Blueprints for a Pan-African Digital Research Infrastructure Take Shape

African research communities are increasingly active in artificial intelligence, data science, Internet of Things, cybersecurity and distributed computing, yet the infrastructure beneath them remains fragmented, unevenly distributed and often unable to sustain large-scale collaborative experimentation. A new open-access paper in EPJ Research Infrastructures, published on 4 September 2026, sets out the vision, methodology and deployment roadmap of DIGITAfrica, a Horizon Europe initiative launched in January 2025 to prepare the design of a pan-African Research Infrastructure in digital sciences. The project brings together thirteen African and European institutions with expertise in digital sciences, research infrastructure design and large-scale experimental platforms, and its explicit goal is not to hand European infrastructure to African researchers but to co-design the technical and governance foundations of an African-led ecosystem of interoperable digital research infrastructures of different scales and specialisations.

The starting point for the design is a sober assessment of the continent’s digital research landscape. The authors identify five recurring barriers documented across multiple studies: uneven distribution and fragmentation of existing research infrastructures; limited funding and investment mechanisms for large-scale infrastructure; constrained computing capacity and limited high-speed connectivity; shortages of skilled personnel able to deploy, operate and maintain complex systems; and insufficient integration of national initiatives into regional or continental frameworks. Within DIGITAfrica itself, these constraints were examined through an iterative demand analysis in five consortium countries—Cameroon, Kenya, Senegal, South Africa and Tunisia—using desk research, expert interviews, a pilot questionnaire completed by 27 research organisations in Tunisia, and a consultation workshop with more than 60 participants. The authors are careful to note that because the five countries were selected through consortium membership, the analysis is exploratory and does not estimate how prevalent individual barriers are across the continent as a whole.

The country assessments paint a picture of real capability coexisting with structural weakness. South Africa hosts a relatively mature digital research environment with national cyberinfrastructure initiatives, high-performance computing facilities and research networks connecting universities and laboratories. Senegal has recently invested in national cyberinfrastructure and high-performance computing, Tunisia maintains dedicated research centres for intelligent systems and digital services, Kenya hosts innovation laboratories and digital experimentation environments within universities, and emerging research communities in Cameroon are increasingly active in artificial intelligence and data science. Yet existing infrastructures—university computing clusters, national research and education networks, specialised laboratories and platforms built independently under national projects—are typically not interconnected through interoperable services or shared access mechanisms. Access to advanced facilities therefore varies significantly between institutions, limiting collaboration and reducing the efficiency of research investment.

The demand analysis also surfaced more specific constraints. Assessed institutions reported that limited access to high-performance computing and specialised hardware restricts computationally intensive experiments, while researchers frequently experience insufficient bandwidth, unstable connections and limited integration with international research and education networks, in some contexts compounded by inconsistent electricity supply. For artificial intelligence specifically, datasets are often fragmented, insufficiently curated or difficult to share across institutions, pushing researchers toward global datasets that do not reflect local contexts and limiting the relevance and performance of AI models in Africa. Shortages of personnel able to operate HPC, cloud, cybersecurity and data-management systems are aggravated by limited career opportunities and the migration of highly skilled researchers. Long-term financial sustainability remains a concern where foreign funding accounts for a substantial share of research expenditure, and dependence on externally operated platforms raises questions of data governance and digital sovereignty that the African Union Data Policy Framework seeks to address through interoperable governance arrangements.

Interest in a coordinated solution is not hypothetical. At DIGITAfrica’s inaugural stakeholder workshop in Cape Town in April 2025, participants from academia, government and industry emphasised the need for federated, interoperable testbeds and platforms supporting experimentation, data sharing and collaborative research across national boundaries, with particular attention to domains relevant to African societies such as digital agriculture, telemedicine, climate monitoring and smart urban systems. A second consultation workshop followed in Tunis on 29 April 2026, and the consortium has catalogued more than 100 stakeholders, grouped into academia and research organisations on one side and government and funding bodies on the other, with a High-Level Policy Forum planned as a venue for strategic alignment and policy dialogue.

Methodologically, the project draws heavily on two European Strategy Forum on Research Infrastructures (ESFRI) landmarks added to the 2021 roadmap: SLICES, which provides infrastructure for experimentation on networking and distributed computing, and SoBigData, which supplies data, methods and computational services for social mining and data-intensive research. Both operate distributed service models in which geographically separated nodes contribute specific capabilities while the whole offers unified access, allowing progressive scaling as institutions join. Both also implement services supporting Open Science and FAIR data management—SLICES through a Metadata Repository Service, SoBigData through a method catalogue of documented, reusable analytical methods—and both run training programmes for researchers and technical staff. The authors stress, however, that these are technical and organisational references rather than transferable templates: deployment conditions, resources, governance models and user requirements differ across the DIGITAfrica partner countries, and each candidate mechanism is assessed against the demand analysis before being adopted, simplified or replaced with a locally maintainable alternative.

The core methodological output is the concept of DIGITAfrica Blueprints: structured specifications for deployable services, interoperability mechanisms and operational guidelines that institutions can adopt, adapt and scale. Each blueprint will specify at minimum a baseline service catalogue covering identity, access, compute, data, experimentation, reproducibility and training; interoperability standards and reference architectures enabling cross-institution and cross-border collaboration, including access policy; and deployment and operation guidelines covering infrastructure setup, connectivity, resource provisioning and lifecycle management. The architecture is organised in five tiers: Tier 0 micro-edge nodes for resource-constrained environments such as rural schools or clinics, designed for low power and operation without stable connectivity; Tier 1 local edge sites at campus scale; Tier 2 regional hubs interconnecting multiple Tier-1 sites; Tier 3 national infrastructures with consolidated governance and large-scale orchestration; and Tier 4 a pan-African federation enabling cross-border collaboration, whose functions are analogous to the Central Hub in the ESFRI reference model, though without requiring all services to be co-located at one site.

Two representative blueprints demonstrate the methodology. The edge-interoperable AI/ML blueprint specifies an experimental framework for deploying and reproducing AI and machine learning workflows across distributed edge-cloud infrastructures, allowing workloads to run close to their data sources—including on offline or resource-constrained edge nodes—with models, datasets and experimental artefacts synchronised opportunistically when connectivity allows. Executing selected processing tasks near the data source can reduce latency and network traffic and supports data-minimising workflows relevant to sensitive applications in health and education, while optional lightweight AI accelerators, energy-aware execution and locally available renewable energy sources such as solar power feature at the lower tiers. The heterogeneous networking blueprint, built on the SLICES 5G/post-5G blueprint and inspired by the GreenDIGIT sustainability framework, adopts dual User Plane Function anchoring so traffic can be handled at either edge or central sites—improving resilience and keeping critical services running locally during backhaul disruptions—and combines WiFi, 5G, LoRaWAN and satellite links in a hybrid multi-access radio strategy.

Underpinning both blueprints is a shared service layer implemented with lightweight open-source technologies: a resource pool orchestrated through Kubernetes-compatible tools, with K3s for lower tiers and Docker Compose for Tier 0; identity management through Keycloak and OpenID Connect using self-contained tokens so services need not query a central portal for every access decision; file storage via NFS and object storage via an S3-compatible interface, abstracted so the same interfaces work across tiers; and browser-based notebook environments through Jupyter and JupyterHub, chosen because they combine executable code, documentation, data and results in a single reproducible workflow. A first operational implementation of the AI/ML blueprint has been validated in three deployment modes—standalone Docker, single-node K3s, and a multi-node K3s cluster with multi-user JupyterHub access—on Raspberry Pi 5 clusters, NVIDIA Jetson devices and Ubuntu virtual machines, and the code is publicly available through the project’s repository.

Early validation results are encouraging but deliberately scoped. In a federated-learning demonstrator for cassava-disease classification on commodity Android smartphones, using a MobileNetV2 feature extractor with a trainable classification head coordinated through the Flower framework, server-side classification accuracy rose from 28.2 percent to 66.9 percent over 100 communication rounds with two heterogeneous smartphones and no client dropouts, while raw crop images stayed on the devices and only model parameters were exchanged; larger-scale experiments over mobile and intermittent connections remain to be performed. A cloud-native 5G network-slicing testbed built on OpenAirInterface validated slice admission, session establishment, control- and data-plane separation and traffic isolation, though real slice-aware user equipment and quantitative quality-of-service evaluation are not yet included. A third educational testbed showed that a K3s configuration reduced CPU utilisation compared with a Docker Compose deployment by 93 percent for Keycloak, 82 percent for PostgreSQL and 74 percent for Redis. Selected components were deployed during training activities in Nairobi, Tunis and Hanoi reaching more than 77 participants, and future work will extend the platform toward Tier-2 regional hubs, with a planned Research Passport intended to control access to services, datasets, GPU clusters and experiments across the emerging federation.

Subject of Research: Design methodology and roadmap for a pan-African digital sciences research infrastructure

Article Title: Designing a Pan-African Research Infrastructure for Digital Sciences: DIGITAfrica Vision, Blueprint Methodology, and Roadmap

Article References: Bruno, R., Natilli, M., Fdida, S., Saucez, D., Makris, N., Barroso, M., Mwangama, J., Fendji, J. L., Ben Saoud, N. B., Ozianyi, V., Gueye, C. A. B., Magedanz, T., Sarli, A. C., & Belloum, A. (2026). Designing a Pan-African Research Infrastructure for Digital Sciences: DIGITAfrica Vision, Blueprint Methodology, and Roadmap. EPJ Research Infrastructures, 10(1), Article 27. https://doi.org/10.1007/s41781-026-00184-5

Image Credits: AI Generated

DOI: 10.1007/s41781-026-00184-5

Keywords: DIGITAfrica, research infrastructure, Africa, digital sciences, Horizon Europe, ESFRI, SLICES, SoBigData, edge computing, federated learning, 5G network slicing, open science