Artificial intelligence is moving from the lab to the command centre, and the European Union is increasingly treating it less like a general-purpose technology and more like a strategic asset with geopolitical consequences. In remarks that underline a growing shift in European thinking, Henna Virkkunen, the EU’s digital chief, warned that AI is becoming a “geopolitical weapon” and that Europe cannot afford to rely on an “unpredictable Washington” for the capabilities that underpin security and competitiveness.
The point is not simply that the United States leads in AI research or that American companies dominate parts of the supply chain. It is that AI—because it can be scaled, deployed quickly, and integrated into decision-making systems—has become a lever states can pull. That lever can be used for influence operations, for economic pressure, for military advantage, and for shaping the rules of the game through export controls, procurement choices, and standards. In this view, the risk for Europe is not only technological dependence but strategic uncertainty: the possibility that access to critical AI capabilities could change abruptly with political priorities, regulatory decisions, or national security assessments across the Atlantic.
Virkkunen’s warning lands at a moment when governments are racing to translate AI ambition into industrial capacity. The conversation has moved beyond model performance benchmarks and into questions that look more like energy policy or defence procurement: Who owns the compute? Who can manufacture the chips? How resilient are data pipelines? What happens when supply routes are disrupted? Which legal frameworks govern cross-border deployment? And how quickly can a country or bloc build alternatives if access to key components becomes constrained?
Europe’s concern is that AI’s strategic value is increasingly tied to inputs—compute, chips, cloud infrastructure, data access, and the ability to train and fine-tune systems at scale. Those inputs are not neutral. They are shaped by national industrial policies, export regimes, and corporate strategies. When those inputs are concentrated in a small number of jurisdictions, the resulting dependency can become a vulnerability. Virkkunen’s phrase about an “unpredictable Washington” captures the fear that even if today’s cooperation is smooth, tomorrow’s constraints might arrive without warning.
To understand why the EU is framing AI as a geopolitical weapon, it helps to look at what “weaponisation” means in practice. It does not always involve autonomous drones or battlefield robots. Often it is subtler: AI-enabled intelligence analysis that compresses the time between data collection and actionable insight; AI systems that improve targeting, logistics, and surveillance; AI-driven cyber operations that accelerate reconnaissance and exploitation; and AI tools that amplify propaganda and disinformation by generating content at scale in multiple languages and styles. In each case, the advantage comes from speed, scale, and adaptability—qualities that can tilt outcomes even when the underlying hardware is not dramatically different.
There is also a second layer: AI can influence the environment in which decisions are made. If one side can model scenarios more effectively, detect patterns earlier, or automate parts of policy implementation, it can shape negotiations and crisis management. That is why the EU’s framing matters. It suggests that AI is not merely a tool that can be used by anyone; it is a capability that can restructure power relationships.
Virkkunen’s message implies that Europe should treat AI capability-building as a strategic programme rather than a purely economic one. That means investing in the full stack, not just in research. It means building domestic or trusted compute capacity, strengthening procurement pathways for public-sector use, and ensuring that regulation supports innovation without leaving Europe dependent on external providers for core capabilities. It also means preparing for scenarios where access to certain technologies is restricted—whether due to export controls, sanctions, or sudden shifts in national security policy.
One unique aspect of the EU’s approach is its insistence that resilience must be engineered, not hoped for. In many technology sectors, Europe has historically relied on a mix of regulation, market integration, and selective industrial support. But AI’s dependence on high-end compute and specialised hardware changes the equation. Resilience requires redundancy: multiple suppliers, multiple routes for data and compute, and the ability to switch architectures or training strategies when constraints appear. It also requires governance capacity—teams that can evaluate risks, audit systems, and ensure compliance across borders.
This is where the EU’s warning becomes more than rhetoric. If AI is a geopolitical weapon, then the ability to deploy it safely and effectively becomes part of national and collective security. That includes the ability to verify what models do, to monitor for misuse, and to prevent vulnerabilities from being exploited. It also includes the ability to maintain continuity when systems are updated or when vendors change terms. Dependence on a single provider can turn routine maintenance into a strategic negotiation.
The “unpredictable Washington” concern also reflects a broader European anxiety about how quickly US policy can change. Even when the US and EU share values, the incentives that drive policy can differ. US decisions may prioritise domestic industrial strategy, national security assessments, or alliance management in ways that do not align neatly with European timelines. For Europe, the problem is not disagreement; it is timing. AI development cycles are fast, and the cost of waiting for clarity can be high. If Europe’s strategic planning assumes stable access to US capabilities, it may find itself behind when conditions shift.
At the same time, the EU is not arguing for isolation. The challenge is to avoid a situation where Europe’s strategic autonomy is limited to what it can negotiate rather than what it can build. Autonomy here does not mean cutting ties; it means having options. Options require investment, coordination, and a willingness to treat AI infrastructure as strategic.
That infrastructure includes compute and chips, but also the ecosystem around them. Training frontier models demands large-scale data processing and energy-intensive operations. Inference—running models in real-world applications—also requires significant resources, especially when systems are used continuously across industries. Europe’s ability to scale AI depends on whether it can secure enough compute capacity at competitive cost, whether it can attract talent and engineering teams, and whether it can create procurement mechanisms that encourage adoption while maintaining safeguards.
The EU’s warning implicitly points to a gap between policy ambition and operational capability. Many European countries have AI strategies, but turning those strategies into sustained capacity is difficult. It requires long-term funding, coordination across member states, and a clear path from research to deployment. It also requires aligning incentives so that companies see a stable market for AI services and government agencies see a reliable supply of compliant, secure systems.
Another dimension is the geopolitics of standards. AI governance is not only about safety and ethics; it is also about shaping how systems are built and how they interoperate. Standards influence procurement requirements, certification processes, and the technical expectations that vendors must meet. If Europe’s regulatory framework becomes a de facto standard, it can increase leverage. But if Europe’s regulatory approach is too dependent on external technical assumptions, it can also deepen dependency. Virkkunen’s warning therefore signals a desire to ensure that Europe’s governance choices are grounded in capabilities Europe can sustain.
There is also the question of data. AI performance often depends on access to high-quality datasets, and data access is shaped by law, privacy rules, and commercial agreements. In a geopolitical context, data can become a strategic resource. If Europe relies heavily on external platforms for data processing or model training, it may face constraints on what can be shared, what can be stored, and what can be used for improvement. That can limit Europe’s ability to tailor AI systems to local needs, including language, culture, and regulatory requirements.
The EU’s framing also highlights a shift in how policymakers think about competition. Traditional industrial policy focuses on manufacturing jobs, export markets, and supply chains. AI competition adds a new layer: the ability to integrate AI into critical sectors such as healthcare, transport, energy, and public administration. When AI becomes embedded in these sectors, it becomes harder to replace. That makes early dependency particularly risky. If a country or bloc adopts AI systems that are difficult to replicate later, it may lock itself into vendor ecosystems.
This is why the EU’s warning resonates with a broader theme: the strategic importance of “switchability.” In other words, Europe needs to ensure that it can change providers, architectures, and deployment models without losing operational continuity. Switchability is not a slogan; it requires technical design choices, contractual flexibility, and the ability to run models locally or in trusted environments. It also requires skills—engineers who can operate and adapt systems, not just purchase them.
Virkkunen’s comments can be read as a call for Europe to treat AI as a strategic capability similar to telecommunications or energy infrastructure. Those sectors are governed by long-term planning because they shape national resilience. AI, in this analogy, is becoming infrastructure for decision-making and automation. If that is true, then the EU’s role is not only to regulate but to coordinate investment and ensure that member states can collectively build capacity.
Yet there is a tension. Europe’s regulatory culture emphasises caution, transparency, and risk management. Those values are important, but they can slow deployment if not paired with industrial support and clear pathways for scaling. The EU’s challenge is to maintain trust and safety while still enabling rapid innovation. If Europe’s governance is perceived as too restrictive, companies may choose to develop and deploy elsewhere. If it is too permissive, Europe may import systems that are difficult to audit or that carry hidden risks. Striking the balance requires both policy sophistication and technical competence.
The “geopolitical weapon” framing also raises uncomfortable questions about how Europe will manage the dual-use nature of AI. Many AI systems can be used for benign purposes—fraud detection, medical imaging, customer service—but the same capabilities can be repurposed for surveillance, cyberattacks, or influence operations. That means Europe’s strategic autonomy is not only about having models; it is about having the ability to control their deployment and to prevent misuse. Control requires governance, monitoring, and enforcement mechanisms that can keep pace with rapid iteration.
In practical terms, Europe may need to expand its capacity for AI assurance: testing, auditing, red-teaming, and verification