Semiconductor legend Morris Chang left Texas Instruments (TI) in 1983. Two years later, Taiwanese government official Li Kwoh-ting recruited Chang to be the head of Taiwan’s Industrial Technology Research Institute (ITRI) — a government linked organisation.
When Chang started in 1985, he had hundreds of engineers working on integrated circuits (IC) under him. But they were far behind what established semiconductor companies were producing and selling. Two years later, under pressure from Li, Chang spun off the IC business under ITRI to form a new company called Taiwan Semiconductor Manufacturing Company (TSMC). And so began the story of the most valuable company in Asia.*
At the time it started, TSMC’s business model was completely novel: it was intended to be a “pure-play foundry”. Although certain chip manufacturers — like Samsung Semiconductor — produced chips for third parties, most manufacturers in the 80s made chips for sale under their own brands. And so if you were a chip design company, say, you would regard these chip manufacturers with some suspicion, since these same manufacturers may well end up competing against you. Worse, there was always the risk of a manufacturer stealing your chip IP.
TSMC offered something new. They wanted to be partners, not competitors with their chip-design clients. To that end, they never produced chips for sale themselves.
But the approach came with a catch: practically no fabless chip companies existed in the late 80s when TSMC started. All the chip companies were also manufacturers — with fully fledged fabrication plants! Companies like Intel and Samsung Semiconductor would only temporarily outsource their manufacturing if they didn’t have enough production capacity or if they were producing the chips at a loss. Hence, the only business that TSMC received for the first few years of its existence was this sort of work — producing the industry’s leftovers.
Things started to change by the late 80s. And Chang was ready for it — during his long tenure at TI, he had already noticed that many IC designers wanted to leave to start their own chip businesses. One major obstacle was the exorbitant cost of setting up their own fabs. TSMC could now offer these budding entrepreneurs a shot — with no need for exorbitant capex. And to prove that point, the graphic chip industry seemingly came out of nowhere in the 1990s, expanding at one point to a field of about 70 companies. The eventual category winner turned out to be NVIDIA.
These startups would all eventually become customers of TSMC (NVIDIA partnered with them in 1998) and TSMC would then use the revenue stream from this growing category of customers to reinvest into R&D. As their net sales ballooned from about US$1.5 billion to US$5 billion from 1998 to 2000, their discretionary capex kept pace — going from 870 million to US$2.4 billion during the same period. As Chang noted in his 2000 Annual Report, “we intend to play a role in facilitating important technology developments in the future. As part of this effort, we plan to continue increasing our R&D investment and aggressively invest in 300mm wafer fabrication capabilities to accommodate emerging market demand.”
TSMC had a flywheel going. The revenue collected from TSMC’s customers allowed them to invest heavily into new technology. In turn, more advanced technologies and improved fab capabilities would attract more customers. And so on. Additionally, TSMC was starting to prove itself as a principled and reliable partner in a trust-sensitive industry — it stuck to its promise of being a pure-play foundry, and never once entered into competition with its partners.
By the early 2000s, TSMC had caught up to the latest chip manufacturing technology. They had finally joined an elite club of 22 companies that could produce the most advanced semiconductors. By the late 2000s, the club had dwindled to 14 companies, and in the mid 2010s, it was down to six. As of press time, in 2022, there are only two cutting edge semiconductor manufacturers left — TSMC and Samsung. These two companies are the only ones with the ability to produce chips based on a 5-nanometer process. What is even more impressive is that TSMC controls an overwhelming 92% of this 5-nanometer market, while Samsung controls the remaining 8%.
How did TSMC become so dominant? A large part of TSMC’s current monopoly over the leading-edge process node may be explained by Chang’s decision to ride the hugely profitable and disruptive wave of mobile chips in 2009. Apple had famously approached Intel to produce the chips for their upcoming smartphone. Intel CEO Paul Otellini declined. He explained that it was a time “before the iPhone was introduced and no one knew what the iPhone would do.” He went on to say that the price Apple wanted for the chip was below Intel’s forecasted cost — assuming the unit numbers that were projected for the iPhone at the time.
This was the “golden opportunity” that Chang referred to in his return as TSMC CEO in 2009 (he left temporarily in 2005). If TSMC could capture the transition to mobile computing, they would have even more capital to reinvest into their R&D, further solidifying their market leadership.
In 2014, TSMC got the deal it wanted. The company would be the sole supplier of Apple’s mobile chips. This more than doubled TSMC’s revenues from $15 billion in 2010 to $55 billion in 2021. These increased revenues, over a ten year period, allowed them to reinvest even more into R&D — eventually resulting in TSMC’s dominant position at the 5nm market.
TSMC’s growth on the coattails of the smartphone revolution may perhaps be best illustrated through the following anecdote. In a February 2022 episode of the Compounding Curiosity podcast, Jon from the Asianometry Youtube Channel described an intriguing property of TSMC’s capex:
“There’s something really interesting happening with TSMC’s CAPEX and their future revenue. If you take the CAPEX for a year, multiply it by about three times, you can broadly predict revenues for the next year (emphasis added). And it makes sense generally because it generally takes a fab a year to get on track. So this trend has been going on basically for a decade.
In 2014, they spent $10 billion in CAPEX. A year later in 2015, they generated $30 billion revenue. 2019, they spent $16.6 billion in CAPEX, they generated $48.4 billion in revenues a year later, 2020. 2020, they spent $18.3 billion CAPEX. In 2021, they generate $56.8 billion. So in 2021, TSMC spent $30 billion in CAPEX. So does that mean in 2022, TSMC is going to make $90 billion in revenue? So, I’ve seen other conservative estimates saying like $72 billion and even their financial projections say its CAGR is going to be 10% to 15%.”
As of writing in 2022, very few challengers can enter the semiconductor market and build their own leading edge fabrication plants to compete with TSMC. They would need to match or exceed TSMC’s capex which is forecasted at US$100 billion for 2021 to 2023. For 2022 specifically, their expected capex totaled US$44 billion. To put that into context, oil company ExxonMobil plans to spend US$20 to US$25 billion capex per year from 2022 to 2027. And even if these challengers had the financial resources to spend this amount of money, they would still need the approval of their local governments.
Why? To build a fab with an 8,000 square metre clean room requires 24 swimming pools worth of ultra pure water every hour, which must be provided by a 120 kilometre network of ultra clean pipes. And 8,000 square metres is a small clean room. The largest clean rooms that TSMC operates in 2022 are 20 times larger, at 160,000 square metres. Few local governments would allow a fab to use 480 swimming pools worth of water every hour, operating 24/7 (to be fair, TSMC reuses each drop of water about three and a half times, but even so, the semiconductor industry consumes 10% of Taiwan’s water). Fabrication plants also use a lot of electricity. TSMC consumed an estimated 5% of Taiwan’s annual electricity in 2019, or about 14 terawatt-hours.
Now let’s say that you had to start all of this from scratch. First, you would need to spend a few billion dollars to build a fab — including specialised, vibration-resistant buildings. This also assumes that you are able to purchase incredibly precise manufacturing equipment, which includes $300 million machines like the extreme ultraviolet (EUV) lithography system from ASML. One problem here is that there is a waiting list for such machines, since TSMC has gobbled up more than 70% of ASML’s annual production. Next, you would need to negotiate quotas with your local government because you won’t be able to use that much electricity or water without getting approvals. And if you somehow manage to accomplish all of these things, you would still need to search for the handful of people who have experience managing the production of these leading-edge chips.
The difficulty in creating another TSMC-like business explains why the company is known as the “Silicon Shield” in Taiwan. It is a deterrent against a military attack from China because it is one of the most important global suppliers for chips — and in particular, an important supplier to the US. This in turn means that the US’s interests are currently aligned with Taiwan’s geopolitical interests. And for as long as that is true, TSMC will get priority when dealing with its local government.
Compare this case with a previous case you've read. What is similar? What is different?
Does this remind you of a similar case? If so, what is different there?
If you have any thoughts, feel free to comment in the member's forum.