Why More Compute Won’t Guarantee AI Infrastructure Africa Needs

On 2 October, the United Nations Development Programme (UNDP) and the GSMA, the global mobile industry body, announced a partnership to scale African-led artificial intelligence under the timbuktoo initiative. Yet UNDP’s own six-country study found that supplying computing power alone does not ensure lasting AI use. For anyone tracking AI infrastructure in Africa, that finding challenges the assumption that more capacity means more AI.

The gap being closed is large. Africa holds roughly 17-18% of the world’s population but only 0.6% of global data centre capacity, according to the Africa Data Centres Association (ADCA).

Investors following the buildout are asking where capital actually converts into usage. This analysis offers a way to separate capacity that will be used from capacity that will sit idle.

Why does compute alone fail to deliver lasting AI use?

The investor’s instinct is simple: hardware is the scarce asset, so buy or build the hardware. UNDP’s study, “Local AI compute infrastructure in Africa”, tests that instinct and finds it incomplete.

What the six deployments show

The study examined six country-hosted deployments in Kenya, Malawi, Rwanda, South Africa, Togo and Zambia, covering 21 AI applications. It followed each from design and procurement through deployment and operation, treating every one as a governance and usage experiment rather than a hardware rollout. UNDP’s source reporting dates publication to 17 September, and its web page was updated on 3 October 2026.

The pattern was consistent. Installed GPUs (graphics processing units, the chips that run AI workloads) and servers were often used intermittently or sparsely where data quality, institutional ownership, operating funds or integration into government workflows was weak.

Main finding Supplying computing power alone does not ensure lasting use of AI.

The study names the conditions that must sit around the hardware:

• Access to usable data
• Skills among operators and users
• Governance arrangements
• Cybersecurity
• Operating funding
• Concrete use cases
• Reliable power and connectivity

For you as an investor, a facility’s value depends on the demand pipeline and institutional backing around it. Diligence should test who will use the capacity, and who pays to run it, before counting megawatts.

Concrete use cases are the missing link in many deployments, and AI adoption in African mining shows where industrial demand for compute is already forming among operators with budgets and data.

Why sovereignty complicates the picture

A February 2026 analysis from New America adds a further wrinkle. It warns that data centres built on weak grids, with limited domestic cloud capability, can entrench dependence on foreign providers rather than deliver digital sovereignty.

That matters because sovereignty concerns can later turn into policy risk for foreign-owned assets.

What the UNDP-GSMA partnership and ATLAS Umoja AI are actually building

The partnership reads as a direct answer to the study’s findings. Announced on 2 October through a UNDP Africa press release, and reported on 5 October by Samira Njoya of We Are Tech Africa, it aims to move locally built AI projects from concept to scaled use.

The partnership under timbuktoo

Its core elements are:

• Easier access to computing capacity for priority AI projects
• Links between African talent and practical use cases
• Support on funding, sustainability and scale-up conditions
• Use of University Innovation Pods, AI laboratories and universities
• Use of GSMA’s operator network to connect startups and public-sector users with telecom operators and digital ministries

ATLAS Umoja AI as the language layer

The partnership builds on GSMA’s ATLAS Umoja AI, launched in July 2026 with backing from the digital ministries of Benin, Kenya, Namibia, Nigeria and Togo. It gathers linguistic data and coordinates AI-related telecom data, infrastructure and policy. The need is stark: GSMA reports that under 0.1% of online resources are in African languages.

What this tells you is that the partnership is demand-side infrastructure. It may improve utilisation prospects for nearby capacity, but with no disclosed capital it is not a funded buildout to invest alongside.

Educational: How power and grid limits shape what can be built

Start with the physical question: can the grid feed it? Power, not investor appetite, sets the ceiling.

Why the grid sets the ceiling

A data centre needs firm, continuous power, because even brief outages interrupt AI workloads. Transmission and distribution (T&D) losses, the electricity lost moving power from plants to users, ran as high as 18-25% in several East and West African nations in 2025, according to ADCA. Generation is also often far from the urban hubs where data centres need to sit.

ADCA’s illustration is blunt: a hypothetical 1 GW campus would consume over 40% of a nation’s generation.

Three countries, three problems

W.Media describes Nigeria as unique in that most, if not all, data centre projects need to be accompanied by a power supply project.

Nigeria’s government said it added 1,000 MW of generation by December 2025, yet the grid remains far short. South Africa has the electricity but struggles to move it, and wheeling (sending privately generated power through the public grid) is the sticking point.

Wheeling reform is the sticking point in South Africa, and regional energy hubs that coordinate cross-border power flows point to one route by which stranded generation could eventually reach demand centres.

Kenya shows renewables alone do not solve capacity. The $1 billion, 100 MW Microsoft-G42 Naivasha project would reportedly need about one-third of Kenya’s roughly 3,000 MW installed capacity, and Semafor reports it is unlikely to proceed.

When you evaluate a project, the power supply arrangement is as much the investment as the building, and bundled generation changes both capital need and risk.

Where capital can work and where it is exposed

The opportunity is real. ADCA puts capacity at 0.6% of the global total, while an April 2026 consultancy analysis cites about 700 MW of colocation power, roughly 1.5% of the global figure, with a contracted pipeline near 1.8 GW by 2028. The sources differ because they measure different things: all data centre capacity versus colocation power.

Kenya’s market was worth about US$266 million in 2025 and is projected to reach US$805 million, with 240 MW by 2031, according to W.Media.

Structural versus cyclical

The sorting matters for you because structural risks need pricing into returns, while cyclical ones may reward patient investors who wait for reform milestones.

Risks to price in

• Overbuild: Axis Intelligence and the Institute of Internet Economics warn the sub-1% share could tempt catch-up builds that demand does not yet justify.
• Underuse: UNDP’s study shows installed compute can sit idle without governance, funding and demand.
• Fuel prices: Nigeria’s diesel reliance exposes operators to cost swings.
• Sovereignty: Governments may later renegotiate or restrict foreign ownership, per New America.
• Capital intensity: Bundling generation with IT infrastructure raises upfront spend.

Semafor reported in July 2026 that many prospective projects are being re-evaluated because of power realities. Kenya’s pattern suggests staged, power-linked builds are the more viable route. The research does not provide a continent-wide power cost per kilowatt-hour or a named quantification of currency risk, so both remain open questions.

What the evidence changes, and what it leaves open

Capacity, power, data and institutions have to move together. The UNDP-GSMA partnership addresses the talent, data and demand side, but not the grid.

Three signals are worth tracking:

1. Whether the partners disclose funding for the programme
2. Progress on grid access and wheeling reforms
3. Utilisation evidence from existing compute deployments

Projects that show movement on all three are better candidates for real usage than those showing only megawatts. Until those signals arrive, treat headlines without disclosed capital or power arrangements as intentions, not infrastructure.

This article is for informational purposes only and should not be considered financial advice. Investors should conduct their own research and consult with financial professionals before making investment decisions. Market projections are speculative and subject to change based on market developments.