The Fab, the Lithography Machine, and the Choke on Compute

How the chokepoint was built

TSMC did not happen by accident. It was founded in 1987 as a joint venture between the Taiwanese government, the Industrial Technology Research Institute, and private investors, under the direction of Morris Chang, a Texas Instruments veteran who had spent years arguing that a company should exist purely to manufacture chips designed by others rather than compete with them. That idea, the pure-play foundry, was not obvious at the time. Every major chipmaker of the era, Intel included, designed and built its own products. TSMC's bet was that if a firm never designed a competing chip, every fabless design house on earth, from Apple to Nvidia to Qualcomm, could safely hand it their most sensitive intellectual property to manufacture.

The bet paid off across four decades of relentless, capital-intensive execution: building fabrication plants that now cost tens of billions of dollars each, refining process nodes generation after generation, and keeping yields high enough that customers stayed rather than diversified. The result, as of the 2025 to 2026 reporting window, is a foundry market share estimated at around 70 percent, with some analyses putting the figure closer to 73 percent, up from about 69 percent in the final quarter of 2024. TSMC alone now accounts for roughly 30 percent of the Taiwan Stock Exchange's benchmark index, a single company standing in for a meaningful share of an entire national market.

ASML's monopoly formed differently but arrived at the same structural place. Extreme ultraviolet lithography, the technique required to print the smallest, most advanced transistor patterns onto silicon, took ASML and a consortium of research partners roughly two decades and tens of billions of dollars to make commercially viable. No competitor followed the same path to completion. By 2025, ASML held effectively 100 percent of the global EUV lithography equipment market and roughly 83 percent of lithography equipment sales of every kind, EUV and otherwise. There is no second supplier of EUV machines to call if ASML cannot ship one.

The economics of an irreplaceable machine

The financial numbers from 2025 show what happens when demand for artificial intelligence collides with supply that cannot expand quickly. TSMC reported full-year revenue of $122.42 billion, a 35.9 percent increase over 2024, with a gross margin of 59.9 percent, a level few manufacturers of anything achieve. ASML reported €32.7 billion in net sales for 2025, up 15.6 percent year on year. Both companies attribute the acceleration substantially to AI-chip demand: the graphics processors, accelerators, and custom silicon that large AI models require, all of which pass through TSMC's most advanced process nodes and, upstream of that, through machines only ASML makes.

That pricing power sits on top of geography that cannot be relocated on a normal corporate timeline. A single advanced fabrication plant takes years to build and requires a supply base, a skilled workforce, and a surrounding network of materials and equipment vendors that Taiwan and the Netherlands assembled gradually, not by decree. The United States, recognizing this, signed the CHIPS and Science Act on August 9, 2022, authorizing roughly $280 billion in total new funding, of which $52.7 billion was direct semiconductor appropriations: $39 billion for manufacturing subsidies and $13 billion for research and workforce development. It is one of the largest industrial policy interventions in recent American history, and it is explicitly an attempt to re-shore, in a handful of years, capacity that took Taiwan and the Netherlands decades to build organically. Subsidy alone does not substitute for the accumulated engineering knowledge, the yield-tuning experience, and the dense supplier networks that make TSMC and ASML hard to copy.

This is the economic thread that runs under the geopolitics: money follows the chokepoint. AI-model developers, cloud providers, and national governments are all, in effect, bidding for a slice of a supply chain with two structurally irreplaceable nodes, and the revenue growth at both companies in 2025 is a direct read of how tight that bidding has become.

A chip war fought with export licenses

The United States government moved early to use this chokepoint as a lever rather than simply try to out-invest around it. On October 7, 2022, the Commerce Department's Bureau of Industry and Security issued sweeping export controls on advanced computing chips and semiconductor manufacturing equipment bound for China, explicitly framed as an effort to slow China's military and AI development. That single announcement is generally treated by policy analysts as the opening act of the modern US-China chip war, and it worked precisely because Washington does not need to control every link in the chain. It needs to control the ones that cannot be substituted, and ASML's EUV machines, along with the most advanced deep ultraviolet tools beneath them, are exactly that kind of link.

The controls tightened through 2025. The US added 42 Chinese entities to its export-control Entity List in March and another 23 in September, and it required Nvidia to obtain a specific license before selling its H20 graphics processor to China at all. The pressure showed up directly in ASML's own numbers: China represented 33 percent of the company's 2025 revenue, but progressively tighter restrictions on deep ultraviolet lithography exports pushed ASML's own management guidance for China's 2026 revenue share down to roughly 20 percent. A government thousands of kilometers from Veldhoven, where ASML is headquartered, reshaped a fifth of a Dutch company's addressable revenue through licensing policy alone.

Then, by December 2025, the policy reversed in part: the Trump administration approved Nvidia's sale of its more powerful H200 chip to China, in exchange for Nvidia paying the US government 25 percent of an unspecified portion of the related revenue, an arrangement that itself signals how thoroughly export policy on these chips has become a tool of direct state extraction rather than simple restriction. Whichever direction the licensing swings, the mechanism stays the same: a government that does not manufacture the chip or the machine that makes it can still decide who gets to buy either one.

Taiwan itself is the risk

The geopolitical thread runs deeper than export licenses, because Taiwan's own economic dependence on chip manufacturing has become inseparable from its political exposure. Taiwan's integrated circuit exports totaled $184 billion in 2022, equal to nearly a quarter of the island's entire GDP. That is not a diversified economy with a strong chip sector; it is an economy substantially organized around one industry, concentrated overwhelmingly inside one company. Every strategic assessment of a potential conflict over Taiwan's political status now has to account for what happens to the global compute supply if TSMC's fabs, most of them physically located on the island, are damaged, disrupted, or cut off, deliberately or as a side effect of conflict.

This has produced what analysts sometimes call the silicon shield: the theory that Taiwan's indispensability to the world's chip supply gives outside powers, particularly the United States, a strong incentive to defend it, because the alternative is a shock to the global AI and electronics economy with no fast substitute. Whether that shield actually deters conflict is contested and unresolved, but the underlying fact it rests on is not: the world's dependence on TSMC is now a factor in how great powers calculate the cost of any move against Taiwan, in either direction.

ASML sits in a comparatively calmer geopolitical position, a firm headquartered inside a NATO member and the European Union, but it is not insulated from the same pressure. Its export licenses are effectively co-managed with Washington, since a large share of the components inside its machines originate from or pass through American technology, giving the US government real influence over what the Dutch company can ship to China regardless of what the Netherlands itself might prefer. Two small democracies, one an OECD founding member and one an island facing direct territorial pressure from a much larger neighbor, now sit at the physical and legal center of a technology every major economy depends on.

What re-shoring can and cannot fix

The CHIPS Act and its European and Japanese equivalents represent a genuine attempt to reduce this concentration, and some capacity has moved: TSMC itself has built fabs in Arizona, and other firms have opened plants in new locations with state support. But subsidy money buys buildings and equipment faster than it buys the tacit knowledge, workforce depth, and supplier density that took Taiwan and the Netherlands decades of continuous operation to accumulate. The 2022 to 2026 window shows re-shoring has begun, not that it has closed the gap. TSMC's share of the global foundry market rose during exactly the years the CHIPS Act was meant to be diluting it, and ASML's lithography monopoly has no credible second entrant on any near-term horizon.

The pattern echoes earlier chapters in this series: a chokepoint that concentrates wealth also concentrates political weight, and every attempt by an excluded power to build an alternative infrastructure runs into the same wall, that the incumbent already has the accumulated advantage that made it a chokepoint in the first place. What is different this time is the tightness of the loop. A printing press could be copied by a rival workshop within years. An EUV lithography machine cannot be copied by a rival nation within a decade, and the compute it enables is now the substrate for a technology, artificial intelligence, that both economic growth and military planning increasingly depend on.