Tips & Tricks for a successful HORIZON-CL4-2027-04-DATA-03 proposal
Opening
17 November 2026
Deadline
Keywords
quantum computing
hardware orchestration
RIA
federated AI
HPC systems
data handling
AI data
edge computing
sustainable AI
decentral processing
Your microfluidic SME partner for Horizon Europe
We take care of microfluidic engineering, work on valorization and optimize the proposal with you
HORIZON-CL4-2027-04-DATA-03: New approaches for decentralized, federated and sustainable AI data processing
The perpetual problem with Europe’s plans for AI is a shortage of computing power, an overconcentration and the costs of energy; these are topics that no one likes to bring up. The Commission is attempting to fund solutions to this through this topic. The idea here is to remove the processing of AI data away from a small group of cloud providers and decentralize it across edge, cloud and HPC systems. This uses computing efficiently and removes reliance on any single processor architecture. If you have a group looking at federated systems, heterogeneous compute orchestration, and energy-efficient AI pipelines, then this call may be of interest to you.
Download the MIC Horizon Europe 2026/2027 Calls Calendar:
Discover more!
Administrative facts: what do we know about the HORIZON-CL4-2027-04-DATA-03 call?
Which call is it, and when is the opening and the deadline?
- Call name: DIGITAL
- Call identifier: HORIZON-CL4-2027-04
- Destination: Developing an agile and secure single market and infrastructure for data services and trustworthy artificial intelligence services
- Topic: HORIZON-CL4-2027-04-DATA-03: New approaches for decentralized, federated and sustainable AI data processing
- Opening date: 17 November 2026
- Deadline: 18 March 2027
- Type of action: Research and Innovation Action (RIA)
What about the budget and estimated size of the project?
- Overall budget for this topic: EUR 35.00 million
- Number of projects expected to be funded: 2
- Budget per project: around EUR 17.50 million
What are the key eligibility and evaluation conditions?
- Standard Horizon Europe RIA thresholds apply (Excellence 4/5, Impact 4/5, Implementation 3/5)
- Participation restricted to EU Member States, Iceland, Norway, plus Canada, Israel, Republic of Korea, New Zealand, Switzerland and the United Kingdom
- Entities established in China are not eligible
- Entities controlled by a non-eligible country must provide guarantees assessed by their country of establishment
- High-risk suppliers of mobile network communication equipment cannot submit guarantees
- TRL expected: start at TRL 3, reach TRL 6 to 7 by end of project
- Transfer of ownership or exclusive licensing of results must be notified to the granting authority; objection possible up to four years after end of action
Scientific range: what does the Commission expect from the HORIZON-CL4-2027-04-DATA-03 grant?
What outcomes are expected?
The Commission seeks practical working models of distributed AI compute architectures. The goals here are to resolve the bottleneck caused by centralized, energy hungry compute. This call wants to see a demonstrably more sustainable and efficient method of running AI workloads across different computing infrastructures from the edge to HPC. The term alternative” is key to understanding the direction of the call. It seeks to demonstrate that it’s possible for European actors to train, fine-tune, and implement AI systems without solely relying on existing hyperscalers.
What is within scope?
- New approaches to distributed/federated AI systems that extend beyond current federated learning and include concepts such as model compression and scaling between multiple compute infrastructures.
- Orchestration of AI workflows over the entire compute range including edge, cloud, HPC and even neuromorphic and quantum computing.
- Use of in-memory computing and hardware/software approximation methods to enhance efficiency.
- Management of data quality and integrity in a distributed setting, including tackling unbalanced datasets, optimizing data transfer volumes, and safeguarding privacy in the distributed environment.
- Post-quantum cryptography implementation for data both in transit and at rest.
- Tools to monitor and measure sustainability of AI pipelines across the edge-to-cloud continuum and from data collection through to deployment.
- At least two use cases across different domains that can be replicated elsewhere.
- Theoretical or benchmark work is not encouraged; instead, demonstrable systems with tangible impacts on energy consumption, latency, and computing diversity are sought.
What are the specifically proposed research directions?
- Federated compute continuum architectures: addressing the entire AI lifecycle, from collection to deployment, within a federated system.
- Heterogeneous hardware orchestration: ensuring AI portability across diverse processors, including neuromorphic and quantum hardware.
- Data consistency and quality in decentralized processing: tools to ensure balanced datasets, reduce data transfer and protect privacy.
- End-to-end energy and sustainability measurement: methods to not only monitor but also enhance energy efficiency of heterogeneous hardware throughout the AI lifecycle.
- Two distinct use cases in different sectors are required and need to be transferable.
Scientific strategy: how can you enhance your chances of being funded through HORIZON-CL4-2027-04-DATA-03?
What scientific choices matter most?
- Don’t focus on federated learning, expand beyond it. The proposal text states that a system similar to a standard federated learning system will receive a low novelty score. The architecture is what they are looking for.
- Heterogeneous hardware is non-negotiable. The call specifies the need to demonstrate heterogeneous compute across both the hardware itself and the infrastructure it uses. Systems only supporting single processor types or compute infrastructures will fail.
- Prioritize energy sustainability from the outset. Proposals have often included energy monitoring as a separate work package but failed to integrate it with the overall architecture. This must be demonstrated through energy costs built into compute tasks.
- Choose your use cases carefully. At least two distinct use cases in different domains are needed, focusing on sectors of strategic importance for European competitiveness, such as manufacturing, energy, health, and mobility.
- Highlight privacy and security by design, and post-quantum cryptography.
Consortium and proposal-writing plan: what works best with this type of call?
- This is a competitive call, with only two projects expected to receive funding at approximately 17.5 million Euros each. Aim for consortia of 8 to 14 partners (potentially more if you have two significantly different use case domains).
- You’ll want some of the leading researchers in distributed systems, a partner with HPC experience, and preferably a cloud or edge platform provider. Include partners with neuromorphic or quantum hardware research experience if possible, even as minor participants, as this signals ambition.
- An innovative SME with expertise in AI orchestration tools, energy-aware scheduling, or federated data management is essential; this adds an exploitable dimension and signals a gap in existing technology that the SME can address.
- The use cases will require end-user partners in at least two different sectors; industrial partners, public infrastructure operators, and health data providers will be highly valued. Ensure these partners are genuinely involved, not just letter-of-support signatories.
- Eligibility criteria are stringent. Carefully verify the legal form of all partners, particularly if there’s any doubt about the control exercised by non-eligible entities.
How would microfluidics contribute to this topic?
Although the call for AI compute infrastructure may seem unrelated to microfluidics at first glance, the latter can provide valuable testing grounds for distributed AI systems. Controlled environments for data acquisition are necessary for the development of real-world use cases in areas like the biomedical and environmental sectors as targeted in the Commission’s call.
- For the health use case, for example, microfluidic systems can acquire continuous streams of relevant biological data (e.g., cellular responses, drug screening results), providing high-quality test data to validate distributed AI systems.
- Since lab-on-chip technology is used in point-of-care settings, it is a perfect example of edge computing. It allows on-device data preprocessing before feeding the results into a larger training dataset in a distributed fashion.
- Another important use case is environmental monitoring, where microfluidic sensors deployed in water treatment plants or in field environments can collect crucial data.
The MIC already brings its expertise in microfluidics to Horizon Europe:
H2020-NMBP-TR-IND-2020
Microfluidic platform to study the interaction of cancer cells with lymphatic tissue
H2020-LC-GD-2020-3
Toxicology assessment of pharmaceutical products on a placenta-on-chip model
FAQ – HORIZON-CL4-2027-04-DATA-03
What is HORIZON-CL4-2027-04-DATA-03 about in one sentence?
It is a Research and Innovation Action that funds new architectures to decentralised, federated and sustainable AI data processing across the edge-cloud-HPC continuum, aiming to decrease reliance of Europe on a limited number of centralised hyperscalers.
How much funding is available and how many projects will be selected?
The overall indicative budget will be EUR 35.00 million and the Commission is only expected to finance 2 projects at around EUR 17.50 million each. There will be stiff competition due to limited number of slots.
When does the call open and when is the deadline?
The call is opened on 17 November 2026, and closed on 18 March 2027, at 17:00 Brussels time. The deadline can be postponed by the Director-General in charge up to two months.
What type of action is this and what TRL is expected?
The activities will commence at TRL 3 and will achieve TRL 6 to 7 at the project completion, which is a significant leap and must have plausible milestones in between.
Who can participate and which countries are excluded?
It is open to legal entities incorporated in EU Member States, Iceland, Norway, as well as Canada, Israel, Republic of Korea, New Zealand, Switzerland and the United Kingdom. Organisations incorporated in China cannot qualify.
What does the Commission mean by 'beyond federated learning'?
The conventional federated learning can be regarded as the baseline. New compute continuum architectures with model compression, distributed orchestration, scaling across multiple infrastructures are expected by the Commission. The submission of a simple federated learning project is likely to have low scores on novelty.
Why are heterogeneous hardware and the compute continuum so important?
The entire aspect of the subject matter is to show that AI workloads are able to execute on a wide variety of processors and infrastructures, such as neuromorphic and quantum hardware. Solutions that address a single chip family or cloud provider lack the intent of the topic.
Why does sustainability and energy efficiency matter for evaluators?
European AI infrastructure is concerned with energy. The call specifically requests the tools which can measure, monitor and enhance end-to-end energy efficiency. The common pitfall is to treat sustainability as a work package, which is not related to the architecture.
What kind of consortium has the best chance of being funded?
The mix of about 8-14 partners distributed systems researchers, HPC providers, cloud or edge platforms, end-users in at least two sectors, and at least one innovative SME capable of bringing orchestration, scheduling or data management tools to the table. Check the Funding and Tenders Portal for more information.
How can microfluidics realistically contribute to this AI compute topic?
Microfluidic systems produce biologic and environmental data streams that are both continuous and of high quality and work under controlled conditions. These can be the natural edges of the network (point-of-care, field deployment), and thus they are tangible and plausible applications to test distributed AI architectures.