AI funding, handled.
AI is novel model architecture, data infrastructure and ML systems pushed past what off-the-shelf tools can do. We secure the non-dilutive funding for the research-grade engineering, not the routine integration.
Research-grade engineering, strong funding fit
AI R&D is eligible where it goes beyond standard application: new model architectures, training and inference methods, and data infrastructure built under technical uncertainty. Routine integration of an existing model does not qualify, novel method development does. Non-dilutive funding covers the research-heavy build.
Why R&D funding fits AI
Four reasons AI companies are well positioned for German R&D funding.
Ongoing, fundable R&D
Model research, training and retraining are continuous R&D, not a one-off. The Research Allowance supports that work year over year.
Compute and hardware count too
Since 2024, depreciation on GPU servers and compute hardware used directly in an R&D project can be included in the eligible base, which matters for training-heavy work.
Researcher-heavy teams
Funding is built around R&D wage costs, and AI teams are full of the ML researchers, data and infrastructure engineers those costs cover.
Strong digital pull
The Horizon Europe digital, industry and space cluster (Cluster 4) and the EIC actively fund AI deep tech, giving teams a route to larger grants on top of the national instruments.
What AI activities qualify
Four categories of AI R&D that consistently qualify. The common thread is technical uncertainty.
Model architecture and methods
- New model architectures or training methods beyond standard baselines
- Optimization, distillation or efficiency methods under uncertainty
- Approaches where established models do not solve the problem
Data infrastructure and pipelines
- Data systems built for novel training or inference needs
- Labeling, augmentation or synthetic-data methods developed from scratch
- Throughput and reliability engineering under new constraints
Applied ML under uncertainty
- Models for domains where standard methods do not yet reach the needed accuracy
- Evaluation and validation methods for a novel model
- Robustness, drift and safety work that resolves technical unknowns
Developer tools and systems
- Tooling, frameworks or runtimes with novel architecture
- Inference, serving or orchestration systems built under technical risk
- MLOps methods developed where no standard solution exists
The line that decides a claim
The line that decides a claim is technical uncertainty. A quick orientation for AI:
Usually qualifies
- New model architectures or training methods beyond standard baselines
- Data or serving systems built under genuine technical risk
- Evaluation and validation methods for a novel model
Usually does not
- Integrating an existing model into an application
- Standard software development, testing and debugging
- Configuration, deployment and data migration
How much AI companies typically receive
Indicative ranges based on R&D team size and intensity. Actual figures depend on eligible costs and the funder's decision.
Mostly the Research Allowance on a core research and engineering team.
The Research Allowance stacked with a ZIM grant on a defined development project.
Toward the Research Allowance ceiling, plus stacked ZIM and Horizon Europe digital funding.
* Indicative figures. The actual amount depends on company size, eligible costs and the programs you qualify for.
The kind of AI work that qualifies
If it carries genuine technical risk and novelty, it usually counts. A few examples:
New architectures and training methods that beat standard baselines.
Pipelines and data systems built for novel needs.
Models for domains standard methods cannot yet solve.
Tooling and serving systems with genuine technical novelty.
Best-fit funding for AI
We usually combine two or three of these for the largest total.
Research Allowance
25-35% of R&D costs back as cash, retroactive to 2022.
Explore program →Grant + equityEIC Accelerator
EUR 2.5M grant + up to EUR 10M equity for AI deep tech.
Explore program →GrantEurostars
~50%, up to EUR 500k per partner for international R&D.
Explore program →GrantHorizon Europe
up to 100% for collaborative digital R&I consortia.
Explore program →AI funding, answered
The novel work around them can qualify: new architecture, training methods or systems that resolve genuine technical uncertainty. Standard integration of an existing model does not. We help you separate the two.
Yes. The Research Allowance is paid out even with no profit, which suits the R&D-heavy early phase typical in AI.
Since 2024, depreciation on hardware used exclusively and directly in an R&D project can be included in the eligible base. We assess which of your hardware qualifies.
Yes. Contract research is eligible at 70% of the cost, so working with a university or institute does not exclude you, it adds to the eligible base.
The Research Allowance and ZIM cover the national base, while Horizon Europe (Cluster 4) and the EIC fund larger, more ambitious programmes. We design the stack so they reinforce each other.
Ask us anything
Tell us about your project and we'll assess your case by hand. No prep, no obligation. Prefer to talk?