EU funding for AI projects: where to look
EU funding for AI projects comes mainly from three places: Horizon Europe funds AI research and collaborative innovation, Digital Europe funds deployment and capacity such as testing facilities and skills, and the EIC funds single companies commercialising high-risk AI innovation.
AI cuts across programmes rather than having one of its own, which is why “where do I apply?” is a harder question here than for most fields. The useful first step is deciding what kind of work you are proposing, because that determines the programme far more than the subject matter does.
Research, deployment, or commercialisation
Research and collaborative innovation → Horizon Europe. Advancing the state of the art, usually in a consortium with research organisations. AI appears both as a subject in the digital and industry cluster and as a method throughout the other clusters — health, climate, mobility, agriculture. Some of the least contested AI funding is in the application clusters rather than the digital one, because the competition there is with domain specialists rather than with every AI group in Europe.
Deployment and capacity → Digital Europe. Explicitly not a research programme. It funds testing and experimentation facilities, common data spaces, advanced skills, and putting existing technology into use. If your project is “make this work at scale in a real setting”, this is a better fit than Horizon, and proposals often lose because they were sent to the wrong one.
Commercialisation by a single company → the EIC Accelerator. High-risk innovation close to market, single applicant, grant with an optional investment component. See EU grants for startups.
Adoption by an individual company → national schemes. In Portugal, Portugal 2030 and the PRR fund digitalisation including AI adoption. Less competitive, more paperwork in your own language, and oriented to investment rather than novelty.
What AI proposals get wrong
Proposing a model rather than a problem. Calls fund outcomes. A proposal whose ambition is a better architecture, with the application chosen afterwards, reads as a solution looking for a problem — and evaluators say so explicitly in their comments.
Ignoring data availability. The single most common gap. Reviewers ask where the training data comes from, whether you have the rights to it, and what happens if the partner holding it withdraws. A project whose data plan is “we will obtain a dataset” is scored down for implementation risk regardless of the science.
Treating trustworthiness as boilerplate. European calls take fairness, transparency, robustness and human oversight seriously, and a paragraph of generic ethics language is visibly a paragraph of generic ethics language. Where the system is high-risk under the AI Act, the regulatory pathway belongs in the impact and implementation sections, not in an annex.
Underestimating the compute and skills line. Budget the compute honestly. A plan that quietly assumes free access to infrastructure you have not secured is a delivery risk that reviewers can see.
The regulatory dimension
The AI Act classifies systems by risk and attaches obligations accordingly. For a funded project this is mostly a design constraint rather than an eligibility question — but calls increasingly expect proposals to state where their intended application sits and how compliance is approached.
The honest way to handle it is to work out the category early, because it changes what you need to build: documentation, data governance, human oversight and post-market monitoring are cheaper designed in than retrofitted in year three.
Getting the current obligations right means reading the current text, not a summary. Timelines for different provisions differ, and a page that quoted specific dates would be stale before it was useful.
Finding the calls
AI-relevant topics appear across several programmes and several clusters, published to different calendars, with the work programmes announcing them a year or more ahead. The searching problem is genuinely harder here than in a single-programme field — we compare the ways to handle it, starting with the portal’s own free notifications.
Frequently asked questions
Which EU programmes fund artificial intelligence projects?
Horizon Europe funds AI research and collaborative innovation, mainly through its digital and industry cluster. Digital Europe funds deployment, testing facilities, data spaces and skills rather than research. The EIC funds single companies bringing high-risk AI innovation to market. National programmes distributing EU structural funds also support AI adoption by companies.
What is the difference between Horizon Europe and Digital Europe for AI?
Horizon Europe funds work that advances the state of the art, done by consortia including research organisations. Digital Europe funds putting existing capability into use — testing and experimentation facilities, data infrastructure, skills programmes. If your project is research, it is Horizon; if it is adoption at scale, it is Digital Europe.
Does the AI Act affect EU-funded AI projects?
It affects what you build, not whether you can be funded. Obligations depend on the risk category your system falls into, and calls increasingly expect proposals to show awareness of the regulatory position of the intended application. Treat it as part of the impact case rather than an afterthought.
Can a company apply alone for AI funding?
Through the EIC Accelerator, yes. Most Horizon Europe AI calls are collaborative and require at least three independent legal entities from three different member states or associated countries. National schemes usually accept single applicants.
By · Last reviewed: 2026-08-12