Extreme ultraviolet lithography (EUV) is the fabrication technique that enables chips below 7nm — the nodes used in datacentre GPUs, smartphone processors, and AI chips. ASML, a Dutch company based in Eindhoven, is the world's sole manufacturer of EUV machines. One ASML EUV machine costs approximately €150 to €200 million, weighs 180 tonnes, requires 40 containers to transport, and is assembled from 100,000 components from 5,000 suppliers in 30 countries.

There is no alternative. Not in China (despite massive investment since 2019), not in Russia, not in Europe other than ASML itself. Any fab that wants to produce advanced chips must go through ASML. And ASML, under pressure from the US and Dutch governments, has not been selling its EUV machines to China since 2023.

TSMC (Taiwan Semiconductor Manufacturing Company) produces approximately 90% of the world's advanced chips (nodes below 5nm). Its customers include Apple, NVIDIA, AMD, Qualcomm, Google, Amazon, Microsoft. The NVIDIA H100 GPUs powering AI models — including those of Anthropic, OpenAI and Google DeepMind — are fabricated by TSMC in Taiwan.

TSMC in Taiwan. On an island of 36,000 km² separated from mainland China by a 130 km strait. In a zone subject to territorial claims by the People's Republic of China. In a region where large-scale Chinese military exercises took place in 2022 and 2023.

Since October 2022, the US government has banned the export to China of advanced chips and associated production equipment. This unilateral decision was applied extraterritorially — forcing ASML (a Dutch company) to cease deliveries to China. It was applied without consulting the European Union.

This precedent says something important for European organisations using US cloud: the United States has demonstrated its capacity and willingness to exercise extraterritorial control over the silicon supply chain. If geopolitical circumstances required it, similar control could be exercised over European access to US cloud services — or over European hardware equipment dependent on US technology.

A Refurbished Grade A Dell PowerEdge R750 bought today contains chips fabricated 2 to 3 years ago. These chips already exist in the physical world. Their availability no longer depends on future TSMC production, US export restrictions, ASML machine availability, or the geopolitical situation in the Taiwan Strait. They are there. They work. They will continue to work for 5 to 7 additional years with standard maintenance.

In a silicon supply crisis — geopolitical (Taiwan conflict), natural (earthquake, typhoon in production zones), or regulatory (new export restrictions) — the patrimonial and strategic value of existing physical servers increases. Demand for qualified certified second-hand hardware rises. The organisation that owns its physical hardware has business continuity that the 100% cloud organisation does not.

Hyperscalers have server stocks — but limited. In an extended TSMC production interruption, they cannot order new H100 GPUs or new datacentre CPUs. Available capacity will be maintained — but not increased. Customers who want to scale up during a global crisis will hit rationed capacity.

Who gets served first? The hyperscalers themselves (their own AI services, critical internal services), then customers in contractual priority order — Enterprise Agreements first, Pay-as-you-go last. The organisation without an Enterprise contract is last in the queue.

The organisation with its own physical server is outside the queue. It does not wait.

The standard Total Cost of Ownership calculation cloud vs on-premise compares: hardware acquisition cost + staff + maintenance + datacentre vs monthly cloud invoice. This is a correct calculation in a stable world. It is insufficient in a world where the silicon supply chain is concentrated on a geopolitically disputed island.

What the standard TCO does not include:
— The geopolitical insurance premium of existing physical hardware
— The cost of cloud rationing in a supply crisis
— The value of business continuity independent of the global supply chain
— The cascade dependency: cloud → H100 GPU → TSMC → Taiwan Strait

Layer 1 — Contractual: Terms Section 14.12 (California jurisdiction), noncancellable commits, egress fees, unilateral price modification. Documented in Terms Under the Microscope.

Layer 2 — Software: proprietary dialects (DynamoDB, BigQuery, Lambda), proprietary certifications, newspeak that destroys neutral skills. Documented in The Newspeak That Costs Dear.

Layer 3 — Material: dependency on TSMC for new GPU servers, dependency on ASML for EUV machines, geopolitical concentration in the Taiwan Strait. This layer is the deepest. It is also the only one the Terms do not mention — because it precedes the Terms. The cloud vendor does not guarantee its own silicon supply chain continuity. Nobody guarantees it.

European Chips Act (2023): €43 billion to double Europe's share of global semiconductor production — from 10% to 20% by 2030. Ambitious objective. Fine-node fabrication capacity: not yet achieved.

Intel Fab in Germany (Magdeburg): planned €33 billion investment, 1.8nm fabrication announced. Delays: pushed back in 2024 due to reduced subsidies and market difficulties. Planned capacity: 2030 at earliest.

RISC-V: open-source instruction set architecture supported by an international foundation. RISC-V chips exist and are produced — primarily by TSMC for customers who order them. The architecture is free. The fabrication remains dependent on the same foundries.

There is no fab in Europe today capable of fabricating chips below 7nm. TSMC and Samsung are the only ones worldwide able to do so at industrial volume. China is investing massively but remains blocked below 28nm due to ASML machine restrictions.

The gap is 10 to 15 years in optimistic scenarios — and assumes announced investments materialise, specialist engineers are trained, ASML machines are available (ASML is itself subject to US pressure on its exports), and markets are there.

The corpus conclusion: RISC-V is the correct horizon. The sovereign foundry is the necessary horizon. Between both and today, there is a 10 to 15-year vulnerability window during which dependency on the Taiwan Strait is structural and unavoidable. It is in this window that the Opération Dindon corpus operates. It is in this window that the CIO's decisions are made.