A Moment of Truth for Europe's AI Strategy

Almost every national AI strategy of the last three years contains the same contradiction. Governments declare sovereignty and control over their digital future while running that future on chips they did not design, clouds they do not own, and models trained on infrastructure they did not build. Sovereignty, it turns out, is less a destination than a direction. And for most of the world, including Europe, the gap between the rhetoric and the reality is wide.

A recent report from Digital New Deal, analysing 25 national AI strategies, puts this plainly. Of all the countries studied, only the United States and China come close to what the report calls full-stack sovereignty: genuine control across chips, computing power, cloud infrastructure, AI models, and data governance. Everyone else is managing dependencies. The question is no longer whether you are dependent. It is whether you chose your dependencies deliberately, or simply inherited them.

Sovereignty was a necessary fiction

Since the COVID pandemic exposed how fragile global supply chains really are, the world has understood that everything runs on chips. AI runs on the most complex and risk-prone version of that supply chain. There are chokepoints at every level: the Dutch machines that are the only way to manufacture the most advanced semiconductors, the Taiwanese factories that produce them, the American tech giants that turn them into usable computing power.

The sovereignty narrative was necessary. It forced governments and businesses to ask questions they had been too comfortable to ignore. But it was always more aspiration than plan. What the Digital New Deal report does well is reframe the conversation. AI sovereignty is no longer primarily about national pride or breaking free from American technology. It is about having real options: the ability to keep operating, adapt, and compete in a world where the rules of the technology stack can change overnight. Export controls, sanctions, a vendor shutting down a service without notice. These are all real risks right now.

Europe has pieces, not a strategy

As a European citizen you can't help but worry these days where AI development is a duopoly game led by US and China. But Europe is not without assets. France has Mistral, one of the world's leading open-weight AI model ecosystems. The Netherlands is home to ASML, without which no one on earth can manufacture the most advanced chips. Norway is building some of Europe's most ambitious AI data centre infrastructure, powered by clean hydroelectric energy. Finland hosts LUMI, one of the most powerful supercomputers in the world. Switzerland has built a genuinely credible sovereign cloud.

But assets are not a strategy. What Europe currently lacks is the coordination that turns these individual strengths into a coherent alternative. The report is direct on this: if Europe pooled its capabilities rather than splitting them across 27 separate national efforts, it could close to within striking distance of the United States and China. The gap is first and foremost organisational and political, not technical.

How close is Europe?

The report scores every country across eight dimensions of AI strength: computing power and energy, semiconductor independence, cloud autonomy, AI models and data, talent, governance, network resilience, and investment. The scale runs to 40. The United States and China both score 36. The best individual European nations land around 26 to 28.

But when European capabilities are assessed collectively rather than nationally, the picture changes. A unified Europe, the report concludes, would score close to 34, within two points of the world's leading AI powers. Go further, and pool serious investment into the hard layers of the stack, the way Japan and South Korea have done with chips and computing infrastructure, and Europe reaches 36. Level with the United States and China.

This is what the report calls the violet scenario, and it is within reach. Japan's Rapidus programme is targeting advanced chip production by 2027. South Korea has funded five national AI model programmes alongside domestic chip development. Europe has comparable industrial depth and research capability. In several areas, including regulatory credibility, data governance, and clean energy infrastructure, it has genuine advantages that neither the US nor China can easily replicate.

Four National Archetypes

The report identifies four broad approaches that countries take to AI independence:

The coordination problem is the problem

What is missing in Europe is not capability. It is the economic architecture that makes collective action rational. European AI development cannot be solved from Brussels alone, and national efforts in isolation keep producing impressive local results that never add up to strategic weight.

The missing ingredient is capital markets and economic incentives that make investing in European AI infrastructure as attractive as investing in Silicon Valley. Initiatives like EU.inc, which aims to create a unified framework for European companies to incorporate and scale across borders, point in the right direction. A fragmented capital market is as significant a barrier as a fragmented chip supply chain.

Pragmatic independence as the new goal

Nobody achieves full sovereignty in this landscape. The United States depends on TSMC for its most advanced chip production and ASML for the machines to build them. China depends on imported lithography equipment it cannot yet replicate. Norway hosts one of Europe's most ambitious AI infrastructure projects and outsources its model layer entirely to OpenAI. Full independence is not available at any price.

The right ambition is pragmatic independence: knowing which dependencies you carry, making deliberate choices about which ones are acceptable, and having a genuine ability to adapt when the ones you accepted stop working in your favour.

Pragmatic independence is not freedom from dependency but favours agency over dependency. It is the difference between a supply chain you have mapped, stress-tested and prepared alternatives for, and one you discover under pressure for the first time.

For European businesses and the policymakers who shape the environment they operate in, this reframe matters. We need to stop measuring progress against the impossible standard of full sovereignty and in stead start measuring it against the standard that actually determines outcomes: how quickly can you adapt, how many paths do you have, and how much of your critical infrastructure depends on decisions made in places you do not control.

Europe has more to work with than the current fragmentation suggests. The violet scenario is an aspiration. But it is an organisational challenge that European ambition, capital, and industrial capability are genuinely equipped to meet, if the will to act collectively proves stronger than the pull of individual national interests.