The Two Chokepoints Inside Every Chip

Western coverage — Bloomberg, the Financial Times, the Wall Street Journal, Reuters — reads the triad as a national security emergency.

The argument: concentration in Taiwan is an unacceptable single point of failure; export controls on ASML tools and AI chips are the price of keeping the lead.

The answer: reshore. Subsidize fabs at home. Deny the machines to Beijing and outrun it.

The risk is dependence. The cure is redundancy — inside allied networks, on allied timelines.

Eastern coverage — Xinhua, the Global Times, the South China Morning Post, Nikkei Asia — reads the same facts as a containment story.

The argument: the controls prove the West fears competition it cannot win on merit; SMIC's progress and Huawei's return show the blockade leaks.

Chinese graphite and gallium controls are framed as symmetrical, lawful countermeasures — not escalation.

The risk is Western overreach. The cure is self-sufficiency: domestic lithography, domestic EDA, domestic everything.

The South — The Hindu, Al Jazeera, the Daily Maverick, Jakarta Post — reads the chip war as someone else's fight with Southern casualties.

The question here isn't who wins the 2nm race. It's who gets priced out of compute, and whose minerals feed the furnaces.

The DRC's cobalt, Indonesia's nickel, Mozambique's graphite: the South digs, the East refines, the West designs. The midstream, again, is where the value — and the veto — lives.

Every export control lands here as a price spike. Then as a delay. Then as a lecture about rules.

How a monopoly gets built: the thirty-year road to one machine

Monopolies are usually built by crushing competitors. ASML's was built by outlasting physics. Extreme ultraviolet lithography — the process at the heart of every advanced chip — requires light with a wavelength of 13.5 nanometres, so short that it is absorbed by air itself and by every known lens material. The machine that generates it fires lasers at droplets of molten tin fifty thousand times a second, catches the resulting plasma flash in mirrors polished to atomic smoothness, and projects chip patterns in a vacuum chamber cleaner than interplanetary space. There was no moment when a rival could simply decide to compete. There was only a three-decade engineering siege, and one Dutch company was willing to lay it.

The siege had funders. In the early 2010s, with EUV still a laboratory promise burning cash, Intel, Samsung, and TSMC — three companies that compete ferociously on everything else — jointly bought into ASML's development program, effectively taxing themselves to keep the project alive. It was an extraordinary admission: the industry's future depended on a single supplier succeeding, so the customers financed the supplier. No government ordered it. The market, left to itself, chose concentration — because the physics left no alternative, and the economics of a ten-billion-dollar R&D program left no second entrant.

This is what the export-control debate keeps missing. Washington's restrictions on ASML's tools are often described as America "denying" China the machines. But America never built the machines. A Dutch company did, with German mirrors from Zeiss, American light-source technology from San Diego, and capital from three chipmakers on two continents. The chokepoint is not American property. It is a Western consortium's thirty-year bet, now repurposed as a geopolitical weapon — and every use of the weapon reminds the rest of the world that the consortium can be aimed at them next. Monopolies sanctioned by physics invite something physics cannot prevent: the political decision to route around them, whatever the cost, whatever the decade it takes.

China's answer is the longest game on the board. Domestic lithography programs, pursued under sanctions pressure, are attempting in years what took ASML decades — with the crucial difference that the target is now visible. The physics is published. The engineering path is mapped. What remains is the grinding work: the supply chains, the metrology, the yield learning that no espionage can shortcut. SMIC's demonstrated 7-nanometre-class output, achieved without EUV, is the proof of concept that the blockade leaks at the edges. The honest assessment — the article's word, "contested" — cuts both ways: contested progress is still progress, and every year of export controls is a year of funded, focused, state-backed catch-up. The monopoly was built by patience. It will be attacked with patience.

The foundry gamble: why a fab takes five years and talent takes twenty

The reshoring wave looks, on a map, like an answer. Fabs rising in Arizona, in Kumamoto, in Dresden — concrete proof that the West heard the lesson of 2021. But a fab is not a factory in the ordinary sense. It is the most complex manufacturing environment humans have ever built: tens of thousands of process steps, each held to tolerances measured in atoms, running around the clock in cleanrooms where a single dust particle is a catastrophe. Pouring the concrete is the easy tenth. The other nine-tenths is recipe, rhythm, and people — and none of it can be hurried.

The people are the binding constraint, and the industry says so openly. A leading-edge fab needs thousands of process engineers who have internalized, through years of apprenticeship, how a specific tool behaves on a specific Tuesday. Taiwan spent four decades building that workforce: the vocational schools, the university pipelines, the culture in which a fab engineer is a prestige profession. Arizona can import the tools. It cannot import the forty years. TSMC's own experience staffing its overseas fabs has been a public education in this arithmetic — the machines travel in crates, the mastery travels in careers.

This is why the "five years" in the article understates the gamble. Five years gets you a building with tools inside. What it does not get you is the yield curve — the slow, expensive climb from first silicon to profitable production, during which every wafer is tuition. Samsung, with decades of memory-chip mastery, remains a distant second in logic foundry precisely because logic yields are a different discipline. Intel, with all its engineering heritage, is still trying to become a foundry at all. The new fabs in the desert and the German plain are bets that money can compress time. The industry's history suggests money can buy everything except the learning.

And the deepest irony of the reshoring wave sits in plain sight: every advanced line on earth, wherever it is built, still needs ASML's machines. Arizona's fabs do not escape the first chokepoint; they relocate the second one. The supply chain's geography changes. Its topology — one supplier at the top, one island in the middle — does not. Reshoring, as currently designed, is redundancy of location without redundancy of control. That may be enough for the war-gamers. It is worth asking whether it is enough for anyone else.

The materials veto: graphite, gallium, and the midstream weapon

If the chip war has a most underrated battlefield, it is the one nobody photographs: the midstream. Mines are photogenic. Refineries are not. But the article's graphite number — 90 percent of battery-grade refining in China — describes a structural fact about the modern economy: the hole in the ground is not the power. The furnace is. Mozambique can dig graphite. Madagascar can dig graphite. Brazil can dig graphite. None of them can turn it into battery anodes at scale, at cost, tomorrow. That capability — the chemical engineering, the energy-intensive processing, the environmental tolerance for doing it — was built in China over twenty years, and it is now a veto.

Beijing's export controls on graphite, gallium, and germanium are best understood as a demonstration, not just a retaliation. Gallium and germanium are produced in small volumes as byproducts of other refining; the market is thin, the expertise concentrated, the substitution slow. By licensing their export, China showed it could reach into Western supply chains at points the West had never mapped — the obscure metals inside the optics, the power electronics, the military sensors. The message was legible, as the article says, because it was meant to be: chokepoints cut both ways, and the West's chokepoints are famous while China's were, until 2023, invisible.

The asymmetry that matters is temporal. An export control on EUV machines bites over years — it slows the adversary's future. An export control on gallium bites over quarters — it squeezes the adversary's present. The West's weapon is a slow strangulation of capability; China's is a fast strangulation of supply. Both sides are now racing to blunt the other's timeline: the West stockpiling critical minerals and funding non-Chinese refining, China pouring capital into domestic lithography. Each race is measured in years. Each side's vulnerability is measured in months.

For the Global South, the midstream weapon is the cruelest cut of all, because the South owns the upstream and captures almost none of the veto. The DRC's cobalt, Indonesia's nickel, Mozambique's graphite leave as ore and return as batteries — the value added, and the leverage, accruing to whoever owns the furnace. Every export control, from either side, lands in the South first as a price: higher input costs, delayed projects, another lecture about rules written elsewhere. The chip war's deepest lesson may be the oldest one in the commodity trade: diggers do not set prices. Refiners do. And the refiners are now at war.

The $210 billion lesson: what 2021 actually taught

The 2021 shortage is remembered as a chip story. It was really a humility story. Carmakers lost an estimated $210 billion in revenue — not for want of advanced processors, but for want of fifty-cent microcontrollers, the dumbest chips in the car. The industry that plans platforms a decade out was stopped by components it had treated as commodities, ordered just in time, from a supply chain it had never mapped past the first tier. The lesson was not that chips are important. Everyone knew that. The lesson was that nobody knew which chips were important until they were gone.

That ignorance was structural, not accidental. Automakers had spent thirty years stripping inventory out of the system — lean manufacturing as religion — while chipmakers had spent the same thirty years consolidating into fewer, larger, more efficient fabs. Both optimizations were rational. Together they built a system with no slack, and systems with no slack do not degrade gracefully; they snap. When demand snapped back from the pandemic faster than supply could restart — fabs cannot be switched on like lights — the queue for mature-node capacity stretched into years, and the car industry discovered it was standing in the same line as everyone else.

The policy response — the CHIPS Acts, the subsidies, the ribbon-cuttings — addressed the shortage's geography but not its logic. Every new fab adds capacity. None adds slack as a design principle, because slack is expensive and shareholders do not pay for insurance they cannot see. The next shortage will not look like 2021: it will not be about microcontrollers but about whatever the optimized-away buffer turns out to be next — substrates, chemicals, the obscure inputs nobody maps. The $210 billion bought the world new factories. Whether it bought wisdom is the question the next disruption will answer, at whatever price it chooses.