geopolitics; asia; china; united states; defence · 13 min
Sharegabrielcerezo.com/article/000034?lang=en
You might assume the most advanced chips in existence come out of the United States, or China, or one of the usual industrial heavyweights such as South Korea, Japan or Germany. In fact there is only one place on Earth where they can be made, an island of thirty-six thousand square kilometres with twenty-three million inhabitants, a hundred miles off the Chinese coast, which China considers its own and whose de facto independence has rested, since 1949, on American willingness to defend it.
Taiwan produces roughly ninety-two per cent of the world's most advanced logic chips, the ones below five nanometres. A single Taiwanese company, TSMC, takes around seventy per cent of global semiconductor foundry revenue, and in advanced chips its share is closer to ninety. Nobody else does what TSMC does at that scale and that quality. Samsung makes advanced chips, though with lower yields and a shrinking market share, while Intel has spent a decade trying to catch up and just needed a capital injection from the American government to avoid disintegration. Put like that it sounds like economics, but if China blockades or invades Taiwan tomorrow the lights of the twenty-first century go out, which makes it a civilisational vulnerability.
Start with what TSMC actually does and why nobody else can do it. It manufactures chips that other companies design (Apple, Nvidia, AMD, Qualcomm, Broadcom and dozens more) and sells nothing to the end consumer. The faster processor in each new iPhone was designed by Apple and built by TSMC. So were the Nvidia graphics cards that train the artificial intelligence models reshaping the global economy, and the advanced semiconductors the Pentagon needs for its latest weapons systems come, directly or indirectly, from the same place.
Advanced semiconductor manufacturing is probably the most complex industrial process humanity has ever developed. A modern three-nanometre chip holds billions of transistors in structures comparable in size to a handful of atoms. Making one takes extreme ultraviolet lithography machines that cost more than three hundred million dollars apiece and are produced by a single firm in the world, the Dutch company ASML, along with cleanrooms where the concentration of airborne particles is millions of times lower than in an ordinary room. It also takes thousands of engineers with decades of process knowledge you can't pick up from a book, and a supply chain of hundreds of ultra-pure materials and components shipped in from all over the world.
TSMC got on top of all this over thirty years of obsessive investment, hiring talent and refining the process a little at a time. Its founder, Morris Chang, an MIT- and Stanford-trained engineer who spent decades at Texas Instruments, set the company up in 1987 on what looked like a modest idea at the time, that TSMC would make the chips other people designed instead of designing and making its own. The "pure-play foundry" model meant designers no longer had to own their factories, and it allowed a degree of specialisation that, thirty years on, nobody has managed to copy. The global semiconductor industry now divides into those who design and those who fabricate, and at the advanced end the second group has, for practical purposes, one member.
The United States and Europe have seen the problem. America's CHIPS and Science Act put up incentives that drew in more than four hundred and fifty billion dollars of private investment. TSMC is building factories in Arizona, Intel received massive federal support and Samsung is expanding in Texas; the European Union launched a European Chips Act of its own, and Japan is subsidising the construction of TSMC factories in Kumamoto. All of this is necessary, though it won't be enough.
The underlying trouble is that advanced chipmaking doesn't relocate the way a car assembly line does. Putting up a building, installing machines and hiring workers is the easy part. What has to be replicated is a whole industrial base that took three decades to develop, lives in the tacit knowledge of thousands of engineers, needs a supply chain of overwhelming complexity and works to error margins measured in atoms.
Take TSMC's Arizona plant. It was announced in 2020 with production projected for 2024, and it slipped, with costs climbing along the way. TSMC had to bring in hundreds of Taiwanese engineers because it couldn't find local staff with the necessary experience, which caused friction with American unions. Once fully operational, the plant will turn out chips at four- and five-nanometre nodes, a generation behind what TSMC makes in Taiwan. The cutting edge, the three- and two-nanometre nodes entering mass production in 2025, stays on the island.
Even if everything goes right, with every planned factory in the United States, Europe and Japan built on schedule and reaching competitive yields, total capacity outside Taiwan will amount to no more than a fraction of global advanced capacity before 2030 or 2032. So the window of vulnerability is open, and it will stay open for at least a decade.
There's a quieter incentive at work as well. TSMC has little interest in moving its most advanced technology off the island, because from the Taiwanese point of view keeping cutting-edge production concentrated there works as an insurance policy. As long as the world needs Taiwan for its chips, the world has an existential interest in Taiwan not being invaded. Analysts call this the "silicon shield", the idea that TSMC matters so much to the global economy that it deters Chinese aggression, since a war would wreck the semiconductor production China itself depends on.
The silicon shield is an elegant concept, but I'd handle it with care. Chip dependency does not stop China from invading Taiwan. American military power does.
China receives more than half of Taiwan's chip exports, so a blockade or invasion would indeed be economic self-harm on a monumental scale for Beijing. History, though, is full of countries that made strategically irrational decisions out of national prestige, ideology or plain miscalculation. Japan attacked Pearl Harbor knowing it could not win a prolonged war against the United States, and Argentina invaded the Falklands well aware that it could not sustain a war against the United Kingdom. Russia, for its part, invaded Ukraine having underestimated both Ukrainian resistance and the Western response. Economic rationality has never been much of an insurance policy against military aggression.
What deters aggression is an adversary's capability to respond and its will to do so, and in Taiwan's case the adversary is the United States. The Seventh Fleet operates in the Western Pacific, and American bases in Japan, South Korea and Guam ring the theatre of operations. American naval and air superiority in the region has narrowed relative to China's military growth but is still real. The signal counts for more than the hardware, though. Washington's policy of "strategic ambiguity" on Taiwan has grown clearer over the years, and Biden went furthest, stating publicly that the United States would defend Taiwan in the event of an invasion.
China knows this. The People's Liberation Army has been building amphibious capability, modernising its fleet and developing anti-ship missiles designed to keep American carriers away from the strait. Still, sending an invasion force across a hundred miles of open sea against a mountainous island with a hostile population and a reasonably equipped military, with the world's foremost military power threatening to intervene, is something even a bold strategist would think twice about, or three times, since if the invasion fails the Chinese Communist Party does not survive the humiliation.
The United States has a second reason it cannot afford to lose Taiwan, and this one has little to do with chips. During the Cold War, the arc running from the Japanese archipelago down through the Ryukyus, across Taiwan, through the Philippines and on to Borneo was called the First Island Chain, and Taiwan is its centrepiece. For seventy-five years that arc has worked as a geographic lock on Chinese naval projection into the open Pacific.
Here the geography is merciless. China's eastern coast faces the Yellow Sea, the East China Sea and the South China Sea, and all three are, in effect, inland waters. Chinese warships heading for the Pacific have to pass through straits watched by American allies, whether the Miyako Strait (between Okinawa and the small Japanese islands of Miyako and Ishigaki), the Bashi Channel (between Taiwan and northern Luzon) or one of the lesser passages among the Ryukyus, and every one of them is under continuous surveillance. For a Chinese submarine trying to reach the deep Pacific, that adds up to a funnel lined with sensors, air patrols and allied submarines.
Taiwan is what holds the whole system in place. Its western coast looks towards China, while the eastern one looks straight out at the open Pacific, with deep water only a few kilometres from shore. If Taiwan falls under Chinese control, the People's Liberation Army Navy gets something it has never had in its modern history, deep-water ports with direct ocean access and no straits to cross or allies to slip past. Chinese nuclear submarines, which today operate out of Hainan and have to thread watched waters before they can vanish, could sortie from Taiwan's east coast into the deep ocean within hours, and tracking them would suddenly become impossible.
The knock-on effects come fast. Yonaguni, Japan's westernmost inhabited island, sits a hundred and ten kilometres from the Taiwanese coast, so Japan's southern flank is compromised the day after Taipei falls. The Philippines is flanked from the north, and the South China Sea, where Washington still contests access, closes. In practice the American perimeter falls back from the First Island Chain to the Second, anchored on Guam, nearly three thousand kilometres further back. American carrier groups that now operate with relative freedom in the Western Pacific would be doing so under direct missile threat from what would by then have become the frontier of China.
Tokyo grasped this before any European capital did. Shinzo Abe put it more bluntly than anyone else dared: a Taiwan emergency is a Japan emergency. He was describing, forty years late, something that had always been true, and it explains why Japan is rearming, reinforcing its bases in the Ryukyus, allowing American deployments to Yonaguni and pushing defence spending to levels that would have been unthinkable a decade ago.
Chips are the argument everyone understands. The lock is older and, I suspect, the more decisive of the two. Taiwan happens to combine both reasons the United States will defend it, the industry the world depends on and a geography without which American dominance of the Pacific doesn't hold.
Dependence on TSMC shapes decision-making in every major power. For the United States, losing Taiwan would cost strategic credibility in the Indo-Pacific and, on top of that, access to TSMC, which is where the technological edge behind its military supremacy comes from. The chips American weapons systems run on, the artificial intelligence feeding targeting platforms like Palantir, the processors running satellites and command-and-control networks all come from Taiwan. A war in the strait would therefore be a systemic event, the kind that reshapes the global economy and the balance of power in ways no model can anticipate.
China faces the same problem from the other side of the mirror. If Beijing invades Taiwan and destroys TSMC's factories in the process, it loses the chips its own economy depends on. Capturing them intact would be technically near-impossible, given how fragile the facilities are, and even then it would need decades to run them without the Taiwanese engineers, who would presumably flee or be evacuated. That leaves a blockade without invasion, which would trigger the very global economic crisis Beijing wants to avoid, since every affected country would line up behind the United States, so force doesn't solve the semiconductor problem by that route either.
In parallel China has tried to build domestic capability, through SMIC and massive state investment in semiconductors, with real but limited results. SMIC can fabricate chips at mature twenty-eight-nanometre processes and, according to some sources, has managed to produce seven-nanometre chips. The gap with TSMC at three- and two-nanometre nodes is enormous, and American export restrictions that keep China away from ASML's EUV lithography machines make closing it very hard. China can make chips for its basic needs, but not the ones it needs to compete in artificial intelligence, high-performance computing and cutting-edge military systems, and that dependency is Beijing's greatest strategic vulnerability, with no fix in sight.
An analogy does the rounds in strategic circles, that semiconductors are the oil of the twenty-first century. Chips, like oil in the twentieth century, are an essential resource for all advanced economic and military activity. Their production is concentrated in a few places, which brings strategic vulnerabilities of its own, and whoever controls the sources of production holds a first-order factor in geopolitical power.
The analogy breaks down at one point. Oil is a natural resource pulled out of the ground, and a country that loses one source can go looking for another; Saudi oil can be substituted, at a cost, by American shale or Norwegian crude. Advanced semiconductors are a manufactured product of technological complexity without precedent, made by a single company with three decades of accumulated advantage in one specific spot on the planet, and there is no substitute on offer and no alternative well to drill.
That is how the world lives today. The phone you are reading this on and the server processing this page depend on an island in the Western Pacific that an authoritarian government claims as its own and that a single company keeps running, and so does the artificial intelligence system assisting medical research, or the processor guiding a precision missile.
Morris Chang, who retired from TSMC in 2018, built what may be the most strategically important company in history. Others are larger and more profitable, but none would be so irreplaceable within any timeframe that matters. If TSMC stopped working tomorrow, the artificial intelligence industry would grind to a halt and smartphone production would collapse; Western defence supply chains would go into crisis, and the global economy would take a shock it would need years to recover from.
Reshoring is necessary and diversification urgent, but to be honest about it, the window of vulnerability will last at least a decade. For that decade, peace in the Taiwan Strait, and with it the stability of the global economy and the supply chain everything rests on, will hang on what it has hung on since 1949: American carriers continuing to sail the Western Pacific, and Beijing believing that using them is not a bluff.
The whole of technological civilisation is propped on a narrow ledge, and what holds the ledge up is a single fleet.