With AI, as with many things, there are rule-makers and there are rule-takers. The United States exercises power through its technology companies, capital markets, platforms, and control over key AI infrastructure. China combines industrial scale with state coordination, manufacturing capacity, and long-term investment. The European Union, often regarded as the third of these ‘Big Three’, imagines itself as perhaps less technologically dominant, but able to shape global vicissitudes through regulation, standards, and the size of its common market.

Yet, this framing has increasingly been called into question—recently in the policy report/speculative scenario Europe 2031. What’s offered is a stark warning that the EU is on a trajectory to lose its position of influence. Its route from rule-maker to rule-taker would not primarily be regulatory. Rather, it would be predominantly material, and fundamentally strategic. An EU dependent on foreign models, cloud infrastructure, compute, platforms, and security systems may retain the formal authority to write rules, while losing its practical ability to determine the conditions under which they operate.

The report is absolutely right in taking this possibility seriously. It is also right that the EU’s central problem is one of political commitment and how this limits institutional capacity. Yet, it limits its own diagnosis by accepting a fundamentally Silicon Valley-driven definition of the AI race and by imagining the EU largely through France, Germany, and the European Commission. The result is a scenario in which the EU appears to have only two choices: compete on American terms or accept irrelevance. A more credible strategy would begin from the EU as it actually exists—with 27 member states possessing different capabilities, vulnerabilities, and sources of leverage—and ask how those differences might be coordinated rather than suppressed.

What Europe 2031 gets right

Europe 2031 is less a conventional policy report than a novella-manifesto. Beginning with the release of DeepSeek R1 in 2025, it follows Caroline Dubois, a European Commission official who gradually realises that the EU is misreading the speed and strategic significance of AI. American systems become essential to cybersecurity, productivity, and scientific research, while Washington increasingly controls and restricts access. China, meanwhile, consolidates its strength in robotics and industrial AI. The EU responds with sovereignty initiatives, procurement mandates, and domestic champions, but fails to build sufficient infrastructure or bargaining power. By 2031, its remaining leverage is concentrated in ASML, leaving EU leaders caught between American and Chinese coercion.

The details are, of course, speculative, as the authors readily acknowledge. The underlying diagnosis is harder to dismiss. The EU does not lack strategies, declarations, funding announcements, or statements of ambition. Rather more crucially, it struggles to turn them into funded, coordinated, and politically durable action: announced investments are too often treated as though they were already mobilised; sovereignty is declared before any infrastructure exists; implementation is assumed to follow naturally from publication.

This is not simply a story about excessive regulation. It is a problem of political will, fragmented responsibility, underpowered industrial strategy, and an unwillingness to accept opportunity costs. Many of the bottlenecks are already painfully familiar: energy, compute, talent, cybersecurity, procurement, data, and adoption. Meanwhile, the unresolved questions are predominantly organisational: who is responsible for acting, who controls resources, and who can be held accountable for delivery? Rules without capacity become dependence management rather than sovereignty.

Where Europe 2031 goes wrong

Scenarios do not merely describe possible futures. Rather, they make some futures vivid and leave others obscured. Europe 2031 generates urgency largely by accepting Silicon Valley’s account of technological inevitability and progress. This development is frontier-led, compute-intensive, AGI-adjacent, and moving rapidly along a single historical trajectory. Once this teleology has been accepted, there is but one race, and the possible and preferable responses follow easily—more compute, faster permitting, looser regulation, deeper integration with hyperscalers, and centralised mobilisation.

I do not intend to make the case that frontier models or compute are unimportant. Nor am I saying that the EU can wish away American technological dominance (a challenge which extends beyond the realm of AI, too). The problem is that AI is not one technology or one industry, and therefore isn’t one race. Rather, AI is a category of technologies, and encompasses models, infrastructure, robotics, industrial systems, scientific applications, public services, and the institutional arrangements through which these are deployed. Capability is therefore not the same as reliable capability, and neither automatically produces useful organisational adoption.

Thus, the way the race is defined determines where agency can be found. If the only meaningful contest is building increasingly general frontier models, the EU is simply behind. If there are multiple contests (which there are), the EU may be behind in some while retaining considerable room to choose where and how it competes. Treating the American trajectory as technological necessity also smuggles in a political-economic preference. The EU is asked to preserve its social models by becoming more like the system from which it is supposedly seeking autonomy: Silicon Valley, but with GDPR paperwork.

The report’s Franco-German focus reinforces this narrowing. Almost all meaningful action occurs through a Paris-Berlin-Brussels constellation. This is understandable if the objective is to reach the EU’s most powerful decision-makers, but it also reproduces the institutional habits the report criticises. Solutions are imagined through the same central actors whose limited coordination and repeated failure to follow through constitute the problem.

The EU’s diversity certainly produces friction, but it is also a strategic resource. Smaller member states have long had to navigate technological dependence, limited scale, energy constraints, public-sector digitalisation and large organisational transformation, and geopolitical exposure without being able to dictate the terms. After all, Estonia brings experience in digital government; the Nordics in energy, public services, and innovation power; the Netherlands in semiconductors and infrastructure; and other member states have their own combinations of industrial, regulatory, and geopolitical capacity. The EU’s problem is not that it contains too many different systems; the problem is that it has failed to turn those differences into coordinated strength.

What the EU should do

Put succinctly, the EU has a commitment problem, and its first task must be to convert urgency into actual commitment. Here the report and I are in full agreement. This calls for mechanisms that bind institutions to delivery: real funding rather than creative accounting, clear ownership of strategic priorities, procurement pathways that create customers for EU-based firms, and public investment tied to deployment rather than another cycle of pilots. Energy, compute, cybersecurity, talent, industrial policy, and public-sector adoption cannot continue to be treated as separate files. Data-centre planning depends on grid capacity; firms cannot scale without customers and infrastructure; public institutions cannot govern systems they lack the capacity to understand.

Most crucially form my vantage point is that the EU must build around EU strengths. The strategic question should not be whether the EU can produce “its own OpenAI” as a matter of prestige. Instead of hunting for its own white elephant, its focus must be where the EU can build capabilities that others need, trust, and cannot readily replace. Industrial and physical AI, robotics, scientific research, energy systems, healthcare, semiconductors and photonics, AI assurance, and trusted data infrastructure are not consolation prizes. They connect to an existing industrial and institutional base in which reliability, specialised knowledge, and legitimacy matter.

This ought to be a deliberately diffused strategy. The EU does not need one centre directing every aspect of AI development. Rather, it needs a coordination architecture through which different member states can lead where they possess capacity and credibility. Coalitions of member states should pursue particular capabilities within shared strategic objectives, rather than waiting for all twenty-seven to move in synchrony. The European Space Agency provides at least partial precedent: common commitments coexist with optional programmes in which states invest according to their interests and strengths. Diversity need not mean fragmentation if institutions are designed to make different capabilities compound.

A strategy of coordinated diffusion will also work alongside leverage built beyond the EU. The aim is not autarky, and there are many middle powers out there who will want to balance their commitments and dependencies vis-à-vis the US and China. The United Kingdom, Japan, Singapore, and other countries with substantial technological capabilities cannot individually match either the US or China, but collectively they can control important parts of their respective AI stacks and support each other’s strategic autonomy. This is something the report points to, but which needs to be further emphasised.

All of this also requires a much more mature approach to both the United States and China. The EU should not be naïve about Chinese state power, surveillance, industrial strategy and coercive capability. Nor should it assume that dependence on the US is safe simply because it is “the devil you know”. Sovereignty does not require owning everything domestically, but it does demand enough capacity and enough alternatives to choose, negotiate, refuse, and recover.

Why the wrong race matters to all of us

Europe 2031 succeeds as a warning because it makes strategic dependence feel concrete. Its central lesson, however, should not be that the EU must run faster along a path defined elsewhere. The EU needs greater AI capacity, but it also needs a clearer account of what that capacity is for and which political and social arrangements it is intended to sustain.

Crucially, these same challenges extend well beyond just the European Union and the European continent. Whilst ASEAN is not the EU, and does not intend to be the EU— lacking an equivalent supranational authority and comprising members that differ even more sharply in capacity, political organisation, and strategic ambition—it nonetheless faces a related danger of becoming a deployment zone for systems, infrastructures, standards, and assumptions developed elsewhere. The broader lesson is that meaningful AI strategy begins from where a region actually stands—from its own institutions, capabilities, vulnerabilities, and potential partners. The greatest danger is not always losing someone else’s race: it is failing to ask whether it was the right race to enter at all.