Germany’s digital minister called on the European Union this week to accelerate its push for AI independence, arguing that the bloc’s reliance on American infrastructure has become a strategic risk. The comments, made in an interview, cited a recent safety incident at OpenAI as the latest evidence that Europe cannot outsource the core of its AI capabilities.
The minister said the EU needs to move faster on three fronts: foundation models, computing power and the ability to test AI systems for safety. Europe has talked about these goals for years, but the gap between its ambitions and its capabilities has widened as American companies pour tens of billions into the underlying infrastructure.
The OpenAI incident, in which a safety test reportedly went out of control at the company’s facilities, sharpened the argument. If Europe depends on American companies for the models it uses and the compute they run on, the minister argued, it has no independent way to verify those systems or to respond when they fail.
The timing is delicate. European companies have been adopting American AI models at a rapid clip, and some of the continent’s largest firms have signed agreements with U.S. providers. The minister is not proposing to cut those ties; the case is for building an alternative, so that Europe is not dependent on a single source.
The ambition has a price tag. Building foundation models requires clusters of advanced chips, trained on infrastructure that costs billions of dollars. Europe has the talent and the research base, but its startups have struggled to raise the kind of capital that flows to American labs, and no European company has built a frontier model that matches the leading American systems.
The compute question is the hardest. Europe’s share of global data-center capacity has fallen as the AI buildout concentrates in the United States, and the continent has been slow to approve the new facilities it needs. Power constraints, permitting delays and fragmented national markets have all slowed the buildout.
The testing question is newer. Europe has regulators with experience in safety standards, and the AI Act gives the bloc a legal framework for oversight. What it lacks is the technical capacity to evaluate frontier models on its own, a gap that matters if the most advanced systems are built elsewhere.
The minister’s comments echo a broader European debate. France has pushed its own AI investments, and the EU has funded research networks and supercomputer projects. But the scale of American spending has made the European effort look like an appendix to someone else’s story.
The industry’s response has been mixed. European startups welcome the attention but warn that policy support without funding will not close the gap. The companies building data centers in Europe say the real constraint is regulatory speed, not capital. Both groups agree on one point: the window for Europe to build an independent AI capability is not open forever.
The policy framework already exists. The EU’s AI Act, which took effect in stages, gives the bloc a legal structure for oversight, and European regulators have been among the most active in enforcing it. What the act does not provide is the technology it regulates, and that gap is what the minister’s push is meant to close.
Europe has assets to build on. The bloc has funded a network of supercomputers and so-called AI factories, research clusters where academics and startups can train models, and France’s Mistral AI has raised billions of euros to build frontier systems. The pieces exist; the question is whether they can be scaled to match American investment.
The compute constraint is the binding one. Advanced chips remain the bottleneck for any AI buildout, and Europe’s access to them depends on export rules set in Washington and manufacturing capacity in Asia. A European AI strategy that cannot secure compute reliably is a strategy with a ceiling, industry executives said.
The political case has its own timeline. European elections and budget cycles move slowly, and the infrastructure that AI requires takes years to permit and build. The minister’s argument, that dependence is a risk, is one that has been made before; what is different this year is that it is being made by a senior official of the bloc’s largest economy.
The minister’s comments also carry a message for European companies. Adoption of American AI tools is not the problem, the argument goes; dependence is. The push is for a dual-track approach, in which European businesses can use the best available models while the continent builds the capability to field its own. The distinction is easier to state than to implement.
The immediate consequence of the minister’s push may be bureaucratic. New programs, new funding vehicles and new safety institutes take years to build, and the technology will not wait. But the argument, that dependence on a single country’s AI infrastructure is a vulnerability, has moved from the margins to the center of European policy in a matter of weeks.
