ASML chief warns EU AI rules risk driving companies abroad

Christophe Fouquet does not lack access. The chief executive of ASML, Europe's most valuable technology company at a market capitalisation of €515 billion, counts the US Commerce Secretary, the founder of Tesla and SpaceX, and the President of the European Commission among his recent interlocutors. Yet the message he delivered from his office in Veldhoven, where construction cranes mark yet another expansion of the company's Dutch headquarters, was blunt: Brussels is regulating artificial intelligence before European industry has learned to walk, and the consequence will be companies moving their AI development elsewhere.

The warning from Europe's most valuable tech company

ASML occupies a unique position in the global technology stack. Its lithography machines, particularly the extreme ultraviolet systems that cost upwards of €150 million each, are the only tools capable of printing the most advanced semiconductor circuits. Those chips power the smartphones, vehicles, data centres and AI systems reshaping the world economy. With more than 44,000 employees and a order book that stretches years into the future, ASML is the industrial backbone of the digital age. When its chief executive speaks, governments listen. Fouquet's warning therefore carries weight beyond the usual corporate lobbying: the EU's regulatory approach, he argues, is actively undermining the continent's only genuine champion in the critical semiconductor supply chain.

The numbers he cites are stark. Ninety-nine per cent of ASML's machine sales go to customers outside Europe. For a company headquartered in the Netherlands, with deep roots in the European research ecosystem, that figure is an indictment of the continent's capacity to absorb its own most advanced technology. The machines are built in Veldhoven; the wafers they print are manufactured in Taiwan, South Korea, the United States and increasingly China. Europe buys almost none of them. That asymmetry, Fouquet contends, is not an accident of geography but a consequence of policy choices that have failed to create demand for cutting-edge compute on the continent.

An industry that sells almost entirely abroad

The 99 per cent figure is not new to those who follow the semiconductor industry, but hearing it from the chief executive of the company that defines the state of the art gives it renewed force. ASML's customers, TSMC, Samsung, Intel, SK Hynix, have concentrated advanced manufacturing in Asia and the United States. Europe's share of global semiconductor manufacturing capacity has fallen below 10 per cent, down from more than 20 per cent in the 1990s. The EU's Chips Act, adopted in 2023, aims to double that share to 20 per cent by 2030, backed by €43 billion in public and private investment. Fouquet's intervention suggests the target is fantasy without a parallel effort to create European customers for the most advanced nodes.

The point is structural. Advanced lithography machines are not purchased speculatively; they are bought when a foundry has confirmed demand for leading-edge chips. That demand comes from companies designing AI accelerators, high-performance processors, and mobile system-on-chips. Europe has few such design houses operating at global scale. ARM, the British chip architecture firm, was acquired by Japan's SoftBank and is now listed in New York. NXP and Infineon are significant players but focus on automotive and industrial microcontrollers, not the leading-edge logic that drives ASML's highest-margin systems. Without a European ecosystem of AI model developers, cloud providers and chip designers consuming advanced silicon, a European fab becomes a subsidy machine for foreign customers.

The AI Act: regulation before innovation

Fouquet's specific target is the EU AI Act, the comprehensive regulatory framework that entered into force in August 2024 after years of negotiation. The legislation classifies AI systems by risk level, imposes strict requirements on high-risk applications, and bans certain uses outright. It is the world's first horizontal AI law, and the Commission presents it as a global standard-setter. Fouquet sees it differently. "We didn't start running, we didn't start even walking, and we already had in front of us all the obstacles to not be able to make even the first step," he said. The metaphor is deliberate: European companies, in his view, are being asked to comply with a rulebook for a race they have not yet entered.

The criticism echoes a broader complaint from European industry that the EU regulates technologies it does not yet produce at scale. The General Data Protection Regulation, the Digital Services Act, the Digital Markets Act, each arrived before Europe had built globally dominant platforms in the relevant domains. The pattern, Fouquet argues, privileges process over outcome. Compliance costs fall disproportionately on smaller firms and startups, the very entities that might eventually challenge US and Chinese incumbents. By the time a European AI company reaches the scale where the AI Act's high-risk obligations bite, its American competitor has already amassed data, users and capital in a less constrained environment.

A united front from Europe's industrial giants

Fouquet is not speaking alone. In early May, ASML joined Airbus, Ericsson, Mistral, Nokia, SAP and Siemens in a joint letter to von der Leyen warning that the AI Act risks hobbling European companies before they can compete with US and Chinese rivals. The signatories represent the spine of European industrial technology: aerospace, telecommunications, enterprise software, and now, with Mistral, the continent's most prominent AI startup. Their intervention is notable for its breadth. These are not companies that oppose regulation on principle; they are companies that depend on predictable, proportionate rules to plan multi-year investment cycles. Their argument is that the AI Act, as drafted, creates legal uncertainty and compliance burdens that deter precisely the experimentation and deployment needed to build European AI capability.

The letter arrived at a moment when the Commission was already preparing a simplification package, adopted earlier this month, which delays certain provisions and introduces a potentially lighter regime for industrial AI applications. Fouquet welcomed the shift but argued the logic should extend beyond machinery and industrial uses. The distinction matters: industrial AI, predictive maintenance, quality control, process optimisation, is where Europe has genuine strength. Generative AI, large language models, consumer-facing applications, these are where the US leads. A regulatory framework that treats both categories with similar stringency, Fouquet contends, misunderstands the competitive dynamics.

Simplification packages and the wrong approach

Fouquet reserves particular scepticism for the Commission's preferred method of regulatory repair: the omnibus package. "The idea that you make something very complex, then simplify it, and that you end up with something good ... is a bit strange," he said. The critique goes to the heart of the EU's legislative culture. Major regulations are negotiated as comprehensive frameworks, often over years, with compromises that accumulate complexity. When implementation reveals unintended consequences, the response is a secondary legislative vehicle, an omnibus, that amends multiple acts simultaneously. The result, in Fouquet's view, is a patchwork that retains the original architecture's flaws while adding new layers of interpretation.

His alternative is straightforward: write the rules with the companies expected to compete under them. That is not a call for regulatory capture but for operational realism. The AI Act's definitions of high-risk AI, its conformity assessment procedures, its data governance requirements, these were drafted by lawyers and policymakers with limited input from engineers building production systems. The consequence, Fouquet argues, is a framework that assumes a maturity of process and documentation that only the largest incumbents possess. A European startup training a foundation model cannot easily meet the same evidentiary bar as Google or Microsoft. The regulation thus becomes a moat for non-European incumbents.

Tech sovereignty, procurement and the Intel cautionary tale

The Commission is preparing a tech sovereignty package for next month, aimed at reducing reliance on US technology, boosting the bloc's data centre capacity and growing its share of the chips supply chain. Fouquet's warnings address its likely pillars directly. On local production, he argues that "no country will ever have it all", a rejection of autarkic ambition in favour of strategic interdependence. On a reported provision that would require governments to give preference to European companies in public procurement, he is dismissive: "To have preference, you first have to have something you can prefer." The point is that procurement preference for non-existent European cloud or AI services achieves nothing but higher costs for public administrations.

The failed Intel project in Magdeburg illustrates the risk. In 2022, then-Commissioner Thierry Breton championed the construction of an advanced semiconductor fab on European soil as the centrepiece of the Chips Act. Intel committed to a €30 billion investment, backed by nearly €10 billion in German state aid. By 2024, Intel's financial troubles, driven by missed process nodes, lost foundry customers and a collapsing stock price, forced a postponement, then an effective cancellation. Fouquet's analysis is unsparing: "If you had a fab like this in Europe, all the wafers that would be manufactured would be exported to the United States. So then you'll be in a situation where Europe subsidizes a big project and the output of this project goes somewhere else." The logic is that without European design houses and end-users for leading-edge chips, a European foundry becomes a contract manufacturer for American IP.

The same logic applies to data centres. The Commission's sovereignty package is expected to include measures to expand European cloud infrastructure, potentially through public funding or coordinated procurement. Fouquet argues the state should not build or operate data centres: "You should leave that to industrial champions because they know how to do it." The champions he has in mind are not only ASML but the telecom operators, the colocation providers, the hyperscalers' European arms, entities with commercial incentives to match capacity to demand. State-backed capacity that sits idle, or serves only public sector workloads, is a waste of capital that could fund AI adoption programmes, skills training, or research compute grants.

What the Commission does next

The Commission's response, delivered through spokesperson Thomas Regnier, emphasises that the AI Act "supports innovation by increasing trust, and trust leads to increased uptake and investments." He points to the recent simplification package as evidence of responsiveness: "clear timelines [and] simpler and innovation-friendly rules for AI." The gap between the two positions is not merely rhetorical. It reflects a fundamental disagreement on the sequence of industrial policy. The Commission believes regulation creates the conditions for trustworthy markets; Fouquet believes markets must exist before regulation can shape them. Both cannot be right in the short term. The tech sovereignty package next month will reveal which view prevails in the corridors of the Berlaymont.