Europe’s concern is shifting from competitiveness to control: whether foreign governments or companies could restrict access to AI systems that public services and firms rely on.
Commission Executive Vice-President Henna Virkkunen has warned that artificial intelligence infrastructure and access controls are becoming instruments of geopolitical power. In a Financial Times interview, the EU digital chief linked AI dependence to Europe’s broader technology-sovereignty agenda, including concerns that foreign access restrictions could disrupt critical services.
The warning is not abstract. The FT reported that Brussels viewed the US decision to temporarily impose export controls on Anthropic’s leading AI models in June 2026 as a reminder that access to frontier systems can be restricted by political decisions. Even if controls are later reversed, the episode shows how quickly a dependency can become leverage.
The Commission’s own Cloud and AI Development Act sets out the policy context. It says Europe needs more cloud, AI and data-centre capacity, and identifies overreliance on non-EU cloud providers as a risk to digital autonomy and resilience. The issue now extends from data location to model access, computing power and service continuity.
EU Today has previously covered how European cloud groups demanded emergency action over Broadcom and VMware, a competition case with sovereignty implications. AI dependence is broader. It concerns whether European administrations, companies and security services can rely on systems whose availability may be controlled outside the EU.
The “kill switch” risk is the clearest political image. If a foreign government can require a provider to cut access, change model availability, restrict compute exports or expose data under national law, European users face a dependency that is not merely commercial. It becomes strategic.
This does not mean Europe can or should isolate itself. The Commission’s tech-sovereignty agenda explicitly describes sovereignty as reducing risky dependencies while keeping markets open. Europe will continue using US technology, cloud services and AI models. The question is whether it has credible alternatives for critical use cases.
Building those alternatives is difficult. Frontier AI requires chips, energy, data centres, talent, capital and massive training budgets. European firms such as Mistral have strong technology, but competing with US hyperscalers and model providers requires procurement support, financing and access to compute. Sovereignty cannot be declared; it must be bought, built and maintained.
Public procurement may be decisive. If European institutions and governments want domestic or trusted AI capacity, they must create demand. That could mean buying European models for public administration, supporting sovereign cloud frameworks and funding compute infrastructure through EU-European Investment Bank mechanisms. Without customers, European alternatives will remain smaller than the dependency they are meant to reduce.
Security services and sensitive public functions are likely to be the first test. A ministry using a foreign AI service for routine productivity faces one level of risk. A police, defence, health or border agency using AI in operational workflows faces another. The EU may need differentiated rules rather than a single sovereignty requirement for all AI use.
The transatlantic relationship will also be tested. Washington may see European sovereignty rules as discrimination against US firms. Brussels will argue that resilience requires choice and control. A trusted-partner scheme could reduce tension, but only if it provides enforceable guarantees that access will not be withdrawn abruptly.
The deeper issue is that AI is becoming infrastructure. Once companies and administrations embed models into workflows, losing access can disrupt operations. That makes AI dependency closer to energy, cloud or payment infrastructure than ordinary software procurement.
Virkkunen’s warning should therefore be read as a policy marker. Europe is no longer discussing AI sovereignty only to create champions. It is asking whether critical functions can be interrupted from outside the Union. That is a harder and more practical question.
The next phase will be judged by execution rather than vocabulary. Europe has no shortage of strategies on cloud, data and AI; the test is whether they translate into compute capacity, procurement demand and viable companies able to serve public and private clients at scale. Dependence becomes political leverage when users cannot credibly switch providers or when regulation is written around infrastructures Europe does not control. Reducing that vulnerability will require open standards, trusted cloud capacity, public-sector demand and investment in European model developers. The goal is not to cut Europe off from global technology, but to ensure that cooperation remains a choice rather than a dependency trap.
