If you want to understand the next twenty years of geopolitics, you do not need a stack of foreign policy journals. You need to understand how a leading-edge chip gets made, because that process is the most concentrated, most fragile, and most strategically loaded supply chain in human history, and almost nobody with real power over it seems to grasp what they are actually holding.
Start with the chip in your phone, your graphics card, your car's driver-assistance system, and by now nearly every military system that matters. Designing it happens mostly in the United States, at Apple and Nvidia and AMD and Qualcomm and Broadcom, and increasingly in China at HiSilicon and a handful of others. Britain has Arm, which makes nothing itself but licenses the instruction set that runs on almost every phone on earth. All of that design work runs on EDA software, and EDA is essentially two American companies, Synopsys and Cadence, with Siemens EDA a distant third. Without those tools you cannot design a modern chip at all, and China has been trying to build a domestic version for over a decade while staying years behind at the leading edge.
The equipment is where it stops sounding like an industry and starts sounding like a single point of failure. The machine that matters most is the EUV lithography scanner, and it is made by exactly one company on the planet, ASML, in Veldhoven in the Netherlands. A standard EUV machine costs somewhere around 180 to 200 million dollars, weighs about 180 tons, and ships in multiple 747 freighters. It works by firing a high-power laser at tiny droplets of molten tin around fifty thousand times a second, collecting the 13.5-nanometre light that comes off the resulting plasma with mirrors made by Zeiss in Germany that are the most precise objects ever manufactured. Zeiss likes to say that if one of those mirrors were scaled up to the size of Germany, its largest bump would stand about a tenth of a millimetre high. ASML's newer high-NA machines, the ones needed for the nodes after this one, run closer to 380 million dollars each. The company ships on the order of fifty EUV systems a year, the global installed base is somewhere around 250, and every one of them goes to TSMC, Samsung, or Intel. China gets zero, by US-led export controls. And EUV is only the headline. You also need deposition and etch and ion implantation and metrology tools, photomasks, and photoresists, and in several of those categories a single Japanese or American or German supplier holds most of the market. The chokepoints are everywhere once you look.
Fabrication is where the concentration becomes almost absurd. TSMC holds around 70 percent of the entire foundry market by revenue, and at the leading edge the number is worse than that. At 7 nanometres and below, TSMC has somewhere north of 90 percent, and at the 3-nanometre class that the newest chips use, its share is around 95 percent. Samsung can technically build 3nm but has been fighting yield problems for years. Intel is clawing back with 18A and is still behind on volume. The overwhelming majority of the world's most advanced chips come out of a small number of TSMC fabs in Hsinchu, Tainan, and Kaohsiung. The last critical step, advanced packaging, is the same story, and TSMC's packaging capacity has been the binding constraint on how many Nvidia accelerators exist in the world. The shortage of H100s and then Blackwell parts was a packaging shortage as much as anything.
So look at the whole chain and ask where the single points of failure are. They are at nearly every layer. This is a supply chain that was optimized ruthlessly for efficiency over resilience across forty years, and it is now the backbone of every economy and every military on the planet.
I am going to be blunt about the part most analysts talk around. The Taiwan question is not really, at the level of great-power behaviour, about democracy or self-determination or the lives of 23 million people. Those things matter morally and they are not what is driving the decisions in Beijing and Washington. What is driving those decisions is that TSMC's fabs are a concentration of strategic capability with no parallel in modern history. Imagine every oil refinery on earth sitting on a single island a hundred miles off the coast of a far larger power that claims the island as its own. That is roughly the situation, except the refineries take a decade to rebuild anywhere else.
China's stated position is simple. Taiwan is Chinese territory, reunification is inevitable, and time is on their side. The American position is deliberately ambiguous and functionally clear: a Chinese takeover would be an unacceptable shift in the balance of power, and the reason is the silicon.
Here is what makes it a genuine nightmare rather than a normal flashpoint. If China invaded Taiwan and won, it would not actually get TSMC, not in any useful form. Those fabs run on a constant drip of equipment, spare parts, chemicals, and expertise from the United States, Japan, the Netherlands, and Germany, and under any invasion scenario that drip stops the same day. Within twelve to eighteen months the fabs turn into extraordinarily expensive paperweights as consumable parts wear out with no replacements. There are persistent and credible rumours, never confirmed, about the ability to disable the machines remotely. The engineers who actually make the whole thing work, thousands of them, would mostly flee. China cannot take TSMC. It can only deny it to everyone else, and from a certain cold strategic angle, denial might be an acceptable outcome on its own. If we cannot have it, nobody can.
This is why Washington has been pushing TSMC to build in Arizona, and the Phoenix fabs are now running. The thing nobody in Washington wants to hear is that those fabs will not match the Taiwan operation for years, possibly a decade, and the reasons are structural rather than fixable with money. Taiwan produces an enormous pool of semiconductor engineers who work brutal hours for salaries that would not fly in the United States, and the Arizona site has had well-documented culture clashes and has had to fly in thousands of Taiwanese engineers, which created its own political backlash. In Taiwan the suppliers are next door and a technician is on site in minutes; in Arizona everything gets trucked or flown in. Taiwan has more than a dozen fabs and a dense ecosystem of hundreds of suppliers built over decades. Arizona will have a few and is building the ecosystem from scratch. Arizona meaningfully reduces the catastrophic risk to the United States specifically. It does not remove the world's dependence on Taiwan and will not for a long time.
China's own push is the most expensive industrial policy in history, somewhere in the range of 150 to 200 billion dollars across funds and subsidies since 2014, and the results are more mixed than either Chinese state media or American threat-inflators will tell you. China is genuinely competitive at mature nodes, 28nm and up, where SMIC produces at scale for everything from power management to cars to consumer electronics, and it is strong in packaging and making real progress in memory. Where it is not competitive, and may not be for a long time, is leading-edge logic. SMIC's most advanced confirmed production is a 7nm-class process using DUV multi-patterning, a brute-force approach without EUV, running maybe 20,000 wafers a month and going mostly to Huawei, at low yields and high cost. Getting to 5nm that way is economically brutal and getting to 3nm is essentially impossible at volume. They do not have EUV and they are not getting it. The export controls, layered on in October 2022 and tightened in October 2023, again at the end of 2024, and again through 2025, are designed precisely to freeze China around the 7nm level, and the directional effect is clear even where the details are debated. China's leading-edge progress has slowed against where it was heading before the controls.
The uncomfortable question is what China does if it concludes it cannot close the gap on its own. The hopeful answer is that it leans into mature nodes, builds out its domestic ecosystem, and quietly accepts inferiority at the leading edge. The darker answer is that the gap itself becomes a reason to act on Taiwan, on the logic that a capability you cannot build, you take or you deny. I think the hopeful answer is the more likely one, but I do not hold that belief tightly, and the stakes are high enough that even a real chance of the darker one should frighten people.
All of this got an order of magnitude more urgent because of AI. Training a frontier model takes tens of thousands of leading-edge GPUs, and running it at scale takes more. Whoever has the most leading-edge compute compounds an advantage in AI, which feeds back into economic productivity, military capability, and scientific progress, which means access to advanced chips is now really about the rate of technological progress itself. Fall behind on compute and you fall behind on everything compute touches, which is starting to be everything. The money tells the story. Nvidia's data-center revenue went from about 15 billion dollars in its fiscal 2023 to 115 billion in fiscal 2025 and then to nearly 194 billion in fiscal 2026. That is not ordinary growth, it is a global scramble to build AI infrastructure before the competition does, and almost every dollar of it routes through TSMC's leading-edge fabs.
The scenario I find both most plausible and most dangerous runs like this. China stays locked out of leading-edge equipment. The AI capability gap between the US-aligned bloc and China keeps widening. Beijing concludes the gap will become permanent if the trend holds. Taiwan is aging demographically and its strategic value is finite, which is to say it matters most right now, while the fabs are still irreplaceable. And a window-of-opportunity calculation starts to form: act while the military balance in the strait is still close enough, before AI-driven autonomous weapons tilt it further. I am not saying this happens. I am saying the structural incentives point somewhere uncomfortable, and most people are not watching, because they have decided chips are boring or too technical to bother with. They are neither. This is the most important supply chain on earth, sitting in one of the tensest regions on earth, underneath the technology now reshaping everything else.
If you care about any other issue at all, AI safety, economic growth, the military balance, climate technology, medical innovation, you should care about this, because every one of those things now runs through a single island in the western Pacific.
For anyone who wants to go deeper, Chris Miller's Chip War is the best single overview, and Dylan Patel's SemiAnalysis is the highest signal-to-noise source on the industry there is. ASML's annual reports and TSMC's earnings calls are surprisingly readable once you skip the financial boilerplate, and the Rhodium Group and CSIS do the best public work on what China is actually achieving versus claiming. This matters more than almost anything trending on a given day. I have not figured out how to make people care about supply chain topology, but it is worth the effort to keep trying.