Acquired: TSMC founder Morris Chang

Category: Acquired Podcast (Deep Dives) · Duration: 175 min · ▶ Watch

Speakers: Ben Gilbert and David Rosenthal · Ben Gilbert, David Rosenthal · Dr. Morris Chang · Morris Chang

Switch language → 中文

Segments (28)

  • 00:00 · Introduction
    • Hosts introduce the episode and the context of their interview with Dr. Morris Chang in Taipei.
  • 03:22 · Meeting Jensen Huang
    • Morris Chang recounts his first interaction with Jensen Huang and how TSMC helped save Nvidia from bankruptcy.
  • 09:33 · The 40nm Crisis
    • Morris Chang discusses the 40nm node yield issues and his decision to return as CEO to fix them.
  • 12:44 · The Layoff Controversy
    • Morris Chang criticizes the previous CEO’s decision to lay off employees based on subjective performance reviews.
  • 17:08 · The 2008 Financial Crisis
    • Morris Chang reflects on the impact of the 2008 financial crisis on the semiconductor industry and management reactions.
  • 50:00 · Splitting TSMC’s Operations
    • Morris Chang discusses the decision by former CEO Rick Tsai to split TSMC’s operations into Advanced and Mainstream groups.
  • 52:15 · The Debate on Organizational Structure
    • Chang recounts disagreements with former president Don Brooks over divisional vs. functional structure and hiring McKinsey.
  • 56:00 · Recombining Operations and Seeking a BD Leader
    • Upon returning as CEO, Chang recombines the operations groups and searches for a new head of Business Development.
  • 59:00 · Appointing CC Wei as Head of BD
    • After Mark Liu declines the role, Chang offers the Business Development position to CC Wei, who accepts.
  • 63:38 · ServiceNow Ad Read
    • The hosts read an ad for ServiceNow, discussing their AI integration.
  • 66:01 · A Crucial Dinner with Apple and Foxconn
    • Chang describes a pivotal dinner at his home with Terry Gou of Foxconn and Jeff Williams of Apple.
  • 73:24 · Negotiating with Apple
    • Jeff Williams pitches for 100% of a fab’s capacity and proposes a 40% gross margin, which Chang pushes back on.
  • 76:44 · The Intel Threat
    • Apple requests TSMC to pause discussions with Intel, revealing Intel as a serious competitor for Apple’s business.
  • 81:00 · Financing the Apple Deal
    • Chang explains the financial planning required to build capacity for Apple, deciding to issue corporate bonds.
  • 86:00 · Customer Deposits and the Dot-Com Bust
    • Chang recalls requiring customer deposits during the dot-com bust to secure capacity, a tactic not used with Apple.
  • 1:40:00 · Apple’s Decision to Use TSMC
    • Morris Chang discusses why Apple chose TSMC over Intel for manufacturing iPhone chips.
  • 1:46:45 · Pricing Negotiations with Apple
    • Chang details the intense pricing negotiations with Apple’s Jeff Williams.
  • 1:50:00 · Samsung’s 16nm Lead
    • TSMC’s delay in 16nm development allowed Samsung to secure initial orders from Apple.
  • 1:54:00 · Qualcomm and IBM’s Foundry Failure
    • The rise of Qualcomm as a fabless customer and IBM’s struggles as a foundry.
  • 2:02:00 · The Learning Curve Theory
    • Chang explains the importance of the learning curve in semiconductor manufacturing.
  • 2:11:00 · The Birth of the Fabless Model
    • How the pure-play foundry model enabled the creation of fabless semiconductor companies.
  • 2:20:00 · Post-Interview Analysis
    • The hosts discuss the key takeaways from the interview, focusing on TSMC’s unique position.
  • 2:30:00 · TSMC’s Fabs and Scale
    • Discussion on the massive scale of TSMC’s fabs in Taiwan’s Science Parks.
  • 2:32:00 · The Fabless Model and Older Fabs
    • How TSMC’s original business plan evolved and the profitability of their older, fully depreciated fabs.
  • 2:36:00 · Learning Curve Pricing
    • Explaining the strategy of learning curve pricing to dominate market share and achieve economies of scale.
  • 2:42:00 · CapEx vs Net Income
    • Analyzing TSMC’s massive capital expenditures relative to their net income to maintain leadership.
  • 2:45:00 · Moore’s Law and Market Growth
    • The relentless demand for computing power driving the exponential growth of the semiconductor market.
  • 2:50:00 · Carve-outs and Wrap Up
    • The hosts share their personal recommendations and thank sponsors.

Specific Prices (13)

Timestamp Item Value Context
06:02 TSMC Revenue $1 billion TSMC’s revenue in 1995.
08:38 Nvidia Projected Revenue $50 million/year The revenue Jensen Huang needed to produce to be considered a major customer.
62:27 McKinsey Consulting Fee A couple million dollars The cost of hiring McKinsey to advise on TSMC’s organizational structure.
74:24 Proposed Gross Margin 40% Jeff Williams’ initial proposal for TSMC’s gross margin on the Apple deal.
74:38 TSMC Existing Gross Margin 45% TSMC’s standard gross margin at the time of the Apple negotiation.
74:39 Target Gross Margin 50% Morris Chang’s goal to push TSMC’s gross margin up to 50%.
82:27 TSMC ADR Premium 20% The premium at which TSMC’s American Depositary Receipts (ADRs) traded compared to the local stock.
94:53 CapEx for Apple Deal $10 billion The estimated capital expenditure required over a few years to build capacity for Apple.
2:17:00 Initial funding request for a new semiconductor company $50 million Gordy Campbell initially asked for $50M to start a new company.
2:18:00 Revised funding request for a fabless company $5 million Campbell revised his request to $5M because he decided not to build a fab, relying on a foundry instead.
2:41:56 Cost to build a new fab $20 billion The current estimated cost to construct a new leading-edge semiconductor fabrication plant.
2:46:13 TSMC Valuation $1 Trillion The potential value of a foundry built by reorganizing the industry away from vertical integration.
2:48:23 AAA Membership $100 The approximate cost of a AAA membership, which Ben highly recommends.

Bottleneck Claims (3)

  • [11:38] 40nm manufacturing yield was a major bottleneck for Nvidia.
    • Evidence: Nvidia was the biggest customer on that node and suffered from the low yield.
  • [79:28] R&D resource constraints prevented simultaneous development of multiple advanced nodes.
    • Evidence: Chang explains that TSMC could not develop 20nm and 16nm at the same time due to limited R&D capacity.
  • [1:50:00] R&D capacity limited TSMC’s ability to develop multiple nodes simultaneously.
    • Evidence: Chang states their R&D was not strong enough to do two nodes (20nm and 16nm) at the same time, leading to a delay in 16nm.

Predictions (2)

  • [09:00, 2-3 years] The Riva 128 chip would save Nvidia and make them a major TSMC customer.
  • [2:10:00, Long-term] The company with the highest volume will ultimately win in semiconductor manufacturing due to the learning curve.

Key Technologies (12)

  • Riva 128: An early successful graphics chip from Nvidia.
  • 40nm node: A semiconductor manufacturing process that experienced significant yield issues.
  • 28 nanometer: A semiconductor manufacturing node that coincided with the smartphone era and created immense demand for TSMC.
  • 28 nanometer: The node TSMC was entering production with during the initial Apple negotiations.
  • 20 nanometer: The node Apple initially requested for their chips.
  • 16 nanometer: The subsequent node that TSMC had to delay focusing on to meet Apple’s demand for 20nm.
  • 20nm and 16nm process nodes: Successive generations of semiconductor manufacturing technology, offering improvements in performance and power efficiency.
  • 2 nanometer process: The next generation of leading-edge semiconductor manufacturing nodes.
  • CMOS sensors: Image sensors used in cameras, manufactured on older semiconductor nodes.
  • CoWoS (Chip-on-Wafer-on-Substrate): TSMC’s proprietary advanced packaging technology crucial for AI chips.
  • EUV lasers: Extreme Ultraviolet lithography used for printing the smallest features on advanced microchips.
  • CDMA: A foundational cellular communication technology developed by Qualcomm.

Companies Mentioned (27)

Nvidia · TSMC · Qualcomm · Apple · Texas Instruments · Texas Instruments (TI) · McKinsey · Boeing · Broadcom · Foxconn (Hon Hai) · Intel · Goldman Sachs · ServiceNow · Element AI · Samsung · IBM · Boston Consulting Group (BCG) · UMC · ARM · AMD · Cadence · Synopsys · ASML · Sony · BCG · Bain · TI (Texas Instruments)

Notable Quotes (9)

Quiet, Morris Chang is calling me. — Jensen Huang (quoted by Morris Chang) @ 07:08

To be a major customer of ours, he would have to produce revenue for us of at least 50 million a year. — Morris Chang @ 08:26

Sometimes you have to let the CEO make his own mistakes and learn from them. — Morris Chang @ 58:02

There was no reason for Terry to just bring any Apple vice president to my home for dinner. It must be someone special. — Morris Chang @ 71:16

Everybody delights in the word confiscate. — Morris Chang @ 86:34

Intel just does not know how to be a foundry. — Morris Chang @ 1:43:00

If you didn’t like the pricing, we would probably be going to a McDonald’s. — Morris Chang (quoting Jeff Williams) @ 1:47:00

I don’t need $50 million anymore. I need only $5 million… I’m not going to build a fab. — Morris Chang (quoting Gordy Campbell) @ 2:18:00

Moore’s law is undefeated. — David Rosenthal @ 2:38:54

Key Topics

TSMC history · Nvidia early days · Semiconductor manufacturing challenges · Corporate leadership and management · TSMC Organizational Structure · Business Development Strategy · Negotiating with Apple · Semiconductor Manufacturing Nodes (28nm, 20nm, 16nm) · Capital Expenditure and Financing · Competition with Intel · Managing Customer Demand and Capacity · Foundry Business Model · Semiconductor Manufacturing · Learning Curve Theory · Apple-TSMC Partnership · Fabless Semiconductor Ecosystem · TSMC · Semiconductor Manufacturing · Learning Curve Pricing · Capital Expenditure (CapEx) · Moore's Law · Fabless Business Model

Takeaways

  • Strong personal relationships between founders can save companies and forge long-lasting partnerships.
  • Taking responsibility for manufacturing errors is crucial for maintaining customer trust.
  • Relying solely on subjective performance reviews for layoffs can be a flawed management practice.
  • A functional organizational structure proved more effective for TSMC’s foundry model than a divisional one.
  • Securing Apple as a customer required a ‘bet the company’ level of capital expenditure and dedicated capacity.
  • TSMC had to carefully manage its R&D resources, choosing to focus on 20nm for Apple at the expense of delaying 16nm.
  • Intel was a significant threat in the race to manufacture Apple’s custom silicon.
  • Strategic financial planning, including issuing corporate bonds, was essential to fund the massive expansion needed for Apple.
  • TSMC’s success was built on a foundation of customer trust and a strict policy of not competing with its customers.
  • The learning curve (experience curve) is a critical driver of cost reduction and competitive advantage in semiconductor manufacturing.
  • The pure-play foundry model enabled the rise of fabless semiconductor companies by drastically reducing the capital required to start a chip company.
  • TSMC’s ability to capture high-volume customers like Apple and Qualcomm fueled its learning curve advantage, allowing it to surpass integrated device manufacturers like Intel.
  • TSMC’s dominance is built on massive, continuous capital expenditure that closely tracks or exceeds its net income.
  • The strategy of ‘learning curve pricing’ involves taking early losses to aggregate volume, drive down costs, and ultimately achieve a monopoly-like position.
  • Older, fully depreciated fabs remain highly profitable for TSMC by producing essential components like CMOS sensors.
  • The relentless demand side of Moore’s Law—the world wanting twice the compute every two years—has driven the semiconductor market’s exponential growth.