However, a study published in 2026 and prepared for the European Commission already seeks to identify the AI fields that could be strategically important for EU research and innovation funding after 2027. The document is neither an adopted funding programme nor an official list of future calls for proposals. Rather, it is an initial technology roadmap, drawing on expert consultations, Delphi surveys (multi-round surveys designed to build expert consensus), publication and patent data, and market analyses to highlight the AI technologies on which Europe should build in the longer term.
One of the most important fields is foundation models. These are general-purpose AI models trained on large and diverse datasets that can subsequently serve as the basis for numerous specific applications. Europe does not necessarily need to focus on simply replicating American general-purpose models. Sovereign, multilingual and sector-specific models—for example, for healthcare, law, public administration or industry—may be far more significant. This is both a technological and a sovereignty issue: if an increasing number of digital services are built on the same handful of foundation models developed outside Europe, this could create substantial long-term dependence.
Closely related to this is the development of AI agents. Agents are no longer merely systems that answer questions, but autonomous or semi-autonomous AI systems capable of planning multiple steps towards a goal, using tools and external systems, and then carrying out tasks. A corporate AI agent, for example, could collect data, analyse documents, query different systems and use the results to prepare recommendations for decision-making. The technological potential is enormous, but reliability, auditability and the limits of such a system’s autonomy are equally important questions. Future funding may therefore seek not simply ‘smarter agents’, but safe and interpretable agent architectures.
Another major direction is AI that does more than recognise patterns: it can reason and explain its conclusions in a more traceable way. This includes causal AI and, in particular, neurosymbolic AI. The latter seeks to combine the ability of neural networks to learn from data with classical, rule- and logic-based artificial intelligence. This could be important because one fundamental weakness of today’s AI models is the limited verifiability of their conclusions. Neurosymbolic approaches may offer lower data requirements, more transparent reasoning and, in some cases, more formal guarantees of correctness.
Advances in human–machine interfaces could likewise become a European strength. Intelligent interfaces, natural-language control, predictive assistance and even collaboration with robots are particularly important for European industry. The aim is not necessarily to replace people. Many applications may instead focus on making human work more effective with the support of AI-enabled machines or software. This approach aligns well with the EU’s frequently articulated vision of human-centric artificial intelligence.
AI safety, reliability and evaluation are also becoming distinct areas of research and funding. The study specifically identifies AI safety and security, AI evaluation, and trustworthy and explainable AI. As AI systems play an increasingly important role in healthcare, financial, industrial and public-administration decisions, demonstrating that a model actually works as intended will become ever more important. This will require new testing methods, benchmarks, security assessments and conformity-assessment technologies. Because of the EU AI Act, this is not merely a research question: over time, a market of its own could emerge for auditing and certifying AI systems.
Another important direction is Green and Frugal AI. One of the challenges of current AI development is that increasingly large models require ever more computing power and energy. For Europe, it could therefore be strategically important to develop models and hardware that deliver adequate performance with lower energy, data and computing requirements. Related to this is neuromorphic computing, which draws inspiration from the workings of the human brain to create extremely energy-efficient computing systems. (The human brain operates at approximately 20 watts, whereas modern AI data centres can require many millions of times as much power.)
Edge AI is also becoming increasingly important: this means running artificial intelligence directly on a machine, sensor, vehicle or other local device rather than in a remote cloud. It enables faster responses, better data protection and less dependence on the cloud. Edge AI is particularly well suited to Europe’s industrial strengths: automation, the Internet of Things (IoT), embedded systems and the semiconductor industry are all fields on which related projects can build.
Among longer-term, higher-risk research areas, the document also highlights the integration of quantum computing and AI, as well as unconventional computing paradigms, potentially including non-silicon-based approaches. These may not become mass-market applications within the next few years, but they could be strategically important for the EU because they concern the computational foundations of next-generation AI systems. The goal here is no longer simply to create better algorithms, but to rethink the hardware and physical principles on which the artificial intelligence of the future will operate.
Under the post-2027 funding scenario, 32% of resources would be allocated to applied research and technology development, 22% to testing and technology infrastructure, 19% to fundamental research, 14% to skills development and 13% to ecosystem building. This could be an important signal for future applicants. In upcoming EU AI programmes, simply demonstrating that a project ‘uses artificial intelligence’ will probably be increasingly insufficient. The emphasis is likely to be on which technological problem it solves, how closely it aligns with Europe’s strategic strengths, whether it reduces dependence on external technologies, whether it can be tested in real-world settings, and whether it can turn research results into competitive European technology. In this sense, the study can be viewed as an early map of EU AI R&D funding priorities after 2027.
Source: European Commission (2026). Study on strategic digital technologies for the EU Research and Innovation funding beyond 2027 – Annex 2: Technology Area Report – Artificial Intelligence. Prepared by Technopolis Group.