Africa's health data problem is not the data. It is power, bandwidth and storage

Most discussion of artificial intelligence in African health care starts with algorithms and ends with ethics. A review in Nature Communications starts several layers lower down, with the question of whether the continent's biomedical researchers can store, move and compute on the data in the first place [s1]. Its answer is a mixed one, and the specifics are more useful than the framing.

What has been built

The review is not a lament about absence. It documents a decade of substantial investment.

H3Africa — a $176 million investment from the US National Institutes of Health and Wellcome to investigate the genetic and environmental basis of communicable and non-communicable disease in Africa while building research capacity on the continent — is the anchor example [s1]. Its bioinformatics network, H3ABioNet, has spent more than a decade building skills across the full range of biological data work, from setting up IT infrastructure through to data management, analysis and interpretation [s1].

Newer initiatives follow the same pattern. The NIH-funded Harnessing Data Science for Health Discovery and Innovation in Africa consortium aims to build a network of African data scientists and to extract maximum value from limited datasets [s1]. The Africa Population Cohorts Consortium and a network of Genomic Centers of Excellence have capacity-building mandates, and the Kenya-led Data Science Without Borders initiative is working on platforms to apply data science to health data [s1].

Connectivity has improved too. Undersea backbone cables on the continent's eastern and western coasts — the EASSy and WACS consortia, composed mostly of African telecommunications companies — carry international traffic, though extending that connectivity inland imposes additional network costs on landlocked countries [s1]. National Research and Education Networks, the specialised providers that serve research and university traffic, have expanded through a series of EU-supported projects running from EUMEDCONNECT in 2004 through AfricaConnect2 to the current AfricaConnect3, with the remaining funding provided by African partners [s1]. A World Bank assessment using the NREN Capability Maturity Model found nine African NRENs qualifying at Level 6, the mature tier [s1]. Four — KENET in Kenya, TENET in South Africa, RENU in Uganda and ZAMREN in Zambia — already support data-intensive applications and the sharing of high-end computing assets [s1].

Where it stops

The review reports results from an HPC Discovery Survey of high-performance computing access. Of 74 total responses, 36 completed the full survey, representing 22 African countries, with the largest share of responses coming from South Africa [s1]. A further 18 respondents used computing resources but did not know what infrastructure sat underneath them, and so could complete only part of the survey [s1].

Among the 36 who completed it, 28 used a locally hosted high-performance computing system and nine used another institution's [s1]. None had access to, or used, cloud-based high-performance computing resources [s1].

That last finding is the one worth sitting with. The commercial cloud is usually presented as the solution to exactly this problem — infrastructure you rent instead of building. The review notes that while public cloud providers have made high-performance computing more accessible in general, many barriers to using commercial clouds persist for researchers in low- and middle-income countries, with the result that reliance falls back on local facilities [s1].

The obstacles respondents named are prosaic and physical: slow internet speeds, system downtime from loadshedding, limited storage capacity alongside inefficient data management, a shortage of bioinformatics skills for big data analysis, and a shortage of computing resources [s1]. One encouraging detail sits inside that: most of the computing environments described were implemented between 2022 and 2024, so the installed base is relatively new [s1].

The two-supercomputer problem

Concentration is the other structural feature. The review names South Africa's Centre for High Performance Computing in Cape Town, hosting a 1 PetaFlop system, and a 3.15 PetaFlop system at the African Supercomputing Center in Morocco, as notable investments [s1]. South Africa also has a national plan — the National Integrated Cyberinfrastructure System — while the Southern African Development Community has developed and approved a regional Cyberinfrastructure Framework [s1]. Most other countries, the review notes, have cyberinfrastructure development that is not guided by national, regional or continental frameworks, roadmaps or implementation plans [s1].

Platform work is trying to bridge the gap between the few institutions with serious compute and the many without. The eLwazi Open Data Science Platform, developed as part of the DS-I Africa consortium, builds on the Terra and Gen3 systems to give researchers workspaces that combine access to IT infrastructure, datasets and analysis tools [s1]. The African Open Science Platform has established three regional nodes: Egypt's National Authority for Remote Sensing and Space Sciences for the north, the African Institute for Capacity Development in Kenya for the east, and the UbuntuNet Alliance in Malawi for the south [s1].

Why this is a health story

The chain from a genomic dataset to a clinical decision runs through every link the survey found broken. Data science for health requires integrating heterogeneous data types — genomics, transcriptomics, proteomics, phenotypic and socio-demographic data, mobility, imaging, climate and infectious disease modelling, verbal autopsies, electronic health records — and doing so demands storage, transfer capacity and skills across the whole data lifecycle [s1].

A research group without reliable power cannot run a multi-day analysis. A group without a science network cannot move a large dataset without the transfer failing partway. A group without local bioinformatics skills cannot use the data it has already collected. None of those are AI problems, and none of them are solved by better models.

What the review asks for

The recommendations are infrastructural and institutional rather than technical: strengthening data and computing infrastructure, developing human expertise, and building toward a sustainable, pan-continental data science ecosystem [s1]. The review frames these as inputs to policy development and strategic planning [s1] — which is another way of saying that the binding constraints identified here are budget lines and national plans, not research questions.

The paper was received in May 2025 and accepted in March 2026 [s1], so its survey describes capacity as it stood before the most recent contraction in international research funding. That timing makes the document useful chiefly as a baseline: a reasonably detailed picture of what a decade of external investment built, published as the terms of that investment are changing.

Sources

• Unleashing potential: assessing Africa's readiness for the data science revolution to impact health, Nature Communications, 13 April 2026