EU-Zuschüsse 2027: Wo ein KI-Startup tatsächlich einen Antrag stellen sollte

One line. We reviewed roughly 250 calls from the European Commission's Funding & Tenders portal for the 2027 cycle and scored each one for a startup building autonomous AI systems. Below is the top of that ranking, the scoring method, and the reasoning behind every call we decided to pursue or skip.

Why EU grants enter the plan at all

Most of the calls below are part of Horizon Europe and the Digital Europe Programme. They matter to a small AI company for three reasons. First, they fund the unglamorous half of the product: data pipelines, energy-efficient inference, standards work, and evaluation that paying customers rarely finance. Second, they force a written, reviewed technical proposal, which is itself useful discipline. Third, Ukraine is an associated country of Horizon Europe, so a Ukrainian AI startup can join or lead a work package as a full partner rather than a subcontractor.

The bet behind the ranking is specific: the market is already crowded with AI wrappers. The teams that win grant work are the ones that can make expensive compute cheap and auditable — not the ones shipping another chat interface. That is why energy-efficient inference and controlled agent orchestration score highest below, while anything that reduces to "a demo" scores lower.

The scoring method

Every call gets a 0–10 score on four axes, then a weighted sum. This is prioritisation support, not a guarantee of winning or of budget size.

Criterion	Weight	Meaning
Fit	0.30	How well the call matches an autonomous-AI stack: agents, model orchestration, cost-efficient inference, OSINT and data work.
Money	0.25	Typical grant or share for that action type.
Gewinn	0.25	Realistic win chance for a small company, assuming a strong EU partner.
Urgency	0.20	How close the submission window is.

Final = 0.30·Fit + 0.25·Money + 0.25·Win + 0.20·Urgency. Tiers: S ≥ 7.4 (invest immediately), A 7.0–7.3 (next months), B 6.4–6.9 (strong reserve), C 5.9–6.3 (selective), D < 5.9 (only as an invited partner).

Budget figures are the typical order of magnitude by action type: RIA/IA ≈ €5–15M per project, CSA ≈ €0.5–2M, cascades and DEP-SME actions ≈ €50–150k. Exact amounts live in each topic's text.

The top of the ranking

The full sheet has 25 positions. Here are the twelve calls we will actually spend time on, ordered by score.

The two S-tier calls are a direct hit for a company whose differentiator is cheap, governable inference: Frontier AI Initiative rewards efficiency in routing and running models, and DATA-09 is literally about energy-efficient AI data processing. The third S is the pragmatic one — AI4SME has a low entry threshold and is the closest practical result.

What we skip, and why

We consciously drop calls where our role would not be technical or where the win chance is lower than the cost of participating: MSCA COFUND and Doctoral Networks (university programmes, not company grants), SPACE and hardware components, EURATOM fusion, the EDF defence fund, New European Bauhaus and construction, and biotech without a wet lab. The rule is the same one we apply to product work: do not ship "another chat", ship a measurable result.

What we do first

1. Phase 0 (now → Nov 2026): cascade open calls — NCCFIUP / FORTIS / GroundWork (deadline 04.11.26) and Innowwide. Low overhead, first €50–150k, plus register PIC/LEAR and start partner search in parallel.
2. Phase 1 (→ Jan 2027): DEP AI4SME (14.01.27) and RAISE Automated Scientific Discovery (02.02.27, the earliest large deadline).
3. Phase 2 (→ Mar–Apr 2027, the core): the Cluster 4 set — DIGITAL-EMERGING-11, DATA-09, EMERGING-04, with DATA-03 und EMERGING-06 as reserve — plus CL5-D3-24 (energy) and MSCA Staff Exchanges for people and network.
4. Phase 3 (summer–autumn 2027): Cluster 3 — ECCC-01 (AI cyber), INFRA-02 (OSINT vs critical infrastructure, a rare exact match to our stack), FCT-01 (online harms).
5. Phase 4 (reserve): Health and Missions calls only when invited as a partner.

The honest risks

• Win scores for large RIA calls stay low without an academic anchor. A coordinating EU university partner, not a lone submission, is what makes the chance real.
• Budgets are approximate; a topic can narrow or shift its scope.
• "Forthcoming" windows must be re-verified on the portal before submitting.
• One track is not one application: we aim to submit two or three in a single wave to diversify the odds.

Where the data comes from

The ranking is built from a sample of the public EC Funding & Tenders portal (call identifier, pillar, programme, deadlines, budgets, change history), captured 2026-09-03. We keep our own dossier archive for this source and use the same archive to re-check the ranking — evidence first, opinion second.

Related surfaces

This ranking feeds two live directories. The startups directory carries the company profiles behind the calls (who is building the autonomous-AI stack, with GitHub and enrichment signals), and the investment funds directory lists the capital side. Grants, startups, and funds are three views of the same funnel — we are linking them into one analytics surface.