TSMC
The world’s biggest chipmaker needs to move beyond Taiwan
Easier said than done
Taipei, a city of over 2m people, stopped moving at 1.30pm on July 17th.
Sirens rang out across the capital as residents rehearsed a civil-defence drill for a Chinese invasion.
Half an hour later, as phones buzzed to mark the end of the drill, the top brass of tsmc, the world’s largest chipmaker, gathered in a hotel in the city centre for their quarterly earnings call.
They brought good news: record profits, good progress on global expansion, a confident forecast of more.
The disjuncture was hard to miss.
As an island contemplated war, its most important company carried on with business as usual.
tsmc has fared exceptionally well in Taiwan, growing into a giant of the global technology industry.
But, for reasons both internal and external, it has embarked on a tricky expansion beyond its home.
In terms of revenue, tsmc produces two-thirds of all chips made by foundries—firms that manufacture semiconductors designed by others.
In the most advanced segment, including processors for smartphones, laptops and data centres, the company’s share exceeds 90%.
The artificial-intelligence boom is powered by the AI accelerator, a type of chip designed to train and run large language models.
Almost all of them are made by tsmc.
Nvidia, the world’s most valuable company, relies entirely on the Taiwanese firm.
So does its closest rival, amd, another chip designer.
Big tech firms like Alphabet, Amazon, Apple and Microsoft, each designing their own bespoke silicon, also turn to tsmc.
Swelling demand from tech firms has pushed tsmc to extraordinary heights.
Between 2014 and 2024 its annual revenues rose from $24bn to $88bn.
tsmc’s market value has reached $1trn, making it the eleventh-most valuable company in the world (see chart 1).
Since the launch of Chatgpt in November 2022, tsmc’s share price has more than doubled.
Unfortunately, the larger and more dominant the firm grows, the more it looks like a problem.
For most of its history, tsmc made all of its cutting-edge chips in Taiwan.
Although the firm has long operated a few overseas plants making less sophisticated chips, it has only recently begun shifting some of its most advanced manufacturing abroad.
Over the past five years it has embarked on a $190bn global expansion.
Of that, $165bn is going to the American state of Arizona, where the firm plans to operate six leading-edge factories or “fabs”.
Replicating tsmc’s precision on American soil will be difficult.
Shielding its core operations from geopolitical risk may prove harder still.
Despite its size and importance, tsmc avoids the spotlight.
In an interview with The Economist, Wendell Huang, the firm’s finance chief, admitted it prefers to “stay low profile” and is “still adjusting” to the more intense scrutiny it receives these days.
That aversion to publicity is woven into its culture.
tsmc was built to let customers shine while it stayed in the wings.
When Morris Chang founded the firm in 1987, chipmakers like Intel, amd and Texas Instruments designed and produced their own semiconductors.
Mr Chang made a contrarian bet: that a firm focused on manufacturing could outperform vertically integrated rivals.
By specialising, tsmc could create manufacturing processes that others could not equal.
By serving many customers, the company could achieve economies of scale and cut costs.
Its bet paid off, and an industry was transformed.
Evercore, an investment bank, estimates that in the first decade of this century more than 20 firms made leading-edge logic chips.
By 2012 only three remained: tsmc, Intel and Samsung, a South Korean electronics firm.
Today only tsmc is thriving.
Samsung has struggled with manufacturing issues at its most advanced fabs.
Intel, once the standard-bearer of the industry, has fallen behind in chip technology and is trying to build a foundry business as its sales dwindle.
The presence of foundries like tsmc made it easier for upstart “fabless” technology firms to focus on chip design and not worry about the manufacturing process.
That shift unbundled the industry, triggering an explosion of chip startups.
Jensen Huang, the boss of Nvidia, has said that his firm “would not be possible without tsmc”.
Many rivals tried to copy tsmc’s pure-play foundry model, but none of them managed to survive at the technological frontier.
Chips with everything
To understand why tsmc is so dominant, peer inside the fabs.
Moore’s law, the idea that computing power doubles roughly every two years, relies on shrinking the transistor, a microscopic electrical switch.
In 1971 a typical processor held 200 transistors per square millimetre.
Nvidia’s b200 ai chip, released in 2024, squeezes in about 130m—making for smaller, more energy-efficient computing.
Manufacturing such devices, with features measured in nanometres (nm, millionths of a millimetre), requires factories that cost $20bn apiece and are capable of producing around 25,000 silicon wafers, each containing many chips, per month.
tsmc’s fabs are enormous.
In Taiwan, it operates four “gigafabs”, each with at least four times the capacity of a typical factory.
Fab 18 in Tainan alone spans 950,000 square metres.
Its cleanroom, the sterilised factory floor where chips are etched layer by layer, occupies one-sixth of that area and is cleaner than an operating theatre.
Few other firms can match tsmc’s scale or precision.
Its yields, the share of chips on a wafer that meet quality standards, are exceptionally high.
Its employees are even more formidable.
Sassine Ghazi, chief executive of Synopsys, a maker of chip-design tools, says that tsmc’s manufacturing discipline is “unbelievable”.
That discipline is shaped by how the firm sees itself: as a manufacturer first and a technology company second.
Insiders describe a culture where employees are pushed to find efficiency gains even when systems are running smoothly.
Any improvement in one fab is swiftly replicated across all others.
Failures are hunted down obsessively.
The company’s finances reflect its rigour.
In 2024 tsmc’s net profit margin was 40%, more than three times the average for rival foundries.
Given its dominance, could it charge more?
Mr Huang of tsmc says the firm is often asked that.
His reply is that it succeeds only when its customers do.
What he does not say, but is true, is that the firm is paranoid about losing ground to rivals.
Pushing too hard risks driving clients away eventually.
Staying ahead is costly.
tsmc outspends its rivals by a wide margin.
TrendForce, a research group, estimates that in 2025 the company’s capital expenditure will amount to between $38bn and $42bn.
Samsung plans to spend around $3.5bn on its foundry arm, Intel between $8bn and $11bn.
tsmc directs spending where it matters most.
In 2025, 52% of revenue is expected to come from chips produced on its most advanced nodes—typically “5nm” and below, although that is a marketing term rather than a precise measurement.
By 2027 the share is expected to reach about 70% (see chart 2).
For years the firm’s growing dominance in chipmaking was hardly noted.
Few outside the technology industry had even heard of tsmc.
That began to change in 2019, when the first Trump administration sounded alarms over America’s reliance on Taiwanese chips.
Then came the pandemic.
Covid-19 closed factories, causing a global chip shortage that halted production in industries from electronics to cars.
As supply chains faltered, governments fretted.
tsmc was no longer a mere manufacturer.
It had become strategic infrastructure.
In 2022 President Joe Biden signed the chips Act, a $50bn package of subsidies and tax credits designed to boost chipmaking in America.
tsmc, which announced it would spend $12bn on a factory in Arizona in 2020, tripled that amount by late 2022.
The ai boom made American politicians even keener to bring more fabs to their country.
President Donald Trump, who derided the chips Act as wasteful, has pushed tsmc to make more chips on American soil by threatening tariffs.
In response, the firm pledged to invest another $100bn in Arizona.
C.C. Wei, tsmc’s boss, explained that a smaller sum “wouldn’t even make Trump open his eyes”.
Res in Arizona
Perhaps sensing that the firm can be swayed, some American officials have floated the idea of tsmc partnering with Intel to help jump-start the American firm’s foundry business.
Mr Huang responds, frankly, that the company is not interested.
He compares such a deal to pouring gasoline into a diesel engine.
tsmc’s processes are not compatible with Intel’s fabs, nor could the firm help run them.
The American government is now considering a stake in Intel.
Attempts to entice and bully tsmc to manufacture more chips outside Taiwan happen to align with the company’s thinking.
Increasingly, the firm seems too large for its island home.
s&p Global, a research firm, estimates that in 2023 tsmc accounted for 8% of the island’s electricity use.
By 2030 its share could rise to nearly a quarter.
Power is not the only limitation.
Stephen Su of the Industrial Technology Research Institute, a tech incubator in Hsinchu, points out that competition for engineers will grow fiercer as the working-age population shrinks.
Taiwan has few immigrants and a fertility rate of just 0.9, whereas a rate of 2.1 is necessary to sustain the population.
Land is another constraint. tsmc’s newest fab in Kaohsiung, which spans 79 hectares, was built on a former oil refinery and required a good deal of soil reclamation.
Finding sites for future fabs is becoming harder.
Expansion abroad brings fresh challenges.
The company is constructing new fabs in Japan and Germany, though its biggest bet is in Arizona.
One fab has begun production; two more are under construction, with another three to follow.
All told, Arizona could house as much as a third of tsmc’s most advanced capacity.
The arid environment has taken some getting used to.
Mr Huang admits the company was surprised by delays in the local permit process.
In Taiwan, he says, managers know what permits are required and “how to handle it”.
In Arizona, they assumed that, with Intel already operating plants nearby, local authorities would be familiar with fab construction.
Things have since improved.
The first Arizona fab is reportedly producing chips for Apple at yields comparable to those in Taiwan.
They are not cheap, though.
Lisa Su, the boss of amd, estimates that chips made in Arizona may cost up to 20% more than those from Taiwan.
The hope is that customers will be willing to pay a premium for supply-chain resilience.
Illustration: Carolina MoscosoA bigger worry is the difficulty of importing an engineering culture.
An executive at a chip-design firm likens TSMC’s operations in Taiwan to a machine with “its own heartbeat” and suggests that workers elsewhere do not possess the same rigour.
Stories of an intense, self-sacrificing work culture abound.
After a powerful earthquake hit Taiwan in 1999, employees quickly filled tsmc’s car park to assess the damage.
In the mid-2010s the company created a “nightingale programme” with engineers working the night shift as it raced to close the gap with Samsung and win Apple’s phone business.
Mr Chang has long maintained that transplanting such a work ethic across borders would be difficult.
The company is trying.
Around 1,000 Arizona-based engineers were sent to its “mother fab” in Tainan for 12 to 18 months of training.
A similar number of Taiwanese engineers later joined them in Arizona.
Over time, the number of transplants ought to fall.
Growing automation may help, too.
Jon Yu of Asianometry, a YouTube channel on business history, notes that tsmc still requires a pool of skilled technicians to operate its fabs.
Given higher labour costs and different work norms abroad, the firm may be more willing to automate.
Silicon shield
Geopolitics adds other pressures, which are harder to anticipate and respond to.
In Taiwan, tsmc is more than just a company.
Locals refer to it as “the sacred mountain that protects the country”, and in hard-to-measure ways it contributes to national security.
As long as China relies on tsmc for its chips, goes the theory, it will hesitate to attack the island.
That makes the firm’s global expansion politically delicate.
Becca Wasser of the Centre for a New American Security, a think-tank in Washington, says that Taiwan faces a difficult balancing act.
It needs to keep enough of TSMC’s operations at home to remain strategically indispensable, while accommodating allies who are eager to reduce their dependence on the island’s fabs.
The events of the past few years may have made the task harder.
Since 2019 asml, a Dutch firm that supplies technology to chipmakers, has been barred by that country’s government from exporting its most advanced tools to China.
That has hobbled Chinese firms’ ability to produce semiconductors at the 7nm level and below.
In November 2024 the American government tightened restrictions further, prohibiting tsmc from offering its most advanced services to Chinese customers.
Some analysts warn that cutting China off from the technological frontier could increase the risk of military action.
The stakes of this balancing act are immense.
In 2022 Mark Liu, then TSMC’s chairman, warned that a Chinese invasion would render the company’s fabs inoperable, since they depend on a “real-time connection with the outside world”.
Elbridge Colby, the Pentagon’s current policy chief, has gone further—suggesting that the fabs should be destroyed if China invades.
Either way, the result would be the same: a global supply chain in chaos.
tsmc sees little value in dwelling on hypotheticals.
Mr Huang believes markets are more pragmatic now.
Though he does not say it outright, the message is that investors have already priced in geopolitical risk and are less anxious about it.
He suggests that if war were to break out, it would not be confined to Taiwan but would engulf its neighbours too.
In that case, there would be much more to worry about than chip production.
That view may be too sanguine.
Even if the Arizona fabs function as well as expected, two-thirds of the world’s most advanced chips will still be made in Taiwan.
The firm’s most advanced manufacturing technology will remain there, as will nearly all its research and development.
Its overseas fabs, by design, will lag at least one generation behind.
Beyond geopolitics, other risks loom.
As tsmc pushes the limits of process technology, it could stumble.
The firm pulled ahead of Intel in 2015 when the American firm faltered on newer nodes.
Its rivals, though bruised, remain formidable.
Samsung recently signed a $16.5bn deal to supply advanced chips to Tesla, a lift to its foundry ambitions.
The ai boom, which has fuelled much of tsmc’s recent growth, could slow down.
Tariffs may sap demand for consumer electronics, which account for 40% of the firm’s revenue.
And the semiconductor industry is notoriously cyclical.
Firms tend to overbuild in good times, only to face a glut when demand slows.
tsmc’s current expansion is bigger than anything in its history.
tsmc’s greatest challenge may be the least tangible.
Mr Huang says outsiders often assume that chipmaking success is simply about money.
Yet, he notes, there are examples where governments have “poured money into a certain company” and still failed.
What they lack is tsmc’s edge, its resilience, discipline and relentless drive to improve.
For the company to keep growing, it must succeed abroad.
That will depend on whether it can export its exacting culture.
For a firm that has weathered typhoons and earthquakes, and lives under the shadow of war, going global may prove the harder test.
0 comments:
Publicar un comentario