People talk about the “sovereign cloud” as if the issue started with the server. You choose a provider, check which jurisdiction it falls under, and verify its credentials. All of this is true, and all of it assumes one thing that’s never discussed: that we know how to build and operate the machines. But servers don’t just fall from the sky. They’re made of chips, and those chips come from an industrial supply chain in which almost none of the critical links are located in Europe. This isn’t a reason to give up. It’s a reason to understand, precisely, what we control and what we don’t.
From Idea to Chip: Who Does What, and Where
A chip isn’t manufactured in a single step. It goes through a long supply chain, where each stage is handled by different players on different continents.
It all starts with the architecture: the instruction set that defines how the processor interprets the commands it receives. Two families dominate the market. The x86 architecture, designed by the American company Intel and shared with AMD, powers most of the servers in use today. The ARM architecture, whose designs are created by the British company Arm—owned by the Japanese firm SoftBank—is gaining ground in data centers. Neither of these two families is European.
Next comes design: companies like Nvidia, AMD, and Apple design a chip based on an architecture, without owning a factory. They are called “fabless” companies—designers without their own foundries. They entrust their designs to a manufacturer.
That manufacturer is the foundry. This is where etching takes place—the process that imprints billions of transistors onto a silicon wafer. For the most advanced chips—those that power artificial intelligence and modern servers—nearly all of this etching is done by a single player: Taiwan’s TSMC. The foundry doesn’t invent the chip: it executes designs with a level of precision that no one else can match.
To etch, the foundry needs machines. The most advanced machines, known as EUV lithography systems, are produced by only one company in the world: the Dutch firm ASML. Without these machines, there can be no fine etching. Even with them, one must know how to use them, which requires years of fine-tuning.
At every stage, therefore, there is a bottleneck. Architecture, design, etching, and etching machines: four links in the chain, and Europe holds only one—that of the machines.
Where Everything Comes to a Head
Three key areas deserve special attention, because they determine everything else.
The first is TSMC. The company manufactures an overwhelming share of the world’s most advanced chips, and nearly all of the chips using the most advanced process nodes. This concentration is no commercial accident: it stems from decades of continuous investment and expertise that competitors are struggling to catch up with. The result is that a large part of the world’s computing infrastructure depends, physically, on factories located on an island whose geopolitical status is disputed.
The second is ASML. The EUV machines it manufactures are extremely complex, each assembled from tens of thousands of parts sourced from specialized suppliers. No one else knows how to produce them. This is where Europe holds a rare trump card—but just one—and it lies upstream in the supply chain, not at the end.
The third is architecture. The x86 architecture remains locked between Intel and AMD, two American companies. ARM, which is more open, is controlled by a company whose ownership and legal structure lie outside Europe. Designing a chip without relying on one of these families would require rebuilding an entire software ecosystem—something no one has done on an industrial scale.
Why Europe Doesn’t Have Its Own State-of-the-Art Foundry
The answer can be summed up in a few words: an advanced foundry costs a sum that few players can afford, and requires an ecosystem that cannot be imposed by decree.
A state-of-the-art semiconductor fabrication plant represents an investment of tens of billions of euros, which must be renewed with each new technology node—that is, every few years. This capital pays off only if the plant operates at full capacity, fueled by massive and continuous demand. Therefore, it requires, all at once, the money to build, the customers to fill the capacity, and the know-how to produce without wasting entire wafers.
This know-how cannot be bought. It is cultivated over decades, within a dense network of suppliers, chemists, process engineers, and teams capable of stabilizing yield. TSMC is not just a factory: it is an accumulated ecosystem that money alone cannot recreate. Europe has the machinery, it has designers, and it has foundries for less advanced chips—which are essential for the automotive and industrial sectors. What it lacks is the combination of capital, captive demand, and cutting-edge etching expertise, all gathered in one place. It’s the same equation that distinguishes a hosting provider from a hyperscaler: not a matter of talent, but a matter of structural resources coming together at the right time.
What the European Chips Act Aims to Achieve, and What It Can Accomplish
The issue is now being addressed at the national level. In the United States, the CHIPS Act, passed in 2022, committed tens of billions of dollars to bring advanced manufacturing back to U.S. soil, with results already visible in the form of factories under construction. The European Chips Act, adopted in 2023, has a specific numerical goal: to increase Europe’s share of global semiconductor production to 20% by 2030, by mobilizing public and private funding and attracting foundries to the continent.
The stated ambition and the resources committed are not on the same scale on both sides. The European initiative combines public funds, authorized state aid, and private investment, without matching the concentration of resources seen in the U.S. program. Above all, the 20% target is measured as a share of a rapidly growing market: to gain ground proportionally, Europe must grow faster than global demand, making the target difficult to meet.
What the plan can realistically achieve is to consolidate what Europe already does well and attract a few new facilities, including foreign foundries that will produce locally. What it will not create overnight is a fully European state-of-the-art foundry, nor the ecosystem that would make it viable. The message is the right one: recognizing that digital sovereignty has a hardware component, and that this component takes a decade to develop, not a single term in office.
What remains is a useful conclusion for decision-makers. The flag on a cloud contract says nothing about the chips that power the service, nor about the machines that etched those chips. One can operate a server in Europe, under European law, while relying on Taiwan for the silicon, the United States for the architectures, and a single Dutch supplier for the etching machines. Knowledge does not eliminate dependence. It allows us to identify it and to decide, link by link, which ones are worth fighting for.
Sources
- European Commission, European Chips Act, overview and 2030 objectives, commission.europa.eu
- CHIPS and Science Act (United States, 2022), text and implementation report
- ASML, annual report and documentation on EUV lithography
- TSMC, financial reports and breakdown by process node
- Arm Holdings, documentation on architectures and the licensing model
- SIA (Semiconductor Industry Association), data on the global semiconductor value chain