The builder - Five factories - Who holds each piece? - The face on the commercial track - What we cannot see - Where to next?
At present Africa cannot make advanced semiconductors, train globally competitive artificial-intelligence models, or raise the capital to do either. Engagement with the global firms that can do these things is not a choice the continent gets to make; it is a necessity. So the question is not whether African operators should work with the hyperscalers — Google, Amazon and Microsoft — but on what terms, and whether we are clear about where a sensible compromise ends and something more like surrender begins.
The starkest example of this dilemma sits with Strive Masiyiwa and Cassava Technologies.
The builder
Masiyiwa is, above all, a builder of things Africa was told it could not have. In 1993 he set out to create one of Africa’s first mobile networks in Zimbabwe. The state telephone monopoly was the only game in town and had no intention of sharing it. It declared him, in his own recollection, “public enemy number one”. He took the government to court, lost, and spent the next five years fighting. He won in the end, arguing that the monopoly violated the constitutional right to free expression in a country where, by his reckoning, seventy per cent of Africans in 1994 had never heard a telephone ring. The ensuing success of Econet Wireless, forerunner of Cassava Technologies, was born in 1998 from that judgment, and a conviction came with it that has never left him: that connecting Africans is an essential developmental goal and that the obstacles to it have to be overcome rather than accepted.
That conviction defines his whole career. It is why, when the African Union made him a Special Envoy for Covid Response in 2020 and asked him to find medical supplies and vaccines, he built a continental purchasing platform and went looking for doses. He had the money and was refused the goods. The year’s production was already committed, richer governments had banned their own manufacturers from exporting, and he found himself, as he put it, corralled into a corner while others ran off and secured the supplies. It was a hard lesson in self-sufficiency: not for the first or last time Africa had to battle for its place at the main table.
Five factories
Five years later, in 2025, through Cassava Technologies — the group that houses his Liquid fibre and cloud businesses, the Africa Data Centres estate and, most recently, Cassava AI— Masiyiwa committed up to US$720 million to build five AI factories across the continent. An AI factory is a data centre built on cutting-edge chips, GPUs, on which AI models are trained. The five sites, in South Africa, Nigeria, Kenya, Egypt and Morocco, will be kitted with 12,000 NVIDIA GPU’s acquired by Cassava. Clients will have access to a range of cloud platforms and AIaaS (AI as a Service) and will be able to
process AI workloads and data within national borders, strengthening data sovereignty, compliance, and trust.
The case for the factories is impressive. Today only about five per cent of Africa’s AI developers can access the computing power needed to train a model. Masiyiwa wants that to rise to fifty per cent by the end of 2027. Cassava also talks, credibly, about models trained in African languages, compute priced in local currency rather than dollars, and the shorter delay that comes from keeping the work on the continent instead of routing every query through Virginia.
Without the infrastructure, without the compute, Africa is a continent of users. We will use other people’s systems, but our own young people will not launch businesses in this space.
The alternative to Cassava is not an African AI industry that springs up indigenously overnight on its own. It is Google, Amazon and Microsoft building directly, with no African operator in the room. Partial control exercised through an African-owned company is plainly better than none. And the money to build at this scale, at this speed, simply does not exist on the continent, so Masiyiwa raised it where it could be raised. The business case looks sound. But we still need to take a closer look.
Who holds each piece?
An AI factory is a stack of layers. At the bottom are the buildings and power; above them the chips; above those the cloud software that turns racks of chips into usable computing; above that the AI models themselves; and at the very top the applications that ordinary users and businesses build. Sovereignty, if the word is to mean anything, is a question asked of each layer in turn: who owns it, who could withhold it and who can replace it?
The buildings are Cassava’s, and so is the power and the land. That is real, and it is not nothing. But it is the bottom of the stack.
The chips are NVIDIA’s. Every GPU in every factory comes from one company, and since November 2025 that company is also a shareholder in Cassava. A supplier and investor wrapped up in one. NVIDIA’s investment is very probably the reason the factories will be built at all, so this is as much an enabler as a dependency; but it means the layer on which everything rests answers to a single foreign firm.
The cloud is American — and which American matters less than it looks. Cassava’s connectivity arm sells private, direct links into Microsoft Azure, Google Cloud and Amazon Web Services, and presents the spread as a virtue: eliminate the hardware, the marketing runs, and focus on growth. Liquid has been reselling Azure and running its direct-connect service since 2017 and it remains the revenue spine. Google is the newer, AI-led, equity-backed partner. The software layer that turns racks of chips into usable computing can be Microsoft’s or Google’s or Amazon’s, and in every case it answers to a company headquartered in the United States. Multi-cloud here does not mean independence from any one vendor; it means dependence on a set of vendors that share a single jurisdiction.
The models sit in the same trap. Cassava’s multi-model platform offers Google, Anthropic and OpenAI. Anthropic’s models reach the continent through Google’s Vertex AI rather than directly. The consumer channel, data-free Gemini access pushed across Cassava’s network from November 2025 is wholly Google’s product, with Cassava as the distribution pipe. Whichever model an African business reaches for, it is renting from an American firm.
One more dependency cuts across all of these. The work of assembling the parts into something a large client can use, the integration where technical know-how actually transfers, is contracted to Accenture, a US consulting giant. The layer where Africa might have built its own capacity to combine these technologies is itself rented.
Which leaves the very top layer, the applications. This is what is actually being offered to Africans: the right to build apps on top of a stack that is, from the chips upward, owned and operated by American firms. That is a genuine opportunity — real businesses and real jobs are built at the app layer, and it is where the “creator not user” claim has some purchase.
But it is neither technological nor economic sovereignty, and the distinction matters more the longer the arrangement runs. A continent that builds on rented compute, rented cloud and rented models is a continent whose digital economy can be repriced, throttled or reshaped by decisions taken in California or Seattle. And, most importantly, whose own capacity to build the lower layers atrophies for lack of any reason to develop it.
Masiyiwa has never hidden the design.
About 15 years ago, we started to invest in data centres, so we’re not new to the data centre business. We started to put in what they call hyperscale facilities to support the likes of Google, Microsoft, AWS, these big tech companies, as they came to put their services on the African continent.
And then there is the law. Every claim made for the factories is a claim about where data sits — within Africa’s borders, resident in a country, hosted at home. None is a claim about who can lawfully reach it. Data held in a Johannesburg data centre, on American chips, in an American-run cloud, owned by a group incorporated through Mauritius, Jersey and London, is physically African and legally reachable from abroad. Most off-site backups are located abroad. And the US CLOUD Act lets American authorities compel American-headquartered firms to produce data wherever in the world it is stored. Nowhere in the public record does Cassava acknowledge the existence of this law.
So the two halves of the sovereignty question fail together. The economic half: nothing below the app layer is African, and the dependency deepens with every year of build-out. The legal half: even the data that does sit on African soil sits within reach of a foreign jurisdiction. A “sovereign cloud” that answers neither has answered only the easiest question of all — the postcode of the building.
The face on the commercial track
None of this is accidental. Google is pursuing continental scale along two tracks at once, as we set out in our earlier piece on Google in Africa. One is commercial: the Equiano and Umoja subsea cables, the Johannesburg cloud region, the enterprise and government customers, the consumer push behind Gemini. The other is institutional: framework agreements with the African Union Commission, the UN Economic Commission for Africa, the AfCFTA Secretariat and Smart Africa, and seats on the very bodies now drafting the continent’s AI and data rules. The first track sells the products; the second helps shape the rules the products will be sold under.
Cassava is the credible African face of the commercial track. That is what the partnership is about: a continental infrastructure group with an authentically African story lends Google a legitimacy that Google could never achieve on its own. When the data-free Gemini app reaches a student in Nairobi over Cassava’s network, what arrives is a Google product wearing an African coat. The arrangement is presented as Africa partnering with Google; it functions, at least as much, as Google acquiring an African front for a market it intends to dominate. That is not a charge against Masiyiwa’s motives, which may be exactly as developmental as he says. It is an observation about what the structure does regardless of motive.
And Google is not the only firm the structure serves. Cassava’s older and, for now, larger commercial relationship is with Microsoft — the Azure resale and direct-connect business it has run since 2017 — and the identical “African digital independence” language was deployed around Microsoft in 2023 before it was ever attached to Google. This is the part that should give the reader pause. The sovereignty wrapper names no hyperscaler, and so it fits them all: Microsoft as the revenue spine, Google as the AI-and-equity future, Amazon available alongside both, NVIDIA beneath. Each dependence is denied by pointing at the others — not captured by Google, because Microsoft is there too; not locked into either, because the platform is “multi-cloud”. But multi-cloud, here, means a choice among American clouds, every one of them reachable by American law. The sovereignty is the wrapper; the contents are interchangeable US hyperscalers.
It would be wrong to read this as proof that African digital sovereignty is hopeless. We have shown elsewhere that African-controlled infrastructure exists, works and crosses borders: of 154 significant facilities, two-thirds are under African control, and a dozen multi-country African-owned operators are already built. Measured against that field, the uncomfortable truth is that Cassava — the loudest voice for African digital sovereignty — is among the more entangled operators, not the cleanest. WIOCC, owned by ten African telecoms companies, and Angola Cables, owned outright by the Angolan state, sit closer to the principle. And the same sponsor-mediated pattern shapes MTN’s relationship with Microsoft and Safaricom’s with Meta: arrangements described as continental keep resting on a few personal relationships between a founder and a Californian executive — a thin foundation for anything called sovereign.
What we cannot see
A private company has the right to keep its terms private, and Cassava is private. But when the assets are of continental weight and partly built with public development money — from the US development-finance agency, from Britain’s development financier, from European funds — the consequence of that privacy is that the public cannot see the terms on which its own digital future is being set. On the central arrangements the record is closed:
- the 2023 Liquid C2–Google partnership — its exclusivity, length and exit terms;
- Google’s and NVIDIA’s exact stakes in Cassava, and any board or veto rights attached to them;
- the commercial settlement behind the Gemini deal — who pays whom, and for what;
- the Cassava–NVIDIA terms;
- the National Sovereign Cloud’s contracts — and, as yet, the name of a single government customer;
- the Umoja overland contract.
None of this is evidence of wrongdoing. It is simply absence that inevitably raises questions. A dividing line no one outside the deal can see is one no one can hold to account.
Where to next?
So where does the dividing line fall? Not at a point on a dial, but at a set of conditions, and they can be named. A compromise stays a compromise while the partner could in principle be replaced, while Africa’s capacity to build the lower layers grows rather than withers, while legal control sits where the data sits, and while the terms are open to those whose future they shape.
Where those hold, working with a hyperscaler is good sense. Where they fail, the same arrangement quietly becomes dependency wearing the trappings of independence.
On the evidence here, the Cassava stack meets almost none of them. The chips, the cloud and the models can be swapped — but only for each other: Azure for Google Cloud for AWS, one American vendor for the next, never out of the American stack altogether. The only layer left wholly to Africa is the one on top; the data beneath it is reachable from abroad whichever vendor is chosen; and the terms are sealed. Selling to the hyperscalers is a revenue line. Depending on their jurisdiction — and on their chips, their cloud and their models — is a sovereignty failure.
Compromises have to be made. The right ones are never going to be easy. But these are urgent challenges that countries, regions and the continent need to prepare for.