On 3 April 2024, at 7:58 in the morning, the ground moved off Hualien, on Taiwan’s east coast. The United States Geological Survey measured magnitude 7.4; Taiwan’s weather agency put it at 7.2 on the local scale. In the science parks on the west side of the island the machines shut themselves down and workers walked out into the open. Ten hours later TSMC, the world’s largest chip foundry, announced that more than 70 per cent of its equipment was running again, over 80 per cent in its newest fabs. No power cut, no structural damage, and the extreme-ultraviolet lithography machines, the most expensive and most delicate tools in the building, untouched. By the end of the third day every plant was back at work. Some wafers in process were scrapped; the bill came to about NT$3 billion net of insurance, half a point of gross margin for the quarter, and the full-year forecast stood. The world did not notice. Outside the fabs the dead numbered fewer than twenty, almost all in Hualien, under rocks that fell into the gorges.

It was a successful test. It was also an easy one. Hualien lies east of the mountain range that splits the island. The fabs lie west, in Hsinchu, Taichung and Tainan. And the west side has a history of its own.
On 21 April 1935, at two minutes past six in the morning, a fault between Hsinchu and Taichung ruptured at magnitude 7.1. The epicentre was at Sanyi, in Miaoli. The dead numbered 3,276, the injured more than twelve thousand. It remains the deadliest earthquake in Taiwan’s history. Forty-five years later, in December 1980, the government opened a science park at Hsinchu, some fifty kilometres from that epicentre, modelled on Silicon Valley and set beside the island’s two technical universities. In February 1987 TSMC was founded there, with capital from the government and from Philips, on Morris Chang’s idea of making chips designed by other people.
On 21 September 1999, at 1:47 at night, the Chelungpu fault, in the same west-central region, moved at moment magnitude 7.6, which the USGS rounds to 7.7 and the local scale gives as 7.3. The dead numbered 2,415, the injured more than eleven thousand, the damage NT$300 billion. Chris Miller, in Chip War, tells what happened inside the fabs. The grid went down, two nuclear plants included. Morris Chang telephoned the government to get power before anyone else. Four fabs out of five were reconnected within a week, the fifth took longer, and the electronics market was back to normal within a month. TSMC was then one supplier among several.

It no longer is. In the second quarter of 2026 TSMC took 72.5 per cent of global foundry revenue, according to the research firm TrendForce, with its 3 and 5 nanometre lines running flat out for artificial-intelligence servers. The second producer, Samsung, holds under 7 per cent. TSMC’s plants in Arizona and Kumamoto cover a fraction of the volume. Miller does the sum in his book: if TSMC’s fabs slid into the Chelungpu fault, the shock would convulse the world economy, and the pandemic lockdowns give only a faint idea of what a single well-placed tremor would do. He wrote that in 2022, when TSMC’s share was smaller than it is today.
Modern fabs do not die by collapse. They are built for earthquakes and in 2024 they proved it. They stop for want of power, for want of ultrapure water, for machines that must be recalibrated one at a time. Every wafer inside at the moment of the shock is scrapped, and an advanced wafer spends two or three months in the fab: what is lost is a whole cycle, not a day. Chip buyers hold one to two months of stock. After that the car lines stop, then phones and computers, and first of all the processors for artificial intelligence, almost all of which come out of Taiwan. 1999 was one week without power at the share TSMC held then. One week without power at the share it holds now has never been seen.

And the sea? Taiwan has two doors for a tsunami. The first is to the south, the Manila Trench, where the South China Sea plate dives beneath the Philippines about a hundred kilometres from the island’s tip. It has produced no giant earthquake in the written record, and that is no comfort: it is locked and accumulating. In the worst case computed by Tso-Ren Wu and Bing-Hong Huang in 2009, a magnitude 9.3 rupture nearly a thousand kilometres long, waves reach 11 metres on the south-west coast; more recent models, with revised sources, give up to 5 metres for southern Taiwan. Either way the wave arrives within half an hour. The second door is to the north-east, the Ryukyu Trench. From there came the only certain tsunami in four centuries of Taiwanese records: on 18 December 1867, after a magnitude 7 earthquake off Keelung, the sea withdrew and returned in waves, flooded the coast and killed hundreds of people, about 580 by current estimates.
The chip parks are not on the beach. Hsinchu, Taichung and Tainan sit inland, kilometres from the coast, on the Strait side, where the water is shallow and wide and damps the wave. A tsunami does not flood them. It hits them sideways, through what has to stand on the water. The port of Kaohsiung, the island’s largest, through which wafers, specialty gases and chemicals come in and chips go out. And the liquefied natural gas terminals, at Yung-An near Kaohsiung, at Taichung and since 2025 at Taoyuan: Taiwan generates about 40 per cent of its electricity from imported gas, heading towards half, and holds a stock of ten to eleven days, according to its Ministry of Economic Affairs. If a wave puts the terminals out of action, the fabs lose power without the ground having touched a single machine. It is 1999 turned inside out: then the shock was on land and the grid fell from the land; here the grid would fall from the sea, and for weeks, not for one.
This is where Amitav Ghosh comes in. In December 2004, days after the Indian Ocean tsunami, the writer went to the Nicobar Islands on assignment, and in The Great Derangement he describes what he found at the Indian military base where his aircraft landed. The hangars and machinery stood inland, away from the water. The officers’ houses stood on the beach, and the higher the rank the closer the house to the sea. When the wave came, the commander’s house was the first hit. Ghosh writes that the base was no improvisation: the state had designed it, military experts and engineers had approved the site. It reflected what he calls the bourgeois faith in the regularity of the world, the conviction that highly improbable events belong to fiction rather than to real life.

The same book recalls the stone tablets on Japan’s north-eastern coast that read: do not build your homes below this point. Ghosh places them on the Fukushima coast and dates them to the Middle Ages. The source he cites, a report by Martin Fackler in The New York Times of April 2011, puts them at Aneyoshi, in Iwate prefecture, far north of the power plant, and dates the best-known stone to 1933, after that year’s tsunami; of the hundreds of stones along that coast, some are more than six centuries old. The point stands. The eleven households of Aneyoshi that had obeyed the stone watched the wave of March 2011 stop about a hundred metres below their homes. Elsewhere the coast had been built down to the sea, and below the oldest stones Japan had put a nuclear power station.
The sentence that comes to mind, that Taiwan built where it should not have, does not hold. Hsinchu has been a city since the seventeenth century. The whole island sits on the seam between two plates, and people have always built on dangerous ground, because dangerous ground is also good ground, with the plain, the water and the harbour. Tokyo, Los Angeles and Naples are where they are for the same reason. Taiwan did what everyone does, and did it well: its fabs rode out a magnitude 7.4 shock in three days.
The madness Ghosh describes is a different one, and it belongs to the world, not to the island. The world let the only factory for an indispensable thing stand on a known fault, one that had already killed twice in a century, because that was cheaper. It knew, and it knew in writing. Miller reports that TSMC’s customers were told the plants can withstand magnitude 9, and that since 1900 there have been only five earthquakes of that size anywhere: the claim has never been put to the test. It is the commander’s house on the beach, rebuilt at planetary scale with the accounts in order. And Ghosh adds the detail that weighs most: those living on the Nicobar beach were the wealthiest and best educated. The islanders lived inland. The more one trusts the regularity of the world, the nearer the front one stands.

The stones have been read. The fault is on the maps. What remains is the question of who decided that the improbable does not count, and what it will take to decide otherwise.

References
Miller Chris, «Chip War: The Fight for the World’s Most Critical Technology», Scribner, New York, 2022 (Italian edition consulted: «Chip War. La sfida tra Cina e USA per il controllo della tecnologia che deciderà il nostro futuro», Garzanti, Milano, 2023).
Ghosh Amitav, «The Great Derangement. Climate Change and the Unthinkable», University of Chicago Press, Chicago, 2016 (Italian edition consulted: «La grande cecità. Il cambiamento climatico e l’impensabile», translated and edited by Anna Nadotti and Norman Gobetti, Neri Pozza, Vicenza, 2017).
Fackler Martin, «Tsunami Warnings, Written in Stone», The New York Times, 20 April 2011.
Hsieh Min-ro, «The Earth Shook at Dawn», Archives of the Institute of Taiwan History, Academia Sinica, Taipei, 2018 (on the Hsinchu-Taichung earthquake of 21 April 1935; primary source cited: Taiwan Earthquake Record, Government-General of Taiwan, 1936).
Yen Ming-Hsuan, Lee Shiann-Jong, Ma Kuo-Fong, «Reconstructed the source rupture and strong ground motions of the 1935 Hsinchu-Taichung Earthquake (ML7.1) in Taiwan», National Central University and Academia Sinica, workshop paper, 2016.
Chung Ling-Ho et al., «The Maintenance of Earthquake Relics: Example of the Chelungpu Fault Preservation Park in Taiwan», National Museum of Natural Science, Taiwan, 2025 Joint Fall Meeting of Korean Geological Societies, October 2025 (Chi-Chi earthquake of 21 September 1999: Mw 7.6, 2,415 dead, NT$300 billion in damage).
United States Geological Survey, «Strong quake and aftershocks rattle eastern Taiwan», featured story, April 2024 (Hualien earthquake of 3 April 2024, magnitude 7.4).
TSMC, statements of 3 and 5 April 2024 on the recovery of its fabs after the Hualien earthquake, as reported by TrendForce and DigiTimes.
TrendForce, ranking of the top ten foundries by revenue, first quarter 2026 (12 June 2026) and second quarter 2026 (as reported by Focus Taiwan/CNA, 10 September 2026).
Wu Tso-Ren, Huang Bing-Hong, «Modeling tsunami hazards from Manila trench to Taiwan», Journal of Asian Earth Sciences, 36, 1, pp. 21–34, 2009.
Qiu Qiang et al., «Revised earthquake sources along Manila trench for tsunami hazard assessment in the South China Sea», Natural Hazards and Earth System Sciences, 19, pp. 1565–1583, 2019, open access. doi: 10.5194/nhess-19-1565-2019
Sun Yu-Sheng, Chen Po-Fei, Chen Chien-Chih, Lee Ya-Ting, Ma Kuo-Fong, Wu Tso-Ren, «Assessment of the peak tsunami amplitude associated with a large earthquake occurring along the southernmost Ryukyu subduction zone in the region of Taiwan», Natural Hazards and Earth System Sciences, 18, pp. 2081–2092, 2018, open access. doi: 10.5194/nhess-18-2081-2018
S&P Global Commodity Insights, «Taiwan vulnerable to LNG supply risks in the event of a maritime blockade», 30 May 2024 (Taiwan Ministry of Economic Affairs data: gas 39.6 per cent of 2023 power generation, 10–11 days of stock).
U.S. Energy Information Administration, «Taiwan Analysis Brief», April 2026 (Yung-An, Taichung and Taoyuan regasification terminals).