Artificial intelligence (AI) is emerging as a general-purpose technology, transforming all sectors of the economy and society, from scientific discovery and industrial development to public administration and defense. As the US and China compete over AI development and its adoption abroad, the European Union (EU) finds itself in the middle of this rivalry, balancing opportunities for cooperation with emerging dependencies. Even countries with established research institutions and specialized industries remain dependent on capabilities concentrated in the US and China. Reproducing every component domestically is costly and, in the near term, often impractical. International cooperation therefore offers an important pathway to strengthening European AI capabilities by connecting research and industry with computing infrastructure, models, investment, and expertise abroad. However, the terms of such cooperation shape which capabilities European actors develop themselves, which they obtain from others, and how dependent they remain on decisions made outside the EU.
This paper maps AI cooperation between EU institutions, member states, European companies, and research institutes and twelve international partners: the United States, China, the United Kingdom, India, South Korea, Japan, Canada, Singapore, Taiwan, Malaysia, the United Arab Emirates, and Israel. It examines cooperation across four areas: infrastructure, research and talent development, commercial partnerships and investment, and government agreements and regulatory cooperation. By identifying emerging patterns of cooperation and the contributions of individual partners, the paper assesses how international partnerships can help address gaps in European AI capabilities while also creating new risks and dependencies. While the EU is often portrayed as lagging behind global AI leaders, the findings show that it is a sought-after partner across these areas, with international cooperation serving as an important strategy for catching up.
Executive summary:
- This paper maps the cooperation of the European Union (EU), its member states, European companies, and research institutes on artificial intelligence (AI) with twelve partner countries: the United States (US), China, the United Kingdom (UK), India, South Korea, Japan, Canada, Singapore, Taiwan, Malaysia, the United Arab Emirates (UAE), and Israel. It covers collaborations in infrastructure, research and talent development, commercial partnerships and investment, and government agreements and regulatory cooperation. While the EU is often seen as lagging in AI, this research shows it is a sought-after partner in all these areas, with external collaborations serving as an important catch-up strategy.
- International cooperation enables the EU to access compute, capital and frontier AI systems it lacks domestically. However, as foreign investors finance and often build Europe’s new data centers, control over these facilities depends on the terms each member state negotiates. Washington can withdraw access to the most capable American models with a decision, as happened in June 2026 with Anthropic’s most advanced models. Some European research programs and suppliers take part in systems in Israel and China that may be tied to surveillance or military uses, which are contrary to EU’s rules or interests. Cooperation with partner countries is therefore essential, but it can also create strategic dependencies that need active management. Cooperation should therefore be seen as a complementary means of developing the EU’s own AI capabilities, not a substitute.
- The EU’s strengths lie in public computing infrastructure, skilled personnel, and development of regulatory frameworks. The EU has introduced the world’s first comprehensive AI law, operates some of the world’s fastest public supercomputers, and has more AI professionals per capita than the US or China. By contrast, its weaknesses include limited frontier-model development, slow AI adoption, scarce private investment, and dependence on US models, cloud services, and hardware.
- The European Commission’s AI Continent Action Plan of April 2025 aims to overcome these weaknesses through developing public AI Factories and Gigafactories and promoting strategies for data, talent, and AI adoption. However, critics doubt whether public computing resources are sufficient, timely, and aligned with demand. Adoption, funding, and computing capacity are also concentrated in a small number of Western member states, and much of Central and Eastern Europe remains behind.
- Foreign capital finances most of Europe’s new AI data centers, with American investors providing the largest share, followed by Japan and Canada. Hardware from the American chip designer Nvidia is used in nearly every AI infrastructure project. The EU’s public supercomputers are built on Japanese and American systems, with European firms working as assemblers and operators. In the semiconductor supply chain, the EU leads in photolithography equipment. AI servers are assembled in Europe by companies such as Taiwan’s Foxconn in Czechia and France’s Bull in Angers.
- Foreign companies are increasingly establishing AI laboratories at European universities and research institutes. Conversely, European companies also open laboratories in Asia and North America. Health and life sciences are the largest area of research cooperation with partner countries, followed by telecommunications and manufacturing. Most skilling programs in Europe come from American companies, led by Microsoft, Google, OpenAI, and Amazon Web Services, and the largest non-American one is Huawei’s Seeds for the Future.
- Indian IT services firms operate the IT systems of many European companies, which increasingly involve AI. China is currently the main partner in automotive AI applications, as some German carmakers develop AI for their Chinese models in joint ventures with Chinese companies. American investors provide most of the foreign capital raised by European AI startups, and European investors mainly fund startups abroad in the UK, Israel, the US, and India. EU companies mainly supply specialized AI systems to governments outside the EU, such as for defense, transport and public administration.
- EU institutions cooperate on AI with Japan, South Korea, Singapore and Canada through respective Digital Partnerships. Since 2022, the EU has structured its AI dialogue with India through the Trade and Technology Council. The Trade and Technology Council with the US is no longer active, but the EU Commission said in April 2026 it was in talks with Washington to set up a new dialogue on digital technologies and markets. AI is also a cooperation area in the Strategic Partnership Agreement with the UAE, under negotiation since December 2025. France, Germany, and Poland are the most active member states in bilateral AI diplomacy, and India, the UAE, and Canada have signed the most agreements with individual EU member states.
- The US is both the primary enabler and main partner of European AI, as well as its largest potential strategic vulnerability. It provides the compute, models, and enterprise software that underpin European business, and its systems are more deeply integrated into European defense than those of any other partner. Washington, however, increasingly treats model capability and access to advanced AI chips as a national security asset.
- Besides the US, the UK is the EU’s closest partner, participating in Horizon Europe and the EU supercomputing network. Canada is also an important research partner, particularly in health and life sciences. Cohere, Canada’s leading frontier model developer, is merging with the German company Aleph Alpha to create a transatlantic sovereign AI provider.
- Taiwan produces the world’s, including Europe’s, AI servers and graphics processing units (GPUs). Taiwan also cooperates closely with Central and Eastern European countries, especially Czechia. South Korean memory makers Samsung and SK Hynix, as well as Taiwan’s TSMC, depend on extreme ultraviolet lithography machines from the Dutch company ASML. ASML in turn counts them among its largest customers.
- Japan is a close partner in the buildup of the European data centers, where its firms invest in significantly. Furthermore, Japanese suppliers are providing crucial parts for the European High-Performance Computing Joint Undertaking (EuroHPC) supercomputers infrastructure.
- China is the main partner in development and deployment of AI for the automotive sector. Chinese cloud providers, namely Tencent Cloud and Alibaba Cloud, are expanding their data center footprint and AI services across the EU. However, only a few member states are actively deepening formal ties with China on AI because of growing security and competitiveness risks posed by China’s AI development and related sectors.
- Beyond regulatory cooperation, India’s IT services firms manage systems for European companies and provide services necessary for successful AI implementation in business operations.
- The UAE invests capital in French and Italian data centers, while actively balancing its approach to AI by maintaining ties to both Washington and Beijing. Singapore also balances between both AI superpowers and serves as the Southeast Asian base of European firms. In Malaysia, European firms mainly supply renewable power, network equipment and AI training to the growing data center industry.
- The EU has commercially beneficial AI cooperation with Israel, especially at the start-up level. But the intertwined nature of civilian and defense AI development in Israel creates risks for the EU that EU-funded or EU-developed technologies might be deployed in ongoing Israeli military operations.
- South Korea is the only partner with a binding cross-sectoral AI law, which borrows the AI Act’s risk-based terms but has no prohibited practices and far lower fines. Japan and Taiwan have framework laws without penalties; China has binding rules for specific services, such as generative AI and deepfakes; and other partners rely mostly on existing laws and voluntary guidance. Most signatories of the EU’s General-Purpose AI Code of Practice are US companies, but no Chinese developer has signed it, even though Chinese open-weight models are increasingly deployed in the EU.