Europe's AI Value Chain Position: The ASML Paradox

Europe regulates AI better than anyone. It also builds almost none of the infrastructure that AI runs on. An EC-commissioned study on critical digital capacities published in May 2026 maps Europe’s position across eight segments of the AI value chain — and the picture is far more precarious than the regulation narrative suggests. Europe dominates exactly one segment. As of Monday 27 July 2026, even that one is under pressure.

Europe’s AI value chain position: the numbers the strategy documents bury

The Technopolis Group report, commissioned by DG CNECT, scores Europe’s market share across every layer of the generative AI stack. The results are stark.

Read that column top to bottom. In six of eight segments, Europe holds less than 15% of the global market. In four of them — raw materials, chip design, chip manufacturing, and cloud — it holds less than 5%. European cloud providers have less than 5% market share combined, against roughly 85% held by US hyperscalers.

Two numbers carry the whole argument. First: Europe holds an 80-90% market share in extreme ultraviolet (EUV) lithography — the equipment segment that makes advanced chip manufacturing possible anywhere in the world. Second: Europe has produced 25 notable foundation models, compared with 61 from the United States — evidence of real progress, but still less than half the US output in the layer that captures the most downstream value.

The single exception is semiconductor equipment, where Europe’s position is anchored almost entirely by one company: ASML. The Dutch firm’s extreme ultraviolet (EUV) lithography machines are essential for manufacturing advanced chips. No one else makes them at scale. This monopoly gives Europe an 80-90% share of the segment.

One company. One technology. One segment. That is Europe’s industrial position in the AI hardware stack.

What is the AI value chain, exactly?

Before the market-share numbers mean anything, it helps to know what each link in the chain actually does. The AI value chain runs from raw material to the AI service a user finally touches — eight stages, each one a precondition for the next.

Raw materials — The physical inputs: silicon, gallium, and rare earth elements needed to manufacture chips and the machinery that makes them. Nothing downstream exists without this layer.

AI semiconductor equipment — The machines that manufacture chips. This is ASML’s segment: extreme ultraviolet (EUV) and deep ultraviolet (DUV) lithography systems that etch circuit patterns onto silicon wafers. Without this equipment, no advanced chip gets made, anywhere.

AI semiconductor design — The blueprints. Companies like Nvidia, AMD, and Qualcomm design the chip architectures (GPUs, TPUs) optimised for AI workloads. Design is intellectual property, not physical output — and it is where the highest margins in the chip industry sit.

AI semiconductor manufacturing — Turning the design into a physical chip, using the equipment from stage two. This is dominated by foundries — TSMC in Taiwan and Samsung in South Korea — that fabricate chips for companies that design but do not manufacture.

Cloud infrastructure & supercomputers — The data centres, servers, and networking that provide the raw computing power to train and run AI models. This is where chips get deployed at scale — AWS, Microsoft Azure, and Google Cloud dominate this layer globally.

Foundation models — Large-scale AI models (like GPT, Gemini, or Mistral Large) trained on broad datasets and adaptable to many tasks. This is the layer most people think of as “AI” — but it depends entirely on the five stages before it.

AI applications — Software built on top of foundation models to solve specific problems: a medical diagnostics tool, a legal document reviewer, a customer service assistant. This is where AI meets a specific industry or use case.

AI services — The professional layer: consulting, integration, deployment, and support that helps organisations actually implement AI applications. This is the segment closest to the end customer, and the one requiring the least capital intensity.

Why the sequence matters: each stage depends on the one before it. A country that is strong in AI applications but has no domestic cloud infrastructure is renting its foundation from someone else. Europe’s problem, as the next section shows, is that its strength sits in a single early-stage segment — while the stages that capture the most value happen almost entirely elsewhere.

The ASML paradox: Europe builds the machines but buys everything else

The structural imbalance is worth sitting with. Europe makes the machines that make the chips — but it does not design the chips (Nvidia, AMD, and Qualcomm do, all US-based). It does not manufacture them at scale (TSMC in Taiwan and Samsung in South Korea do). It does not run the cloud infrastructure those chips power (AWS, Azure, and Google Cloud control roughly 85% of the European cloud market). And it does not train the large foundation models that run on that cloud — 61 notable models originate from the US, compared to 25 from Europe.

ASML captures revenue from equipment sales. But the value generated by AI — the models, the applications, the services, the cloud margins — accrues overwhelmingly to US and, increasingly, Chinese firms. Europe sells the pickaxe but does not mine the gold.

This would be a stable, if uncomfortable, position if the pickaxe monopoly were secure. It is not.

27 July 2026: the last stronghold under pressure

On Sunday, reports emerged that a Chinese state-backed consortium involving Huawei and SiCarrier has begun producing domestic immersion deep ultraviolet (DUV) lithography machines. The group plans to deliver five DUV systems in 2026 to Chinese chipmakers including SMIC. ASML shares dropped approximately 6% on Monday, wiping roughly $44 billion in market value.

Context matters here. This is DUV — older-generation lithography, not the cutting-edge EUV technology where ASML has no competitor. ASML’s EUV monopoly is not directly threatened yet. Some analysts called the sell-off disproportionate, noting that five machines do not constitute high-volume manufacturing capability.

But the signal matters more than the current output. China is systematically building the capacity to replace ASML’s older tools — tools that still account for a significant share of ASML’s China revenue. The trajectory points in one direction. DUV today, with EUV ambitions to follow. If Europe loses lithography dominance, it holds a meaningful industrial position in precisely zero segments of the AI hardware stack.

The market understood this instantly. The question is whether European policymakers do.

Regulation without industrial position is a tariff on your own teams

The EU AI Act is the world’s most comprehensive AI regulatory framework. That is a genuine achievement. But regulation operates differently depending on whether you are producing AI or consuming it.

If you are a US or Chinese AI provider, EU AI Act compliance is a market-access cost — absorbed at scale across a global customer base. If you are a European SME or research organisation, compliance is a cost you bear while depending on those same non-EU providers for every layer of the stack. You are paying the regulatory overhead without capturing the value the regulation is meant to govern.

The EC report itself flags this dynamic. It notes that “compliance costs faced by the private sector might slow down AI deployment” and that SMEs particularly struggle with high licensing costs, vendor lock-in, and compliance obligations. When the infrastructure is foreign, regulation does not protect your market position — it adds friction to your adoption of someone else’s technology.

This is not an argument against regulation. It is an argument that regulation alone is not a strategy. You need something to regulate from.

What would an actual industrial AI strategy require?

The EU is not unaware of the gap. The InvestAI initiative, announced in early 2025, aims to mobilise €200 billion for AI investment across the EU, with €50 billion in public funding and €20 billion dedicated to AI gigafactories. The AI Continent Action Plan focuses on five areas: computing infrastructure, high-quality data access, AI in strategic sectors, AI skills, and talent development.

The scale of ambition is real. But the current market-share data suggests the distance to cover is enormous. Europe’s realistic opportunities, based on the Technopolis analysis, sit in three segments where its share is growing:

• Foundation models (5-15%, increasing): Mistral AI is the flagship example, but Europe needs more than one globally competitive model developer. The report notes the EU has produced 25 notable models — a base to build from, not a position of strength.
• AI applications (5-15%, increasing): Vertical AI — domain-specific tools for healthcare, manufacturing, climate — is where European startups secured $5.8 billion in VC funding, accounting for 63.5% of all AI deal value. This is Europe’s clearest near-term opportunity.
• AI services (5-15%, increasing): Deployment, integration, consulting. The professional services layer where proximity to European customers and regulatory knowledge create genuine advantage.

None of these recover the hardware gap. But they represent areas where European organisations can build value rather than simply consuming technology built elsewhere.

The EC report is refreshingly honest about where Europe stands. The data is not ambiguous. What remains to be seen is whether the investment commitments match the structural reality — or whether Europe continues to lead in regulation while falling further behind in every layer of the stack that regulation is meant to govern.

If your organisation is navigating this landscape — building an AI strategy while managing EU regulatory obligations and non-EU provider dependency — get in touch.

If you found this useful

• EU CADA Sovereignty Tiers: What the Four Cloud Levels Mean — the new procurement framework for cloud sovereignty in the EU
• Enterprise AI in Europe: The Hidden Cost of Vendor Lock-In — how AI adoption is deepening software dependency concentration
• AI Adoption Risks Every European SME Should Know About — the specific barriers European SMEs face when adopting AI

Browse all Perspectives or get in touch →