Europe’s third way for AI is easier said than done

On June 3, The European Commission revealed Its European Technological Sovereignty Package – a Chip 2.0 Act, a Cloud and AI Development Act, an open source strategy and a roadmap for digitization in energy. The stated ambition is striking: Europe must take control of its data, supply chains and future.

The impulse is understandable. Europe remains highly dependent on non-EU suppliers for basic digital technologies and demand for computing is growing significantly. In a world where the United States and China are each pursuing complete AI dominance, drawing other economies into their orbits, Europe risks being drawn into the AI ​​order on terms set by others.

The Commission deserves credit for combining regulatory ambitions with industrial policy. But the package combines two distinct things: owning stack AND CHECK that. Until Europe understands this distinction, its third way will remain more statement than architecture.

The AI ​​race is primarily a calculation race. Whoever controls that infrastructure shapes the technology and set the conditions in which others access it. Europe is not a meaningful player: it produces a number of models with limited global reach, attracts only a fraction of global AI investment, and runs the most sensitive workloads through US hyperscalers.

What makes this doubly disappointing is that Europe owns upstream hardware assets – a global monopoly in extreme ultraviolet lithography through ASML and frontier semiconductor research through Belgium’s IMEC – yet exports these advantages to the US, South Korea and Taiwan.

Chips Act 2.0 partially addresses this, but building competitive fabrication capacity lasts a decade. The Cloud and AI Development Act introduces a useful sovereignty assessment framework, but assessment is not infrastructure.

The most instructive comparator is China. Unlike the US, which has been on the brink, China started out dependent on foreign equipment and responded with a two-pronged strategy: using diplomatic and market leverage to extract concessions from Washington, while building a domestic ecosystem around Huawei’s Ascend chips and state-led procurement.

Chinese firms unable to get Nvidia hardware now rely on Ascend for much of their computing. This is what closing a hardware dependency gap looks like – but it won’t be quick or cheap.

The most difficult comparison is India. India’s digital public infrastructure, such as the account aggregator framework, was a real innovation: the population-scale identity and payment railways that Europe has studied with admiration. But AI sovereignty has proven far more demanding. India imports almost all its advanced semiconductors, depends on foreign cloud termination hyperscalersand lacks the power infrastructure for large-scale GPU clusters.

India Stack showed that building digital rails for public purposes is scalable; checking full AI stack it is not, even for a country with strong software talent, 1.4 billion users and considerable political will. India has struggled in part because it pursued sovereignty across the full stack rather than focusing control on layers that directly affect strategic exposure. The result has been widespread effort and limited leverage everywhere.

Therefore, the right restatement for Europe is not “can we own stacks?” (the answer is almost certainly no, not completely, not quickly), but “can we control how it behaves within European society and shut down our most critical hardware dependencies?” These are more manageable questions.

Control means incorporating compliance, audit trails and risk thresholds into procurement contracts as market access conditions, not as afterthoughts. It means building a genuine evaluation capacity so that European public authorities can test AI systems for reliability, security and biases rather than relying on vendor assurances. It means using Europe’s single market as a lever – requiring data residency, algorithmic auditing, open interfaces and sovereign back-up options from any firm operating at scale in Europe.

Specifically for hardware, this means treating ASML and IMEC not simply as export champions, but as anchors for a European semiconductor industrial policy – ​​using public procurement, Gigafactories AI and the EIB to create the captive demand that China’s state-led procurement created for Huawei.

Without a reliable domestic customer base, the infrastructure investment in the Commission’s package will struggle to become self-sustaining.

The Commission on Technological Sovereignty’s package is a start. What needs to be done is a control architecture based on honest industrial strategy – one that learns from China’s hardware trajectory, avoids India’s proliferation trap, and uses Europe’s regulatory market power not as a substitute for infrastructure, but as a lever that makes infrastructure investment worthwhile.