Introduction
European policymakers are moving to make personal data more accessible for artificial intelligence development, even as the continent's reliance on foreign-built models and overseas infrastructure deepens. Recent cloud outages, export restrictions on advanced AI systems, cyberattacks, and experiments in automated political persuasion have exposed the vulnerabilities of this dependence. The central question is no longer simply who builds the most powerful AI, but whether European democracies can safeguard their economies, national security, and democratic institutions without depending on decisions made in other jurisdictions.
What Happened
On October 21, 2026, ambassadors from EU member states agreed on the Council's negotiating position for the Digital Omnibus, a legislative package intended to streamline European digital rules, including provisions of the GDPR. One element addresses the use of personal data for training and operating AI systems under the legal basis of legitimate interest in Article 6 of the GDPR.
This approach is not entirely new. The European Data Protection Board acknowledged in December 2024 that legitimate interest may, under specific conditions, permit the development and use of AI models. However, the Council clarified that creating an AI system is not automatically a legitimate interest and must be evaluated case by case. The proposal also changed the planned right to object: an earlier Commission draft included an unconditional right to object, but that was never implemented. Instead, the Council pointed to Article 21 of the GDPR, which allows individuals to object based on grounds related to their particular situation.
A major technical hurdle remains. Once data has been used to train a large AI model, removing its influence in a verifiable way—known as machine unlearning—is extremely difficult without retraining the model from scratch. The Digital Omnibus text is not yet law; the European Parliament must still adopt its position before inter-institutional negotiations begin. Meanwhile, a deeper question persists: even if data rules are simplified, who actually has the computing power, energy, and capital to turn that data into cutting-edge models?
Why This Matters
The resources required for frontier AI are growing rapidly. According to Epoch AI, the computing power used to train the most advanced language models has increased roughly fivefold annually since 2020, while training costs have risen about 2.4 times per year since 2016. At the same time, algorithmic improvements have roughly halved the compute needed to achieve the same performance every eight months. This creates a paradox: catching up to existing performance is becoming cheaper, but pushing the frontier forward demands ever-greater resources.
Development is highly concentrated. Stanford's AI Index attributes 50 significant models to the United States and 30 to China in 2025, compared with only two in Europe—one French and one British. The gap is not fixed, however. In March 2026, the leading U.S. model outperformed the top Chinese model by only 2.7 percent in user-preference rankings, and an open-weight model briefly matched the best U.S. system in early 2025.
Europe is not standing still. In June 2026, the Commission selected the EUROPA consortium, led by Italian firm Domyn, to build a state-of-the-art, open-source European model across all 24 official EU languages using public computing resources. The group can access up to 2.5 percent of the EuroHPC network's capacity for one year, though the model had not been released by November 2026. The Commission has also called for up to seven AI gigafactories, with application deadlines running from December 2026 to February 2027. Companies such as Mistral AI are pursuing their own infrastructure. Still, the timing lags behind the pace of frontier development, and U.S. investment operates on a vastly different scale.
The risks are already visible. Surveys by Germany's digital association Bitkom found that 71 percent of companies using or evaluating cloud services rely on American providers, and 85 percent believe Germany is too dependent on U.S. cloud infrastructure. In AI, 76 percent of German firms using the technology depend on ChatGPT, while no surveyed company reported using Mistral.
Technical failures illustrate the exposure. In October 2025, a DNS flaw affecting Amazon Web Services in the U.S. East region disrupted banking services, mobile networks, and AI-powered search tools. Political interference is also a factor: in June 2026, U.S. export controls forced Anthropic to suspend access to its Claude models for foreign users, including its own employees, before access was restored weeks later. Legal risks persist as well; a Microsoft executive testified before the French Senate that he could not guarantee French citizens' data would never be handed to U.S. authorities under the CLOUD Act.
Key Takeaways
- The EU's Digital Omnibus could simplify GDPR-compliant data use for AI, but it is not yet law and does not solve Europe's infrastructure gap.
- Frontier AI development is concentrated in the U.S. and China, while Europe has released only a small number of significant models despite recent initiatives like EUROPA and AI gigafactories.
- European businesses and public services depend heavily on American cloud providers and AI models, leaving them vulnerable to outages, export controls, and foreign data-access laws.
- AI is accelerating both cyberattacks and cyberdefense, shortening the window between vulnerability disclosure and exploitation.
- Research suggests AI chatbots can measurably influence political attitudes and civic actions, raising concerns about the integrity of democratic debate.
- Europe's strategy should focus on reducing dependence on single suppliers and ensuring essential services can operate if key AI systems become unavailable.
Conclusion
Europe faces a strategic choice. Making it easier to use personal data for AI is only half the equation; the continent must also invest in developing and deploying the technology within its own borders. The goal is not necessarily to outpace the United States or China in raw model performance, but to ensure that hospitals, energy grids, banks, and government agencies can continue functioning even if a critical foreign AI system is restricted or fails.
The challenge extends beyond Europe. Any democracy that relies on technology controlled elsewhere must ask how much autonomy it can afford to surrender. Democratic institutions cannot depend on the goodwill of foreign governments or private corporations to keep essential services running. The decisions made today about data, infrastructure, and model development will determine whether European citizens retain control over their own democratic future.

Discussion
Join the conversation
Thoughtful reactions, questions, and follow-up ideas help shape the next story.