Smarter, Not Harder: A Research-Driven Agenda for Europe’s AI Sovereignty

As the EU lags behind the US and China in AI, a smarter focus on software, decentralised systems, and human capital may be the best way forward. By harnessing algorithmic innovation and synthetic data, while avoiding costly hardware races, Europe can chart its own path to AI sovereignty. This approach ensures competitiveness and is in line with Europe’s core values.

The stakes in the global AI race go far beyond technological progress, touching on economic resilience, security, and societal well-being. However, the European Union (EU) faces significant challenges that jeopardise its position in this geo-economic competition, especially compared to the United States and China.

Despite initiatives such as the European Chips Act and the creation of “AI factories” using supercomputing capacity, a persistent investment gap and structural inefficiencies are hampering domestic progress. In this context, what role can software and research play in supporting the EU’s plan to establish itself as a sovereign player in Artificial Intelligence?

Europe’s AI gap: status quo and bottlenecks

To begin with, the gap between Europe and its global competitors is evident in the development of frontier AI models. Data from Epoch AI shows that the United States has consistently led the way in building large-scale models with billions of parameters, leveraging unparalleled computing resources and access to massive datasets.

China has also made rapid progress in the past two years, supported by strategic policies and state-directed investment. In contrast, Europe has made only modest progress, according to this data, with smaller AI models and significantly limited computational resources.

This discrepancy reflects not only a lack of access to advanced chips and data centres, but also Europe’s reliance on non-European cloud providers.

Küsters, Weil and Carugati (2024), Europe‘s Path to Competitiveness in the Global AI Race, FNF Global Innovation Hub Policy Brief

Europe’s stringent regulatory framework, including the General Data Protection Regulation (GDPR), further exacerbates these challenges.

While GDPR sets essential standards for data protection, it limits the availability of data needed to train advanced AI systems. In contrast, the US and China are using vast amounts of data to fuel their AI development. The regulatory burden, while helpful for privacy governance, risks inadvertently stifling innovation by creating compliance challenges that deter investment in AI projects.

This comparative lag extends to innovation output, as private company data suggest. The United States continues to dominate AI publications and patents, reflecting a strong synergy between academia and industry. China follows closely, excelling particularly in patents, reflecting its focus on commercialisation and applied research. Europe, on the other hand, suffers from weaker links between academia and industry and a pronounced talent retention problem.

The result is an ecosystem that struggles to match the speed and impact of its global competitors.

Unlocking potential: solutions for a competitive edge

However, within these challenges lie opportunities for Europe to play a distinctive and competitive role in the development of AI. One promising but under-researched avenue is the adoption of distributed computing models, in particular so-called federated learning. This approach enables AI training across distributed systems, optimising existing computational resources without centralising data or infrastructure.

By reducing reliance on resource-intensive centralised systems, federated learning can address the computational bottlenecks that currently constrain Europe’s AI ambitions. Supporting these types of computational approaches also matches the decentralized nature of Europe’s economy and the principles of subsidiarity and proportionality.

Synthetic data also offers a transformative opportunity for European firms. By generating artificial datasets that simulate real-world conditions, synthetic data can circumvent the constraints imposed by GDPR while maintaining high-quality standards for AI training.

Recent advances, such as simulated environments for AI development in the field of medicine, highlight the potential for synthetic data to democratise access to training resources, mitigating the traditional advantages of data-rich regions such as the US and China.

While media reports like to focus on high-end chips, the most critical frontier is algorithmic innovation. Recent breakthroughs, such as, for example, DeepMind’s JEST and advances in AI architectures beyond the well-known transformers, suggest an untapped potential for software-led advances to improve AI capabilities without a proportional increase in hardware requirements.

Europe can capitalise on these developments by investing in smaller, more efficient AI models (so-called small language models, or SLMs) that achieve competitive performance with reduced computational requirements. Moreover, hybrid neuro-symbolic systems combine symbolic reasoning with neural networks to create AI systems that are both resource efficient and capable of complex tasks.

Recommendations: a smarter path to AI sovereignty

What are the implications of these findings? In terms of strategic direction, Europe needs to focus on strengthening its talent and innovation ecosystem. Greater investment in education and training programmes is essential to build a skilled workforce capable of addressing “hard” bottlenecks in AI development.

Europe’s talent pool, combined with a robust open-source ecosystem, can foster collaboration and reduce dependence on external technologies. In particular, open-source projects provide a collaborative framework for accelerating AI innovation in a decentralized fashion while also helping to reduce cybersecurity risks.

Fair competition should be another key aspect of Europe’s next AI strategy. The currently popular pursuit of “national champions” through mergers and state aid risks stifling competition and innovation.

Instead, promoting a level playing field for start-ups and small and medium-sized enterprises can unlock the potential of a diverse and decentralised AI landscape. Such an approach is also in line with Europe’s core values of fairness and subsidiarity, ensuring that the benefits of AI development are widely shared.

Finally, the EU institutions must reshape their investment strategy to better align with emerging AI paradigms. Traditional approaches that emphasise hardware-centric competition with the US and China are both costly and unsustainable. Instead, Europe should channel its resources into areas where it can still lead – such as algorithmic efficiency, software innovation, and distributed computing. These smarter priorities leverage Europe’s strengths, including its regulatory expertise and decentralised culture, to achieve AI sovereignty on its own terms.

Ultimately, Europe’s path to AI leadership lies not in mimicking the strategies of its global competitors, but in charting a course that plays to its comparative advantages. By focusing on and supporting software innovation, distributed systems, and human capital, the new EU Commission can help build a European AI ecosystem that is both competitive and sustainable. This strategy not only addresses the geopolitical challenges of the global AI race, but also ensures that the transformative benefits of AI are accessible to all segments of society.

Note: This essay draws on insights from the paper “Europe’s Path to Competitiveness in the Global AI Race”, commissioned by the Friedrich Naumann Foundation for Freedom. The study, published in early December 2024, explores the implications of AI developments, with a focus on the core hardware and software supply chains dominated by the US and China. The full study is available at this link: https://shop.freiheit.org/#!/Publikation/1826