Africa’s AI Gap: Infrastructure, Policy, and the Long Road to Frontier Models

Africa is home to 18% of the world’s population. The African Development Bank projects that AI could add $1 trillion to the continent’s GDP by 2035. Yet Africa holds less than 1% of global data center capacity and produces just 0.89% of global AI publications. That gap between potential and present reality is the central tension of AI on this continent.

I have been watching this space closely, both as someone building in tech and as someone from Malawi—a country where ‘AI strategy’ is not a phrase you hear in government corridors. Conversations about AI in Africa often fluctuate between expectations of leapfrogging and resignation about being left behind, but the reality is more complex. Progress exists, yet deep-rooted structural problems cannot be solved by a few well-funded startups.

The digitization problem

AI runs on data, compute, and connectivity. Africa is short on all three.

Internet penetration ranges from 26% to 68% across Africa, notably below the global average. About 600 million Africans lack electricity, making the continent home to over 80% of the world’s population without power. Nigeria’s grid delivers only four hours of reliable power daily, while 17 data centers require 137 MW to operate.

The picture of the data center is especially grim. Africa has roughly 360-409 MW of operational data center capacity - less than 1% of the global total for a continent with 1.4 billion people. McKinsey estimates that the top five markets will need to grow from 400 MW to 1.5-2.2 GW by 2030 just to meet demand. Only 5% of African AI innovators have reliable access to advanced computing. There are an estimated 7 million GPU hours of unmet demand for training models over the next three years.

Africa’s mobile-first approach—projected smartphone penetration of 88% by 2030—shapes AI deployment. In Nigeria, 86% of web traffic is mobile, illustrating the importance of on-device inference and compact models over cloud-based systems that require stable, low-cost connectivity.

What this looks like up close: Malawi

These numbers become real when you look at a single country. Malawi, where I am from, is landlocked. To get internet to Malawi, data has to travel through several other countries first. Each neighbor adds a transit fee, and every mile of cable is another opportunity for a break or a power failure. The result: some of the highest internet costs in the world and less than 1% of the population with a fixed broadband connection.

Brian Munyao Longwe, a technologist working on Africa’s internet infrastructure, recently published a blueprint for how he would spend $25 million to fix Malawi’s connectivity. His plan is not regarding flashy apps or free Wi-Fi in parks. It is about plumbing. The biggest item: $9 million to lay high-speed fiber along the Nacala railway corridor to Mozambique’s coast, giving Malawi a direct path to undersea cables instead of routing through expensive middlemen. Another $4.5 million for open-access fiber networks in cities - shared infrastructure that any ISP can use, driving competition and pushing prices down. A $4 million wholesale 4G LTE fixed wireless network to solve the last-mile problem, where a small entrepreneur could start an ISP and serve thousands of homes without needing millions to build towers. Solar panels and batteries at network hubs to remove the “diesel tax” imposed by unreliable grid power. And $2 million for local caching and Internet Exchange Points, so that when two people in Malawi email each other, the data does not have to travel to a server in Europe and back.

That last point sticks with me. It is the same absurdity as having to fly over Europe to reach another African country. When I raised this with Brian, he pointed out that the lowest-hanging fruit is last-mile access - get enough people meaningfully connected, and that pressure moves all the other bottlenecks. He is right. Our excessive dependence on mobile internet creates a strain on service delivery that compounds every other problem. Fixed broadband connections are what “meaningfully connected” actually looks like, and companies like Malawi Telecommunications Limited are not delivering it at scale. This is where converged networks and wholesale infrastructure models become essential.

Longwe’s blueprint totals $25 million, a negligible fraction of hyperscaler AI infrastructure spend. The basic gap starts at basic connectivity—something the rest of the world takes for granted.

Signs of progress

Progress is visible. Meta’s 2Africa submarine cable connects 33 countries and is projected to add billions to GDP. Google’s Equiano cable and the upcoming Umoja line expand continental links. Data center investment from firms like IFC and Equinix is increasing. However, these investments in cables and infrastructure are prerequisites, not the solution itself.

The policy fragmentation problem

Here is where Africa’s challenge begins to look fundamentally different from that of other regions. As of July 2025, only 16 of 55 African Union member states have launched national AI strategies. Kenya published its National AI Strategy 2025-2030. Egypt has the most developed strategy on the continent. But the majority of countries have nothing.

The data governance landscape is fragmented in ways that actively hinder AI development. While 44-46 countries now have data protection laws, the Malabo Convention - the AU’s framework for cybersecurity and data protection - took nine years to enter into force (adopted 2014, effective June 2023). South Africa, Nigeria, and Kenya, three of the continent’s four biggest tech markets, still have not ratified it. At least 10 countries have explicit data localization requirements, including Rwanda, which requires all personal data to be stored domestically unless authorities permit otherwise.

This matters because AI needs data to flow across borders. A Nairobi-based startup trying to build a model that works across East Africa faces a patchwork of incompatible regulations. The cost of compliance across multiple jurisdictions is prohibitive for small teams.

Compare this with what Europe just did. The EU Startup and Scaleup Strategy, adopted in May 2025, includes 26 measures and offers a “28th Regime” - an EU-wide corporate legal framework with 48-hour incorporation. Whatever its flaws, it represents the kind of continental coordination that Africa lacks. Tunisia passed a Startup Act in 2018 - the first in Africa - with 20+ policy measures. Cote d’Ivoire followed with regulatory sandboxes and tax incentives. But these are country-level efforts. There is no continent-wide equivalent.

For Africa to unlock the promise of digital trade, accelerating the ratification and alignment of the AfCFTA Digital Trade Protocol must become a continental priority. Governments should take coordinated action to bring the protocol into force swiftly and implement interoperable systems that foster cross-border digital growth.

No frontier models

Allow me to be blunt about where things stand. In 2024, the United States produced roughly 40 frontier AI models, China about 15, and the EU around 3. Africa produced zero. The scale of investment required to build frontier models is staggering - hyperscaler capital expenditure (Amazon, Google, Microsoft, Meta, Oracle) is projected to exceed $600 billion in 2026 alone. Africa’s entire tech VC ecosystem raised $4.1 billion in 2025 - a strong year, up 25% from 2024, but orders of magnitude below what frontier model development demands.

Africa will not build GPT-5. That is not the point. The question is whether Africa can build AI that solves African problems, in African languages, for African contexts. And on that front, real work is happening.

Lelapa AI in South Africa built InkubaLM, a 0.4-billion-parameter language model trained on 2.4 billion tokens across five African languages: Hausa, isiZulu, isiXhosa, Swahili, and Yoruba. It outperforms larger models regarding sentiment analysis tasks in these languages. Their commercial API, Vulavula, provides speech recognition, translation, and sentiment analysis for African languages. In Nigeria, researchers built N-ATLAS, an open-source multilingual LLM for Yoruba, Hausa, Igbo, and Nigerian-accented English.

The Masakhane community - a grassroots NLP collective of over 2,000 researchers - has produced more than 200 papers on African language AI. They maintain benchmarks, datasets, and models that would not exist if the work were left to big labs optimizing for English and Mandarin.

The startup ecosystem is growing. There are now 207 AI startups functioning in Africa, up from 104 in 2022. Finance (22), agriculture (20), and healthcare (20) are the leading sectors. And the exits are starting to matter: InstaDeep, founded in Tunisia, was acquired by BioNTech for $682 million - one of the largest tech acquisitions in African history.

The research and talent gap

Sub-Saharan Africa accounts for just 0.89% of global AI publications, according to the Stanford AI Index 2025. Only about 3% of global AI talent is based on the continent. The brain drain to the US, UK, Canada, and France is real and ongoing.

But the talent pipeline is building. The Deep Learning Indaba - Africa’s flagship machine learning conference - grew to 1,000 attendees in Kigali in 2025, with sponsors including Google DeepMind, Meta, Microsoft, and OpenAI. IndabaX satellite events now run in dozens of countries. The AfricaNLP workshop runs annually at top ML venues like ICLR and ACL.

Big tech is investing, though the scale is modest relative to their global spend. Google Research Africa operates labs in Accra and Nairobi, backed by a $37 million AI Community Center in Ghana. Microsoft’s Africa Development Center in Nairobi and Lagos represents a $100 million investment. OpenAI chose the University of Lagos as Africa’s first host for its Academy program. The AMMI program at AIMS, backed by Meta and Google, offers a fully funded Master’s in Machine Intelligence in Rwanda and Senegal.

Governments are starting to respond. Ghana approved a $250 million AI Center in April 2026. Ethiopia’s Prime Minister announced the establishment of Africa’s first AI-focused university at the February 2026 AU Summit. The UAE and Ghana signed a $1 billion agreement for an AI hub in January 2026.

These are encouraging signals. But 207 startups and a handful of research labs do not constitute an ecosystem capable of keeping pace with a global AI race where the leading players are spending hundreds of billions annually.

What needs to happen

The AU Continental AI Strategy, adopted in June 2024, lays out a five-pillar framework: harnessing benefits, building capabilities, limiting risks, encouraging cooperation, and stimulating investment. The AI 10 Billion Initiative, launched by the AfDB and UNDP at the 2026 Nairobi AI Forum, aims to raise $10 billion by 2035 to unlock 40 million jobs. The Africa Declaration on AI, endorsed by 49 countries at the April 2025 Global AI Summit on Africa in Rwanda, signals political will.

But declarations are not execution. Here is what actually needs to happen:

Continental coordination on data governance. The AfCFTA Digital Trade Protocol and Smart Africa’s Cross-Border Data Exchange Guidelines are steps in the right direction. But the Malabo Convention’s ratification history shows how slowly these things move. The major economies - South Africa, Nigeria, Kenya, Egypt - need to lead rather than lag.

Shared compute infrastructure. No individual African country can afford to build the computing capacity needed for serious AI research. A continental or regional shared compute facility - modeled on CERN or similar multilateral science infrastructure - would let researchers across the continent access GPUs without each country duplicating the investment. The World Economic Forum’s work on green computing in Africa points in this direction, but it needs political backing and funding.

Investment in African language AI. There are over 2,000 languages spoken across Africa. The work that Masakhane, Lelapa AI, and others are doing on language models is foundational. It needs sustained funding, not project-based grants that dry up after 18 months.

Power grid modernization. You cannot run data centers on four hours of grid power a day. AI infrastructure investment without energy infrastructure investment is building on sand. Africa’s renewable capacity is growing - 15.4 GW solar, 9.2 GW wind, 39.3 GW hydro as of 2024 - but grid infrastructure and storage investment lag far behind.

Talent retention. The brain drain will not stop through appeals to patriotism. It stops when researchers and engineers can do competitive work, earn reasonable compensation, and access computing resources without leaving the continent. Google and Microsoft’s research labs in Africa help, but they employ dozens, not thousands.

The window is narrowing.

Africa’s AI future depends on solving the same infrastructure and governance problems that hold back everything else on this continent. That is not a glamorous insight, but it is honest. There is no AI leapfrog without reliable electricity, affordable bandwidth, interoperable regulations, and trained people who choose to stay.

The building blocks are emerging: submarine cables landing on both coasts, data center investments flowing in, policy guidelines being drafted, and homegrown models proving that African-language AI is viable. The number of AI startups on the continent doubled in three years. VC funding is recovering. At least, political rhetoric now includes AI as a priority.

But the global AI race is accelerating faster than Africa is building its foundation. The US and China are competing to shape AI adoption on the continent - over 52 African countries have signed agreements with China totaling $700+ billion in engineering deals over the past decade. Africa risks becoming a consumer of AI systems designed elsewhere, for other contexts, in other languages, rather than a participant in developing the technology.

Back in Malawi, someone has already drawn up a $25 million plan to fix the country’s internet plumbing. It is not a trillion-dollar moonshot. It is fiber along railway lines, solar panels on towers, and shared infrastructure that enable small businesses to compete. That is the kind of concrete, unsexy work that actually moves the needle - and it is the kind of thinking that needs to scale across the continent.

The trillion-dollar question is whether African governments, institutions, and the private sector can coordinate at a continental scale, invest in the boring yet essential infrastructure, and create the conditions for a real AI ecosystem - not just a handful of success stories. The clock is ticking, and declarations alone will not build data centers or train models.