Jensen Huang Twice Turned Down the TSMC CEO Job: I Already Have a Job
In 2013, Morris Chang seriously considered handing the 26-year-old TSMC to Jensen Huang. Huang replied with just five words: ‘I already have a job.’ Twice, the same sentence. At the time, TSMC's market cap was 10 times Nvidia's.
The video won't play here. Listen to the audio instead:
The argument · tap a timestamp to hear it
No money to tape out, so build it in a computer first
Nvidia's first two chips both failed, and by the end of 1996 the company had been cut down to just over thirty people. The normal process in the chip industry is tape out, test, fix the design, repeat — two or three rounds is common — but Huang only had enough money for one tape out. He took a third of his cash, about $1 million, and bought a machine that could simulate an entire chip in a computer, to find as many errors as possible before going to the real fab, and then decided the first version would go straight to mass production. His manager asked him how he knew it would work the first time. He replied: I know it must be right, because if it isn't, the company is gone. In August 1997 the third chip shipped, and in its first four months it sold over 1 million units.
Betting the lifeline on one fab, with no contract in hand
From 1998 on, Huang put almost all of Nvidia's chips on TSMC alone. At the time most fabless chip companies worked with two or three fabs at once, playing them off each other on price while keeping a backup. Nvidia went down this road thoroughly enough. More critically, Huang made this bet without a long-term contract — around 1997 the two sides discussed a formal manufacturing arrangement, but day-to-day orders were still placed one at a time, and TSMC made no commitment to reserve a certain amount of capacity long term. The bet rested on only two things: Morris Chang's verbal promise, and Huang's line that sounded almost like bragging.
When TSMC makes a mistake, Nvidia is the one that goes bankrupt
In 1998 a chemical step at the end of a TSMC production line went wrong, causing massive breaks across an entire batch of wafers; more than half the chips in Nvidia's shipment were ruined. Because it was hard to tell which ones were usable through normal sampling, Nvidia set up a temporary test line and hired a large crew to screen them one by one, and TSMC sent two production planners from Hsinchu to California for a full month. That quarter Nvidia's revenue fell by more than half from over $28 million, with a net loss of nearly $10 million, and it ultimately had to sell part of its equity to several downstream companies for cash. This is the fundamental risk of the design-only, no-manufacturing model: the fab is someone else's, the mistake is someone else's, but the money lost, the undelivered goods, and the brink of bankruptcy are yours alone.
A customer diverting a little business has to be written into risk factors
In 2002 TSMC's most advanced process generation had poor yields, Nvidia's new flagship shipped half a year late, and after launch it was nicknamed the hair dryer by review sites for its power draw and heat, while rival ATI used new products to overtake it. In March 2003 Nvidia announced an alliance with IBM, handing its next-generation flagship to IBM's new fab in New York state, in a contract estimated at over $100 million. The same day the two companies jointly issued a press release reaffirming their partnership, but three months later TSMC filed its annual report with the U.S. SEC, putting the matter into risk factors for the first time, stating it could not guarantee it would not lose Nvidia's business in the future. The previous year Nvidia had contributed about one-fifth of TSMC's revenue; that year its share fell from 20% to 15%. It turned out IBM's yields were also poor, and in the summer of 2004 Nvidia's terminal chips returned to TSMC, with the flagship following in 2005.
After a pizza dinner, a 48-hour ultimatum
After the 40nm yield incident, Nvidia demanded compensation. Then-CEO Rick Tsai insisted TSMC was not at fault and would not pay a cent, and Huang said if they couldn't agree they would go to arbitration. After Morris Chang returned in 2009, he spent half his time on this matter, talking for dozens of hours with everyone involved, and then spent three full weeks studying it behind closed doors with his staff. In early July he emailed Huang, saying only that he would come to Silicon Valley next week, and to be at his house at 6:30 p.m. on July 15 for salad and pizza, without a word about business. At 8 p.m. the two went into the study and closed the door. Chang gave his settlement proposal: over $100 million in compensation, TSMC's only and final offer, not open to bargaining, valid for only 48 hours, otherwise arbitration. Within two days Huang accepted. The figure was roughly 3% to 4% of TSMC's net profit that year, and about 3% of Nvidia's annual revenue.
Morris Chang wanted Huang to succeed him, and was refused twice
In 2013, when Morris Chang was choosing a successor for the second time, he seriously considered Huang, from outside the company. In his autobiography he listed three reasons: character, career, and expertise in semiconductors, plus the two had 16 to 17 years of personal friendship. Chang asked Huang to his face whether he was interested in being TSMC's CEO, spent about 10 minutes explaining expectations and compensation, and specifically stressed that the pay would certainly be more than Huang currently earned. Huang did not interrupt for a full 10 minutes, asked not a single question, and finally said just five words: I already have a job. A few weeks later Chang made another phone call, without changing his wording or lowering his terms, and Huang's answer did not change a single word. At the time TSMC's market cap was $90 billion, fully 10 times Nvidia's.
The AI bottleneck moved from inside the chip to packaging
Training a model is essentially doing repeated operations on massive amounts of numbers. The processor does the computing, the numbers are stored in memory, and in the past the two were packaged separately and soldered onto the same circuit board, with data transmitted through copper traces in the board. The traces were long and limited in number, and they became insufficient for AI's compute volumes. TSMC's approach was to use wafer fab processes to build a large piece of silicon as a base, place the processor and high-speed memory side by side on it, and put them in the same package. This technology is called COWOS. In 2011 the two companies together produced the first test chip using this packaging, and in 2016 the first mass-produced chip shipped, though few people paid attention at the time. In early 2023 TSMC could supply only about 8,000 COWOS wafers per month, while Microsoft, Google, Meta and a group of model companies were all waiting for Nvidia's cards, and their orders all got stuck at the same packaging step. TSMC expanded capacity roughly tenfold in three years, about 60% of it going to Nvidia.
A design flaw is Nvidia's fault, not TSMC's
In the summer of 2024, Nvidia's next-generation AI chip was found to have a design flaw before mass production, yields could not come up, and shipments were delayed by a quarter. Rumors soon circulated in the market that the two companies were blaming each other. On October 23, Huang was asked directly by a reporter in Copenhagen, and said flatly that these were all fake news, and made clear that the design problem was 100% Nvidia's fault, and that TSMC was helping them solve the yield problem and rebuild the chip at an astonishing speed. This partnership has lasted so long not because no mistakes were ever made, but because after problems arose, both sides rarely shifted blame to the other.
In their own words · checked verbatim
I know it must be right, because if it isn't, the company is gone.
我知道它一定是对的，因为它如果不对，那公司就没了。
Jensen Huang14:11
Please don't do this to us, because in the future we will be your biggest customer.
请你不要这样对我们，因为将来我们会是你们最大的客户。
Jensen Huang20:16
TSMC has learned to dance with 400 partners, while Intel has been used to reading the dance for so many years.
台积电已经学会了跟400个舞伴一起跳舞，而英特尔这么多年一直习惯读舞。
Morris Chang1:20:22
Figures
| Nvidia market cap | $5 trillion | 1:00 |
| Morris Chang's age when he founded TSMC | 55 | 8:08 |
| Nvidia's third chip shipments in first four months | over 1 million units | 14:11 |
| Nvidia's share of TSMC revenue (2001) | about 17% | 26:22 |
| 40nm settlement compensation | over $100 million, about 3% to 4% of TSMC's net profit that year | 48:39 |
| 2013 TSMC vs Nvidia market cap | TSMC $90 billion, 10 times Nvidia | 1:27:26 |
| TSMC COWOS monthly capacity in early 2023 | about 8,000 wafers | 1:19:21 |
| Nvidia's prepayment to TSMC at end of 2021 | nearly $7 billion | 1:13:18 |
| Largest customer's share of TSMC 2025 annual report revenue | 19% | 1:25:25 |
Glossary
- COWOS / Chip-on-Wafer-on-Substrate
- TSMC's advanced packaging technology that places the processor and high-speed memory on the same silicon interposer to solve the AI compute transmission bottleneck.
- Tape-out
- Running a chip design through a complete real production line to produce the first batch of wafers, used to verify whether the design can be implemented.
- Yield
- The proportion of chips cut from a wafer that are actually good, directly determining the cost of each good chip.
- Fabless chip company
- A company that only designs chips and does not build its own factory, outsourcing all manufacturing to a foundry.
- Silicon photonics
- Technology that introduces optical signals to transmit data between chips, replacing slow and hot copper traces.
How to listen
Founders and investors watching the semiconductor supply chain, AI compute bottlenecks, and long-term founder relationships; anyone who wants to know why TSMC is hard to replace and why Nvidia can't do without it.
The dinner opening and biographical setup from 0:00-3:00 can be fast-forwarded; the core mechanics start in the middle.