2021 chip shortage
• TSMC is building a $12 billion fab in Arizona and one in Japan — driven largely by government subsidies; Morris himself has said leading-edge manufacturing makes no business sense outside Taiwan's accumulated ecosystem
• ASML sales to China have been restricted (Dutch export licenses not renewed, likely under US pressure); China cannot yet acquire the EUV machines needed for leading-edge nodes
• The knowledge problem is binding: even with machines, operating EUV requires decades of operational experience that can't be transferred quickly
• Analogy to Toyota's NUMMI experiment: Toyota tried to replicate its production system in a joint US plant; it didn't fully work, even in peacetime

Intel's decline

• Intel's dominance of PC and server CPUs remains real but the growth is elsewhere — AI workloads run on GPUs and custom accelerators, not x86 CPUs
• Indecision has been costly: Bob Swan prepared to outsource Intel chip manufacturing to TSMC; Pat Gelsinger reversed course and wants Intel itself to become a foundry — neither direction has been executed cleanly
• ASML is now larger than Intel by market cap: a sole-source supplier of one machine category to essentially one dominant customer has more value than the company that once defined the industry
• The Rosenthal postulate, per the hosts: never make strategic decisions based on pricing — Intel passing on Apple, Ford not hiring Morris Chang over $1

Financials

• IPO on Taiwan Stock Exchange in 1994 at a $4 billion market cap; listed on NYSE in 1997
• 27-year revenue CAGR of 17.4%; market cap CAGR of ~20% — reaching $550 billion at the time of original recording in 2021, over $1 trillion by 2025
• 2020: $48 billion revenue, $20 billion operating profit; $17 billion of that ploughed back into capex
• 2021 capex guidance raised to $30 billion for the year, $100 billion over three years
• Raised prices ~20% in 2021 — the first price increases since the pre-Morris learning-curve era; the clearest possible demonstration of pricing power
• 28 billion in cash and cash equivalents on the balance sheet; effectively self-funding