Bargaining Chips: Could the EU Leverage ASML to Influence U.S. AI Policy?

ASML's monopoly on advanced chip-making machines gives the EU rare leverage over global AI development. Using it would mean accepting major costs.

Summary

• What’s happening: Dutch firm ASML has a monopoly on machines essential for advanced AI chips, giving the EU potential leverage over global AI development.

• The stakes: Frontier AI companies in the U.S. could one day pose catastrophic risks affecting the EU, from misuse to misalignment.

• A very big lever? The EU could impose export licensing on ASML technology, forcing U.S. companies to adopt better safety measures.

• Bottom line: This strategy would pose major costs for the EU, but may be justified if policymakers judge catastrophic AI risks to be sufficiently serious.

‍

The Dutch firm ASML is the only company in the world that commercially produces Extreme Ultraviolet (EUV) lithography machines, essential for manufacturing AI chips at nodes below 7 nanometers. These machines are used by companies like Taiwan’s TSMC to manufacture AI chips designed by companies like Nvidia.

Chips provide computing power (compute) – one of the three pillars of AI development alongside algorithms and data. Lower-node chips offer huge performance advantages, and as AI models scale exponentially, only advanced lower-node chips can keep up.

Despite this crucial contribution, the EU is asserting little control over the behavior of the frontier AI companies that use ASML technology.

While those companies are not operating from the EU, there are growing concerns that frontier models pose cross-border catastrophic risks. These could range from misuse risks, arising when bad actors weaponize AI systems for harmful purposes, such as cyberattacks or weapons, to existential risks caused by AI misalignment, which occur when AI systems pursue unintended goals. The urgency of AI risk mitigation is heightened by the U.S.-China race for AI leadership, which is currently prioritizing capability gains over safety.

As evidence grows that frontier AI may pose severe risks, the EU could feel compelled to use all tools at its disposal to ensure foreign companies developing the technology are adopting appropriate safety measures.

EUV export controls towards China are already in place due to U.S. pressure to slow Chinese AI development. Could the EU use similar leverage to shape the behavior of U.S. companies?

Despite ASML’s strategic importance, the implications for EU AI policy remain severely understudied – a gap our research seeks to address.

What licensing requirements would improve safety practices?

ASML could limit the speed of frontier AI development by restricting lithography machines necessary to develop new chips. The EU could force ASML to do this by imposing export licenses for chips made with ASML’s EUV tools. These licences could require chip companies, such as Nvidia, to only sell top-end chips to companies and cloud providers that meet specific safety standards.

EU-level export controls of EUV tools, though politically difficult, could offer advantages by distributing the costs of potential U.S. retaliation among member states, reducing vulnerability to external pressure and helping establish the EU as a geopolitical actor in AI.

Here’s how EU export licences on EUV tools could work in practice.

1. Policy instrument

Just as the U.S. restricted AI chip access through its Foreign Direct Product Rule (FDPR) or (now repealed) AI Diffusion Framework, the EU could develop mechanisms to prevent EU-enabled technology from causing spillover harm. In practice, this could mean that when Nvidia designs a new chip and contracts TSMC in Taiwan to manufacture it using ASML equipment, that chip would require EU export licensing before Nvidia could sell it to AI companies.

2. EU country roles

Qualified majority voting in the EU Council could override national decisions, including Dutch preferences. This would also distribute political and economic costs across all 27 member states rather than leaving the Netherlands vulnerable to targeted retaliation.

3. Legislative process

The EU Commission could propose the new regulation which would then require Council approval via qualified majority voting. The regulation could include fast-track procedures to ensure a speedy Council decision – akin to the EU’s anti-coercion instrument, which requires the Council to make a determination within 10 weeks of receiving a formal proposal.

4. Targeting Taiwan

Taiwan’s TSMC produces a clear majority of the world’s most advanced chips. Because the U.S. mostly outsources the fabrication of its AI chips to Taiwan, export controls towards the U.S. would for now do little to reduce America’s chip access – at least until the U.S. scales up its own domestic production, which is expected to be a gradual endeavour. To maximise the leverage of export controls in the short term, the EU’s best strategy would be a credible threat of restricting exports to Taiwan, with the expectation that this can change the behaviour of U.S. companies.

5. Restrictions and conditionality

The EU could declare that companies wanting to purchase advanced AI chips must demonstrate compliance with AI safety standards. This could include submitting to regular audits by third party organizations; implementing specific transparency measures for frontier AI training; or adhering to compute usage caps for models above certain parameter thresholds.

To balance strategic leverage and cost, the EU’s controls could initially target only the newest EUV machines. This would limit the ability of chip companies to scale production, given that leading chips typically stay relevant for frontier AI training for less than five years. But it would be a more measured policy, as it wouldn’t affect the production capacity that TSMC and others already have.

A potentially more powerful way to disrupt chip production capacity would be for ASML to refuse to provide maintenance services, a potent threat given the 10-30 year lifespan of ASML machines.

6. Enforcement mechanisms

The EU could establish a new central enforcement body to coordinate with national export control agencies to track the flow of EU-enabled chips through the global AI supply chain.

Companies who fail to comply with EU audits could face graduated responses, ranging from warnings and fines for initial breaches, to being cut off from future chip access for serious violations.

The system would allow the EU to escalate or relax restrictions as required. To incentivise the U.S. to keep retaliations in proportion, the EU could threaten to expand controls to include lower-end EUV equipment. If cooperation improved, restrictions could be eased.

This framework could ultimately serve as the enforcement backbone for a potential international AI safety agreement, giving the EU concrete tools to ensure that its technological contributions to global AI development align with European values and safety standards.

Reality check

Implementing export licences on ASML technology would face significant challenges.

First, the EU’s dependence on U.S. frontier AI models and military support – particularly regarding Ukraine – means Washington would have powerful leverage in any confrontation over semiconductor policy. Any EU attempt to restrict ASML exports against U.S. interests would risk severe retaliation at a time when Europe can least afford it.

Second, the EU is not a unified actor. EU decision-making involves lengthy negotiations where competing national interests often result in watered-down compromises. Despite EU dual-use regulation, export controls remain primarily a national responsibility, with limited EU oversight. This creates structural barriers to quick, effective and coordinated action.

Third, any restrictions would significantly impact ASML, which generated €28.3 billion in revenue in 2024, and would ripple through the entire European economy.

Fourth, ASML itself is dependent on U.S. talent and technology and thus vulnerable to retaliation. The U.S. accounts for over 8,000 ASML employees and two out of the five board members. About 10% of ASML’s technology is American, meaning 10% of the components, software, or intellectual property by value in their machines originates from the US.

Under the U.S. FDPR, even this relatively small percentage, as defined by the ‘de minimis rule’, gives the US legal authority to control where ASML can export their equipment. EU controls that conflict with US preferences could create a legal standoff where the U.S. retaliates using their FDPR to restrict ASML’s exports to even more third party countries. However, the more likely retaliation would be for the U.S. to simply block the export of U.S.-built parts for ASML’s machines using standard export controls.

To overcome these challenges, the EU would need to drastically reduce its dependency on U.S. technology and strengthen its economic sovereignty. While the EU is already seeking to reduce dependency through its Economic Security Strategy, the EU Chips Act, the European Defense Fund, and diversified trade partnerships, current efforts are challenged by EU personnel constraints and fragmented funding and implementation.

Expanding the EU’s strategic toolkit

To strengthen the EU’s position if it one day decides to use ASML as a strategic tool, EU policymakers could focus on:

• more resources and staff to deliver its Economic Security Strategy;

• improving coordination across funding instruments, such as the European Innovation Council, the European Investment Fund, the Chips Joint Undertaking, and the Smart Networks and Services Joint Undertaking;

• establishing a dedicated fund to compensate affected stakeholders like ASML and affected member states.

Crucial to all these efforts is building the political capital for stronger coordinated action at the EU rather than national level.

ASML’s monopoly on lithography machines provides the EU with unique leverage over AI development. Using it would require the EU to accept major costs, and may only be viable in high-stakes scenarios such as credible threats of AI catastrophe.

The EU faces a choice: build this strategic capacity proactively, or risk having no lever to pull when it matters.