Supply chain realities¶

The global semiconductor supply chain has achieved a level of centralisation and fragility that would make a competent risk manager reach for the emergency whisky while updating their CV. Most of the world’s most advanced chips are made by one company, on one island, using equipment from a handful of suppliers, in a geopolitical situation that could charitably be described as tense and more accurately as one diplomatic incident away from catastrophic.

TSMC occupies a position in the global technology supply chain comparable to The Patrician’s position in Ankh-Morpork: not officially in charge of everything, yet practically nothing happens without its involvement, and its sudden removal would be spectacular, expensive, and probably involve at least one small war. The arrangement worked adequately while demand was predictable and tensions manageable. Neither condition currently holds.

One company, one island¶

TSMC manufactures around 90 per cent of the world’s most advanced semiconductors, a dominance built not on anti-competitive behaviour but on decades of technical excellence, enormous capital investment, and competitors concluding that the leading edge was too difficult or too expensive to chase. A leading-edge fab costs somewhere between €15 and €20 billion before it produces a single chip and takes years to build, so capacity cannot follow demand surges; it is allocated, through long-term contracts, premium pricing, and TSMC’s own assessment of priorities. When demand spikes, as it has for AI accelerators, lead times stretch, prices rise, and the foundry effectively decides which products exist.

The single farm in contested territory¶

Taiwan’s situation adds catastrophic tail risk to an already precarious structure. Fabs are enormous physical infrastructure, deeply woven into the island’s power, water and logistics, and cannot be relocated; conflict, or even a credible threat of it, would halt most advanced semiconductor production worldwide, with consequences usually estimated in the trillions. The arrangement resembles Ankh-Morpork depending for its entire food supply on a single farm sited in contested territory. Everyone understands this is dangerous. Everyone agrees it ought to change. Nobody can change it quickly, because relocating the farm takes a decade of infrastructure work, so everyone hopes nothing disrupts the current arrangement while remaining acutely aware that disruption is plausible.

Chokepoints all the way down¶

The chain is a series of narrow gates rather than one. ASML is the only maker of the extreme ultraviolet lithography machines the leading edge requires, so TSMC’s growth is constrained by ASML’s output. Hundreds of specialised materials come from single suppliers or nearly so. Packaging and testing have their own capacity limits and their own geographic concentrations. Each specialisation made economic sense; together they form a system in which disrupting any component can stall the whole, as the pandemic demonstrated and as more recent episodes keep re-demonstrating. Efficiency and resilience were traded, deliberately, and efficiency won, because disruptions were rare and margins are not. This is, after all, the first foundation of the digital stack, and it is run for profit, not for resilience.

Diversification at the speed of concrete¶

Subsidy programmes in the United States, Europe, and China are building fabs outside Taiwan, and they are necessary, expensive, and slow. A fab without the surrounding ecosystem of suppliers, equipment, and accumulated expertise is a building; the expertise takes years to grow and cannot be bought retail. Distributing production for resilience also costs more than concentrating it for efficiency, which is precisely the trade the market declined to make on its own. Meaningful geographic diversification is measured in decades, and planning that assumes otherwise is hope wearing a spreadsheet. Until then, the sensible posture is the unglamorous one: inventory buffers, design flexibility, contingency plans, and attention to a strait most procurement teams could not previously have found on a map.

The clerk’s brief¶

June 2026: Brussels tries again¶

On 3 June 2026 the Commission adopted the Chips Act 2.0 proposal as part of a wider technology sovereignty package, aiming at investment conditions, demand stimulation, supply-side reinforcement, and reduced dependencies, after the first act mobilised more than €52 billion. The proposal now enters the legislative process. The clerks note that the second act arrives with more modest verbs than the first, which they consider progress of a kind.

February 2026: The war is priced again¶

Bloomberg Economics re-modelled a US-China war over Taiwan in February 2026, holding to the order of magnitude of its earlier estimate: a shock around 10 trillion dollars, with the largest single component being the missing Taiwanese semiconductors. A blockade scenario does roughly half the damage of a war, which still puts it well ahead of the pandemic. The clerks observe that the models keep being rerun and keep producing the same first digit, and that nobody reruns a model they believe.

November 2025: A truce, of sorts¶

According to the Dutch government’s update on the Nexperia affair, in November 2025 China moved to enable the resumption of supplies from Nexperia’s facilities in China, and the Netherlands subsequently suspended its order under the Goods Availability Act as a constructive step taken with European and international partners. Chips began flowing again. The clerks note that both sides demonstrated their chokepoints work as designed, and that this is the kind of lesson each party files for later.

October 2025: The rehearsal, with car parts¶

On 30 September 2025 the Dutch minister invoked the Goods Availability Act against Nexperia, the first use of the act since 1952, citing governance shortcomings and a risk to European economic security; China answered with an export block on all Nexperia locations in China. By mid-October ACEA warned that carmakers’ stocks of these unglamorous, essential chips would last a few weeks while qualifying alternative suppliers would take months. A mature-node commodity part nearly stopped European vehicle production. The clerks underline the ratio: weeks of inventory against months of substitution, and note it generalises.

June 2025: The island draws its own line¶

In June 2025 Taiwan added Huawei and SMIC, among hundreds of entities, to its strategic high-tech commodities entity list, requiring licences for Taiwanese firms to supply them, aligning Taipei with the US control regime. The silicon shield, the argument that Taiwan’s indispensability protects it, is now itself an instrument of policy. The clerks decline to judge whether a shield becomes stronger or more brittle when it is waved.

April 2025: The auditors mark the homework¶

The European Court of Auditors’ special report on the EU’s microchips strategy, published April 2025, found the Chips Act very unlikely to deliver the 20 per cent global production share targeted for 2030, with the Commission’s own forecast pointing to 11.7 per cent, and noted the Commission controls only around a tenth of the public funding involved. An urgent reality check was recommended. The clerks observe that the reality check arrived on schedule; reality, as usual, had arrived first.

The chokepoint map¶

The map has not changed shape since the file opened; it has only acquired legends. Taiwan holds around nine tenths of the leading edge, and the cost of losing it has been estimated at roughly 10 trillion dollars since Bloomberg Economics first modelled it in January 2024. The equipment gate is Dutch: ASML noted in December 2024 that US export rules already covered its systems and that Dutch restrictions could follow. The materials gate is Chinese: two waves of rare earth export controls in April and October 2025, the second suspended until November 2026, which is a pause with a timer on it. And the demonstration was European: Nexperia, October 2025, weeks of stock against months of requalification. Every gate on the map has now been exercised at least once by its keeper. The clerks’ standing assessment is unchanged: the first foundation is run for profit, resilience is bought only after each rehearsal, and the next rehearsal is unlikely to use car parts.