I spend a good part of my year working with African governments on digital projects, and there is a version of the same question waiting in every meeting room. Sometimes it’s asked directly, sometimes it hides inside a procurement decision, but it’s always there: is it already too late for us?
It’s a fair question. It’s also the wrong one - and the way it’s wrong took me years of sitting in those rooms to articulate.
The Reality Check
Let’s not romanticize anything. The gaps are real, and they are structural.
Around 600 million Africans still live without access to electricity - roughly 43% of the continent, the largest unelectrified population in the world. You cannot run an AI revolution, or even a reliable school network, on power that isn’t there.
Connectivity tells the same story. In 2024, about 38% of Africans used the internet, against 68% globally. And for those who can connect, it costs more of their income than anywhere else: a basic 2GB mobile plan eats over 4% of average monthly income - more than double the UN’s 2% affordability target.
Every AI conversation on the continent happens on top of those two numbers. Ignore them and everything else is theater.
The Extraction Pattern
Here’s what makes the gaps dangerous rather than just painful: while the infrastructure lags, the data doesn’t. It flows out.
Personal information, cultural patterns, behavioral data - collected here, used to train models elsewhere, monetized somewhere else entirely. What flows back are systems trained mostly on wealthy-country data, carrying wealthy-country assumptions: the loan model that has no concept of your economic reality, the health system that wasn’t trained on people who look like you.
History students will recognize the shape. Railways were once laid to move resources out, not to connect the people living along the tracks. Some of today’s infrastructure follows the same logic with better branding.
And dependency has sharp edges. In February 2025, Reuters reported that the US threatened to cut Ukraine’s Starlink access during minerals negotiations - a claim Elon Musk denied, but one that made every government relying on foreign-controlled connectivity sit up. The technical layer is no more comforting: Oxford researchers showed that Starlink terminals could be remotely disabled and forced into stow mode - a vulnerability since patched, but a reminder of where the off switch lives. Starlink is now licensed in more than 20 African countries, and while its footprint is still small, the trajectory is plain: critical connectivity, controlled from elsewhere.
This is the part the “left behind” framing misses entirely. Being left behind is not a position on a timeline. It’s a position in an architecture.
The question was never whether Africa adopts AI. Adoption is guaranteed. The question is whether Africa owns any layer of the stack it adopts.
What Ownership Already Looks Like
That question has answers - and they predate the current hype cycle.
InstaDeep, founded in Tunis with reportedly two laptops and $2,000, was acquired by BioNTech in a deal worth up to $682 million - Africa’s biggest startup exit, built on world-class AI research. Lelapa AI in South Africa builds Vulavula, language AI for isiZulu, Xhosa, Sesotho and Afrikaans - languages the big labs treat as afterthoughts. Kenya’s Jacaranda Health built UlizaLlama, the first open-source Swahili LLM, to power maternal-health support. In Nigeria, Awarri is building Yoruba, Hausa and Igbo voice AI and the crowdsourced datasets those languages never had. Zindi has grown a community of tens of thousands of African data scientists across the continent, solving local problems in open competition. Zenvus puts soil sensors and analytics in the hands of Nigerian farmers. mPharma uses predictive inventory to keep pharmacy shelves stocked and medicine prices down across nine countries.
The energy story - the supposed disqualifier - is quietly becoming an asset. Morocco built Noor Ouarzazate, one of the world’s largest concentrated-solar complexes, and targets 52% renewable capacity by 2030. Microsoft and G42 are building a $1 billion geothermal-powered data center in Olkaria, Kenya - AI compute running on volcanic steam. Ethiopia earned around $140 million exporting hydroelectric power in a single year. Compute follows cheap clean energy, and Africa has more untapped renewable potential than any other continent.
The connectivity dependency is being hedged too - not solved, hedged. Eutelsat’s Konnect satellites already serve over 200,000 people in sub-Saharan Africa through solar-powered WiFi hotspots. Vodafone and Amazon’s Project Kuiper plan satellite backhaul to extend 4G and 5G deep into the continent. In rural South Africa, community-owned networks like Zenzeleni run solar-powered mesh internet as local cooperatives. Nigeria’s data protection regime now requires sensitive data - including health data - to be stored in-country. Through Smart Africa, governments are coordinating data-center and cloud policy continent-wide instead of negotiating alone.
And last month, the clearest signal yet: at the Global AI Summit on Africa in Kigali, African states endorsed the Africa Declaration on Artificial Intelligence and a proposed $60 billion Africa AI Fund. In the same season, Cassava Technologies announced NVIDIA-powered “AI factories” - GPU clusters starting in South Africa and expanding to Egypt, Kenya, Morocco and Nigeria, built so African data can be processed on African soil.
None of these are charity. They are positions being taken in the architecture.
The Honest Reckoning
Now the part the conference keynotes skip.
Declarations are not deployments. The continent has a long history of summit documents that outran implementation, and the $60 billion fund is, for now, a pledge - the gap between announcing capital and writing checks has buried better ideas.
The funding asymmetry is brutal. The sums celebrated as continental milestones are rounding errors against what a single US lab raises in a quarter. Owning a layer of the stack at that disadvantage requires choosing layers carefully - languages, energy, distribution - not matching the leaders dollar for dollar.
The talent math cuts both ways. The strongest African AI engineers are courted globally, and every Kigali declaration competes with a relocation package from London or San Francisco. Ecosystems like Zindi’s exist precisely to give that talent local problems worth staying for - but the pull is real and constant.
And the electricity gap will not close as a side effect of AI enthusiasm. Six hundred million people without power is a generation-scale infrastructure project. The honest version of the energy story is potential, mostly untapped - not capacity, already delivering.
The Question in the Meeting Room
So when the question comes - is it already too late? - here’s what I’ve started answering.
Too late for what? To build a frontier lab that outspends the giants - probably, yes. To own the layers that decide whether the next billion users are customers of their own future or data sources for someone else’s - no. That contest is barely underway, and the entries that matter look like Swahili LLMs, geothermal data centers, community networks, and data laws with teeth.
The lights staying on is not a metaphor here. It’s the actual work.
And if I had to bet on who finds a way to build under constraints that would stop everyone else, I’d still put my money on the continent that has been doing exactly that for centuries. Not because optimism is pleasant - because I watch it happen, one meeting room at a time.