Photo credits: Marcel Oliveira
By Niamh Teather
ASML gives Europe leverage no other region can easily replace. The question is whether Europe is willing to use it
The most valuable machines in the AI economy are made in Veldhoven, a town in the southern Netherlands. They are lithography machines: tools that use light to print the microscopic patterns from which advanced computer chips are made. About the size of a bus, the newest versions can cost as much as $400 million. Inside them, droplets of molten tin are flashed by a laser 50,000 times a second into a light fine enough to etch circuitry to tolerances measured in atoms. The Dutch firm ASML is the only company that makes them. Without them, the most advanced chips Washington is so anxious to control cannot be made. In the race for the AI frontier, ASML is Europe’s rare point of leverage.
In a hypothetical scenario published this June by a group of researchers including MIT’s Michiel Bakker, the White House eventually moves to bring ASML under American control. It does so after Europe has spent years mistaking a thin version of sovereign AI for strategy, while leaving its one real source of leverage unused.
Sovereign AI, as governments use the term, is mostly incoherent. No country owns the whole stack. Europe’s mistake is not that it depends on others. Every country does. The goal is not self-sufficiency but interdependence: relationships where dependence runs both ways and can be managed on better terms.
InvestAI embodies that mistake: €200 billion on paper, perhaps €50 billion in public money, with the rest dependent on private investment. It is designed to build European alternatives – European compute, European gigafactories, a champion to wave at the Americans. There is a real case for sovereignty: diversify dependence so that no single supplier can switch Europe off. But building a smaller domestic version of the American stack is insurance, not power. It may reduce Europe’s reliance on others; it gives others no reason to rely on Europe. ASML does.
Yet Europe is letting Washington spend its chokepoint advantage. Under US pressure, Dutch and American export controls have tightened, reducing ASML’s China business. China accounted for 36% of ASML’s sales in 2024 but the company expects this share to fall to roughly 20% in 2026. Those controls may well be justified. But Europe is enforcing them through the one asset that gives it bargaining power, without asking for anything in return.
A chokepoint you refuse to bargain with does nothing for you. Europe is doing Washington a favour and calling it prudence.
Chatham House, the international-affairs institute, makes the strongest objection to this kind of argument. In an April article, it argued that AI export controls are a weak bargaining chip: advanced chips leak, and DeepSeek showed how better algorithms can make less advanced hardware go further. That weakens the case for treating GPUs (the chips used to train AI models) as a permanent chokepoint. It does not settle the case against ASML.
Extreme ultraviolet (EUV) lithography is different, because its supply is uniquely concentrated. ASML makes more than 90% of the world’s advanced lithography machines and every EUV system in existence; there is no rival product to divert, and no second source to smuggle from. A GPU can slip through a third country or be partly offset by better software. An EUV system can do neither: it is far harder to replicate, and its installation, servicing and upgrades stay tied to ASML’s own engineers. Better algorithms may reduce the compute needed to train a model. They do not remove the need for the most advanced chips, or the lithography that makes them. Chatham House is describing a chokepoint whose durability is in doubt. ASML’s is more durable, for now.
That does not make it a clean Dutch trump card. ASML sits at the centre of a European industrial ecosystem: the German firm ZEISS supplies the ultra-precise optics, while TRUMPF, also German, provides the high-power lasers that generate EUV light. Its systems also rely partly on American technology, Europe would need allied cooperation to sustain any controls, and China has every incentive to build around the monopoly. The leverage is shared and perishable. It is still leverage.
What would using that leverage look like? The next time Washington asks The Hague to tighten the screws on ASML’s China sales, the answer should be neither yes nor no. Instead, it should be what do we get? The price should be a US-backed guarantee of access to frontier compute – reserved capacity, not pay-as-you-go access, at top-tier allocation and for a fixed period. This would not be sovereignty but it would secure access while Europe works to reduce its reliance elsewhere.
Dutch cooperation on the next round of controls should be conditional on that guarantee, with both commitments set to lapse together. The difficulty is institutional. Dutch export law, US restrictions and commercial cloud allocation sit in separate systems. That is why Europe receives nothing for the controls it already enforces. The agreement would have to create the linkage.
Do that, and Europe avoids becoming the second-tier customer Bakker’s scenario imagines for 2029, with its cloud compute halved by a decision made in a room it does not sit in. Europe already has a legal lever at the chokepoint. It should start charging for the cooperation it gives. The bargain is simple; the consequences are not.
This would be ugly. It would simultaneously use a civilian company as geopolitical leverage, risk damage to ASML’s commercial interests, invite Chinese retaliation, and require Europe to say no to an ally accustomed to getting its way. Denmark’s dispute with Washington over Greenland is a reminder that an alliance does not erase asymmetry. This is realpolitik of an unlovely kind, and the dignified European instinct is to shoe-shine it into the vocabulary of strategic autonomy.
But the alternative is not dignity. It is Bakker’s 2031 ending, in which Washington stops asking for the chokepoint and simply seizes it, and Europe is left to choose between becoming an American protectorate and handing the future to Beijing. Better to be the party that played its one card badly than the party that was too well-mannered to play it at all.
None of this requires bad faith. Cutting exports keeps the Americans happy. Building a domestic model looks like progress. Waiting for consensus is how the EU held 27 countries together in the first place. Every move is reasonable in isolation; the sum is a continent that mistakes owning things for holding power, and gives away the one asset others fear.
ASML’s capability is still rooted in Veldhoven, and it will be there next year and the year after. ASML is not a substitute for reducing Europe’s wider dependencies but it is the leverage that Europe has today. The question is whether Europe understands what it holds before someone else decides for it, and whether it can bring itself to do the unlovely thing that holding it requires. So far, it has never treated the card as anything more dangerous than an export.
