![\"\"](\"https:\/\/villpress.com\/wp-content\/uploads\/2026\/07\/image-8.jpg\")
<\/figure>\n\n\n\nAlmost all of the coverage that followed asked the same question, how do four cables fail at once, and who is to blame? It was the wrong question. The useful one is the opposite: what kept working?<\/p>\n\n\n\n
What stayed up is the more interesting question<\/strong><\/h2>\n\n\n\nInternet exchange points remained operational across the affected countries, according to the Internet Society\u2019s post-mortem. An exchange point is an unglamorous room in which local networks plug directly into one another instead of routing through Europe. Where one existed, a Nigerian bank\u2019s traffic to a Nigerian customer never needed the Atlantic; it stayed in Lagos. Niger\u2019s exchange was not operational at the time.<\/p>\n\n\n\n
Second, cross-border terrestrial fibre carried traffic overland to countries whose sea links were still up, the regional grid doing what regional grids are for.<\/p>\n\n\n\n
Third, and most tellingly, the countries whose most-visited websites are hosted or cached on their own soil were markedly less disrupted than their neighbours. South Africa and Kenya rode out the week in a way Liberia and Benin could not, not because they had more cables, but because more of what their citizens wanted was already inside the country.<\/p>\n\n\n\n
Put those three together and a definition falls out, an operational one rather than an ideological one. Digital sovereignty is not what you own. It is what still works when the link to the outside is cut.<\/strong> Call it the cut test. It is a diagnostic, not a doctrine, and its virtue is that it is answerable: point it at any layer of the stack and ask what happens when the outside stops cooperating. The answers, layer by layer, tell a stranger and more useful story about Africa than the familiar one about being behind.<\/p>\n\n\n\nThe one layer Africa already governs<\/strong><\/h2>\n\n\n\nIn 2025, some $1.4 trillion moved through mobile money accounts in sub-Saharan Africa, about two-thirds of the entire world\u2019s mobile money value, on the GSMA\u2019s numbers. The region holds more than half of all registered accounts on earth. By the GSMA\u2019s modelling, mobile money was adding around $190 billion a year to sub-Saharan African GDP as of 2023. <\/p>\n\n\n\n![\"Africa](\"https:\/\/villpress.com\/wp-content\/uploads\/2026\/07\/image-9.jpg\")
<\/figure>\n\n\n\nCut the cable to Europe and a Kenyan sending money to Kisumu is inconvenienced rather than stranded: the transaction clears on domestic rails, under a domestic licence, at a price set by a domestic regulator, settled in domestic currency. Nigerian bank apps and USSD services did stumble in March 202, but that was a connectivity failure upstream of the payment system, not a failure of who controls it.<\/p>\n\n\n\n
Here is the part worth sitting with. None of that sovereignty came from owning the technology. The handsets are foreign. The towers are foreign. The silicon is foreign. Safaricom did not build a semiconductor industry before launching M-Pesa. What Kenya\u2019s regulators did instead was let a telco hold customer float, and then insist the resulting rails stay licensed, switched and settled at home. Sovereignty was won at the layer where the rules are written and the customer is touched, not at the layer where the equipment is manufactured.<\/p>\n\n\n\n
The pattern is now being extended deliberately. The Pan-African Payment and Settlement System, launched by Afreximbank and the African Union in January 2022, exists so that a Nigerian importer can pay a Kenyan supplier in naira and shillings without the payment detouring through a correspondent bank in London or New York. The Bank of Central African States joined in July 2026. Progress is real but slower than the announcements imply: by early 2025, fifteen central banks had signed up while live commercial banks existed in only twelve countries. A continental switch is easy to launch and hard to fill.<\/p>\n\n\n\n
Even so, the direction is unambiguous. Payments remain the only layer on the continent where the honest answer to the cut test is: mostly fine.<\/p>\n\n\n\n
\u00a0Why the data went abroad, and why the law could not bring it back<\/strong><\/h2>\n\n\n\nMore than forty African countries have passed data protection or localization frameworks over the past decade. Institutions including the IFC and the GSMA nonetheless estimate that upwards of 80 percent of African data is hosted outside Africa. The gap between the statute and the server is the most important number in this story, and it has almost nothing to do with political will.<\/p>\n\n\n\n![\"Africa](\"https:\/\/villpress.com\/wp-content\/uploads\/2026\/07\/image-10.jpg\")
<\/figure>\n\n\n\nMalawi\u2019s number, though, carries a twist. A further 62 percent of its content is served regionally, overwhelmingly out of South Africa. Tanzania shows the same shape: 10 percent local, 44 percent regional. The countries with almost no domestic hosting are not simply dependent on Frankfurt and Virginia. They are increasingly dependent on Johannesburg. What looks like failure at the national scale is, at the continental scale, the early outline of a solution.<\/p>\n\n\n\n
The economics explain the rest. Africa had roughly 223 data centres across 38 countries by mid-2025, with South Africa, Kenya and Nigeria accounting for about 41 percent of them. Capacity here is expensive \u2014 on industry estimates around $10 million per megawatt, and closer to $15 million in Nigeria, where operators report grid supply of about four hours a day and diesel makes up the difference. Nigeria\u2019s seventeen facilities together draw roughly 137 megawatts. A law can require that data stay in the country. It cannot conjure the megawatt the data would sit on.<\/p>\n\n\n\n
Estimates of what the continent actually has diverge, and it is worth saying which to trust. The Africa Data Centres Association put operational capacity at 307 MW in 2025; McKinsey works from about 0.4 GW; ITWeb, citing industry figures, puts Africa at 0.6 percent of global capacity. One commercial research house reports a 2025 figure roughly three times higher, apparently on a broader definition of IT load. The lower cluster is consistent across independent sources, and it is the one used here.<\/p>\n\n\n\n
The kill switch is a contract<\/strong><\/h2>\n\n\n\nIn March 2024 there was an actual physical interruption, but the more relevant threat is contractual control. After a U.S. executive order in May 2025, the International Criminal Court\u2019s chief prosecutor lost access to a Microsoft-hosted email account and moved to a Swiss provider. Microsoft disputes that it suspended service to the court and in February 2026 asked a UK parliamentary committee to correct its earlier testimony; the facts remain contested. The practical consequence was already clear: by October 2025 the ICC began migrating to OpenDesk, an open-source platform supported by Germany\u2019s state centre for digital sovereignty. Gartner expects European sovereign-cloud spending to more than triple between 2025 and 2027.No cable was cut in The Hague. An account was.<\/p>\n\n\n\n
A data centre you own, running software you licence, under a legal system you do not vote in, is a warehouse,\u00a0not a sovereign.<\/em><\/pre>\n\n\n\nStating where a diagnostic fails is what makes it usable, so state it plainly: the cut test misses every dependency that is legal or commercial rather than physical. Under section 2713 of the US CLOUD Act, American providers can be compelled to produce data in their possession, custody or control regardless of which country it is stored in. Localisation answers the question \u201cwhere does the disk sit?\u201d It does not answer \u201cwho can be ordered to unlock it?\u201d<\/p>\n\n\n\n
The same logic runs through the cables. Google\u2019s Equiano, at 144 Tbps, is wholly owned by Google. Meta\u2019s 2Africa, at 180 Tbps and completed in November 2025, encircles the continent across 45,000 kilometres and 33 countries \u2014 built by a consortium in which Meta sits alongside MTN\u2019s Bayobab, Orange, Vodafone, Telecom Egypt, stc\u2019s center3 and WIOCC. Researchers writing in The Information Society note that each of these two systems dwarfs the combined capacity of Africa\u2019s other cables by roughly an order of magnitude. The Atlantic Council\u2019s Digital Forensic Research Lab has found that Google, Meta, Microsoft and Amazon between them own or lease around 70 percent of global undersea bandwidth.<\/p>\n\n\n\n
The result is genuinely double-edged, and pretending otherwise would be dishonest. West Africa went dark in 2024 partly because it was leaning on a handful of ageing systems; the new cables are precisely the redundancy the region lacked, and African operators hold real equity in one of them. But the redundancy arrived courtesy of the two companies whose own services generate much of the traffic. Africa got a better road, built by the trucking company.<\/p>\n\n\n\n
Compute is failing the test faster than anything else<\/strong><\/h2>\n\n\n\nWhich brings the argument to the layer where the answer is worst and the clock runs quickest.<\/p>\n\n\n\n
Oxford researchers mapping the world\u2019s publicly available AI compute found it in just 32 countries, with the United States and China operating the overwhelming majority of specialized AI data centres between them. Most of the world sits in what the authors call compute deserts. Africa\u2019s presence on that map is in the single digits.<\/p>\n\n\n\n
The scarcity is felt at the level of an individual researcher. An analysis of Zindi\u2019s network of 11,000 African data scientists found that only about five percent had access to the computing power their work required, and that the first one percent were simply those with GPUs physically in their offices. The GSMA notes that a single GPU costs the equivalent of 75 percent of GDP per capita in Kenya and 69 percent in Senegal. The absolute price is much the same everywhere. That is exactly the problem.<\/p>\n\n\n\n
The most serious African answer is Cassava Technologies, which announced in March 2025 that it would build the continent\u2019s first AI factory in South Africa with Nvidia, targeting deployment by that June. By November 2025 it had acquired 12,000 Nvidia GPUs and launched GPU-as-a-service; in March 2026 it announced deployment in South Africa with Nigeria, Kenya, Egypt and Morocco to follow; by May 2026 the Cape Town facility was preparing to go live and a second was planned for Johannesburg. Its stated ambition is 12,000 to 13,000 GPUs across the entire continent.<\/p>\n\n\n\n
For scale: a single site in Memphis, Tennessee was reported in January 2026 at around 555,000 GPUs. Published figures for that cluster vary by source; the order of magnitude does not. Cassava\u2019s continental target is roughly two percent of one American building.<\/p>\n\n\n\n
That comparison is not a verdict on Cassava, which is doing something difficult and necessary that nobody else on the continent is attempting at scale. It isolates the variable that actually matters: on current trajectories Africa will not out-build the frontier, and any sovereignty strategy premised on doing so is a strategy for permanent disappointment.<\/p>\n\n\n\n
\u00a0Fifty-five sovereignties, one uneconomic market<\/strong><\/h2>\n\n\n\nIf every country insists that its citizens\u2019 data live within its own borders, Africa gets fifty-odd subscale data centres, each too small to be efficient, each sitting on a grid that cannot reliably power it, each raising the cost of serving its own market. Sovereignty pursued at the national grain produces less resilience, not more \u2014 because the smallest facilities are the ones most likely to fail the cut test in the least dramatic way: by being switched off during load-shedding.<\/p>\n\n\n\n
The African Union has, to its credit, seen this coming. Its Data Policy Framework, the Continental AI Strategy adopted in July 2024 and the AfCFTA Digital Trade Protocol ratified in February 2024 all lean the same way \u2014 harmonised rules, conditional rather than absolute localisation, and free movement of data within the continent. The strategy\u2019s own ambition is for AI to add $1.5 trillion to African economies by 2030. That is a projection, and a stretching one, but the direction it sets is unmistakable.<\/p>\n\n\n\n
Implementation is the problem. The Malabo Convention, the AU\u2019s binding instrument on cybersecurity and personal data, was adopted in 2014 and took until June 2023 to enter into force. On the AU\u2019s own July 2024 tally, 21 member states had signed and 16 had ratified, out of 55. South Africa is not among them; nor, on that list, is Egypt. By a separate count, 22 of 54 countries had a national AI strategy by 2026, up from 16 in 2024.<\/p>\n\n\n\n
The Malawi finding points at the way out. A country serving 3 percent of its content locally and 62 percent regionally has not failed; it has pooled. Regional hosting is what national sovereignty looks like once it is made affordable. The unit that can actually pass the cut test in Africa is not the country. It is the region.<\/p>\n\n\n\n
The constraint has moved to the power station<\/strong><\/h2>\n\n\n\nOn 24 July 2026, the City of Cape Town approved two hyperscale data centres that will together draw around 174 megawatts. That single approval is close to half of the continent\u2019s entire installed data centre capacity today. McKinsey expects Africa to move from roughly 0.4 GW now to between 1.5 and 2.2 GW by 2030, requiring $10\u201320 billion of investment.<\/p>\n\n\n\n
The evidence keeps pointing the same way. Microsoft\u2019s $1 billion Kenyan project with G42 is built around geothermal power, with 100 MW planned. Nigeria\u2019s Kasi Cloud commissioned a carrier-neutral campus in Lekki in June 2026, designed to scale to about 100 MW. Raxio reported that its contracted power grew sixfold in the first half of 2026 against the same period a year earlier. In each case the pitch to capital is not bandwidth. It is firm power.<\/p>\n\n\n\n
And the constraint is severe. Around 600 million Africans still live without electricity \u2014 a figure that shifts with methodology, from about 565 million in the Tracking SDG7 baseline to roughly 590 million in the IEA\u2019s most recent investment report, but never in a direction that changes the argument. Per capita electricity use in sub-Saharan Africa has been broadly flat for three decades. Meanwhile a large data centre can consume up to 19 million litres of water a day for cooling, on figures put to South African councillors \u2014 which is why, when eThekwini approved an AI data centre agreement in Durban in March 2026, a senior opposition councillor withheld support pending transparency on its energy and water demand.<\/p>\n\n\n\n
That councillor was doing something more sophisticated than obstruction. He was asking the sovereignty question in its correct, current form: not whose flag is on the building, but who bears the cost of keeping it running, and whose supply gets shorter when it does.<\/p>\n\n\n\n
The unglamorous agenda<\/strong><\/h2>\n\n\n\nFor founders. <\/strong>The sovereign layer is the one Africa has always won, rails, rules and the last mile. Payment infrastructure, identity, clearing, logistics, health records. Compute is a rented input, and the firms that survive the next decade will be the ones that treat it as such while owning the relationship with the customer and the licence that governs it.<\/p>\n\n\n\nFor investors. <\/strong>The scarce asset in African digital infrastructure is no longer fibre; it is firm, dispatchable, contractable power, and the transmission and storage to move it. A hyperscale tenant is the best offtaker a private power project has ever been offered on this continent: predictable, round-the-clock, multi-decade demand. Cape Town\u2019s 174 megawatts is the demand signal. The generation to meet it is the trade.<\/p>\n\n\n\nFor ministers. <\/strong>Localisation mandates without regional harmonisation are a tax on your own economy. The higher-return moves are cheaper and far less announceable: fund the internet exchange, subsidise local caching and CDN presence, ratify the continental instruments so data can move within Africa as freely as it already moves to Frankfurt, and buy substitutability. Open standards and exit-ready contracts are how you stop a foreign compliance department from becoming your policy.<\/p>\n\n\n\nThe rockslide off Abidjan did the continent an unintended favour. It ran, for free, the experiment no government would ever have commissioned: it switched off the outside world for a few weeks and let everybody see which parts of their national life were actually theirs. Payments held. Exchanges held. Locally hosted content held. Everything routed through somewhere else went dark.<\/p>\n\n\n\n
The next cut will not be geological, and it will not announce itself with an outage graph. It will arrive as a sanctions list, a licence revocation, an export control on a chip, a clause invoked by a compliance officer eight thousand kilometres away. Africa cannot stop that switch from existing. Nobody can.<\/p>\n\n\n\n
What it can still decide, layer by layer and within this decade, is how much of ordinary life is wired through it.<\/p>\n\n\n\n
A note on sources and method<\/strong>
Outage figures are NetBlocks network measurements as reported by the BBC and Reuters on 14 March 2024, with restoration dates from cable operators and the Internet Society\u2019s April 2024 outage report. Mobile money data is the GSMA\u2019s State of the Industry Report on Mobile Money 2026 (covering 2025). Content-locality measurements are preliminary results from the Internet Society Pulse research fellowship, published June 2026. Compute geography draws on Lehdonvirta, W\u00fa and Hawkins, \u2018Compute North vs. Compute South\u2019 (AIES 2024) and subsequent Oxford Internet Institute mapping reported in June 2025; African compute-access figures are from Alliance4ai and Zindi via UNDP, and the GSMA.
Where credible sources disagree, on Liberia\u2019s outage depth, on Africa\u2019s installed data centre capacity, on the number of people without electricity, and on the Microsoft\u2013ICC episode, the disagreement is stated in the text and the figure used is identified. Forward-looking figures from McKinsey and the African Union are projections and are marked as such. No figure in this feature has been estimated or interpolated by Villpress.<\/strong><\/pre>","protected":false},"excerpt":{"rendered":"On 14 March 2024, a stretch of seabed off C\u00f4te d\u2019Ivoire gave way. Four submarine cables,\u00a0WACS, SAT-3, ACE and MainOne,\u00a0failed within hours of one another. There was no attacker, no ransom note, nobody to negotiate with. A rock had moved. By that afternoon C\u00f4te d\u2019Ivoire\u2019s internet was running at roughly four percent of its ordinary […]<\/p>\n","protected":false},"author":1,"featured_media":12287,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"_acf_changed":false,"_mi_skip_tracking":false,"footnotes":""},"categories":[2109],"tags":[2247],"ppma_author":[331],"class_list":["post-12279","post","type-post","status-publish","format-standard","has-post-thumbnail","category-villpress-intelligence","tag-infastructure"],"acf":[],"authors":[{"term_id":331,"user_id":1,"is_guest":0,"slug":"pastakutmanwen","display_name":"Staff Writer","avatar_url":{"url":"https:\/\/villpress.com\/wp-content\/uploads\/2025\/05\/Logo.png","url2x":"https:\/\/villpress.com\/wp-content\/uploads\/2025\/05\/Logo.png"},"author_category":"1","first_name":"Staff","last_name":"Writer","user_url":"http:\/\/villpress.com","job_title":"Staffs At Villpress","description":"The Villpress Staff Writers are an in-house team of experienced editors and industry experts dedicated to producing clear, insightful content. As part of Villpress, they cover the latest trends and innovations across business, technology, artificial intelligence, advertising, and more, delivering stories that inform, engage, and add real value to readers."}],"_links":{"self":[{"href":"https:\/\/villpress.com\/de\/wp-json\/wp\/v2\/posts\/12279","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/villpress.com\/de\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/villpress.com\/de\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/villpress.com\/de\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/villpress.com\/de\/wp-json\/wp\/v2\/comments?post=12279"}],"version-history":[{"count":2,"href":"https:\/\/villpress.com\/de\/wp-json\/wp\/v2\/posts\/12279\/revisions"}],"predecessor-version":[{"id":12288,"href":"https:\/\/villpress.com\/de\/wp-json\/wp\/v2\/posts\/12279\/revisions\/12288"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/villpress.com\/de\/wp-json\/wp\/v2\/media\/12287"}],"wp:attachment":[{"href":"https:\/\/villpress.com\/de\/wp-json\/wp\/v2\/media?parent=12279"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/villpress.com\/de\/wp-json\/wp\/v2\/categories?post=12279"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/villpress.com\/de\/wp-json\/wp\/v2\/tags?post=12279"},{"taxonomy":"author","embeddable":true,"href":"https:\/\/villpress.com\/de\/wp-json\/wp\/v2\/ppma_author?post=12279"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}
The one layer Africa already governs<\/strong><\/h2>\n\n\n\nIn 2025, some $1.4 trillion moved through mobile money accounts in sub-Saharan Africa, about two-thirds of the entire world\u2019s mobile money value, on the GSMA\u2019s numbers. The region holds more than half of all registered accounts on earth. By the GSMA\u2019s modelling, mobile money was adding around $190 billion a year to sub-Saharan African GDP as of 2023. <\/p>\n\n\n\n<\/figure>\n\n\n\nCut the cable to Europe and a Kenyan sending money to Kisumu is inconvenienced rather than stranded: the transaction clears on domestic rails, under a domestic licence, at a price set by a domestic regulator, settled in domestic currency. Nigerian bank apps and USSD services did stumble in March 202, but that was a connectivity failure upstream of the payment system, not a failure of who controls it.<\/p>\n\n\n\n
Here is the part worth sitting with. None of that sovereignty came from owning the technology. The handsets are foreign. The towers are foreign. The silicon is foreign. Safaricom did not build a semiconductor industry before launching M-Pesa. What Kenya\u2019s regulators did instead was let a telco hold customer float, and then insist the resulting rails stay licensed, switched and settled at home. Sovereignty was won at the layer where the rules are written and the customer is touched, not at the layer where the equipment is manufactured.<\/p>\n\n\n\n
The pattern is now being extended deliberately. The Pan-African Payment and Settlement System, launched by Afreximbank and the African Union in January 2022, exists so that a Nigerian importer can pay a Kenyan supplier in naira and shillings without the payment detouring through a correspondent bank in London or New York. The Bank of Central African States joined in July 2026. Progress is real but slower than the announcements imply: by early 2025, fifteen central banks had signed up while live commercial banks existed in only twelve countries. A continental switch is easy to launch and hard to fill.<\/p>\n\n\n\n
Even so, the direction is unambiguous. Payments remain the only layer on the continent where the honest answer to the cut test is: mostly fine.<\/p>\n\n\n\n
\u00a0Why the data went abroad, and why the law could not bring it back<\/strong><\/h2>\n\n\n\nMore than forty African countries have passed data protection or localization frameworks over the past decade. Institutions including the IFC and the GSMA nonetheless estimate that upwards of 80 percent of African data is hosted outside Africa. The gap between the statute and the server is the most important number in this story, and it has almost nothing to do with political will.<\/p>\n\n\n\n<\/figure>\n\n\n\nMalawi\u2019s number, though, carries a twist. A further 62 percent of its content is served regionally, overwhelmingly out of South Africa. Tanzania shows the same shape: 10 percent local, 44 percent regional. The countries with almost no domestic hosting are not simply dependent on Frankfurt and Virginia. They are increasingly dependent on Johannesburg. What looks like failure at the national scale is, at the continental scale, the early outline of a solution.<\/p>\n\n\n\n
The economics explain the rest. Africa had roughly 223 data centres across 38 countries by mid-2025, with South Africa, Kenya and Nigeria accounting for about 41 percent of them. Capacity here is expensive \u2014 on industry estimates around $10 million per megawatt, and closer to $15 million in Nigeria, where operators report grid supply of about four hours a day and diesel makes up the difference. Nigeria\u2019s seventeen facilities together draw roughly 137 megawatts. A law can require that data stay in the country. It cannot conjure the megawatt the data would sit on.<\/p>\n\n\n\n
Estimates of what the continent actually has diverge, and it is worth saying which to trust. The Africa Data Centres Association put operational capacity at 307 MW in 2025; McKinsey works from about 0.4 GW; ITWeb, citing industry figures, puts Africa at 0.6 percent of global capacity. One commercial research house reports a 2025 figure roughly three times higher, apparently on a broader definition of IT load. The lower cluster is consistent across independent sources, and it is the one used here.<\/p>\n\n\n\n
The kill switch is a contract<\/strong><\/h2>\n\n\n\nIn March 2024 there was an actual physical interruption, but the more relevant threat is contractual control. After a U.S. executive order in May 2025, the International Criminal Court\u2019s chief prosecutor lost access to a Microsoft-hosted email account and moved to a Swiss provider. Microsoft disputes that it suspended service to the court and in February 2026 asked a UK parliamentary committee to correct its earlier testimony; the facts remain contested. The practical consequence was already clear: by October 2025 the ICC began migrating to OpenDesk, an open-source platform supported by Germany\u2019s state centre for digital sovereignty. Gartner expects European sovereign-cloud spending to more than triple between 2025 and 2027.No cable was cut in The Hague. An account was.<\/p>\n\n\n\n
A data centre you own, running software you licence, under a legal system you do not vote in, is a warehouse,\u00a0not a sovereign.<\/em><\/pre>\n\n\n\nStating where a diagnostic fails is what makes it usable, so state it plainly: the cut test misses every dependency that is legal or commercial rather than physical. Under section 2713 of the US CLOUD Act, American providers can be compelled to produce data in their possession, custody or control regardless of which country it is stored in. Localisation answers the question \u201cwhere does the disk sit?\u201d It does not answer \u201cwho can be ordered to unlock it?\u201d<\/p>\n\n\n\n
The same logic runs through the cables. Google\u2019s Equiano, at 144 Tbps, is wholly owned by Google. Meta\u2019s 2Africa, at 180 Tbps and completed in November 2025, encircles the continent across 45,000 kilometres and 33 countries \u2014 built by a consortium in which Meta sits alongside MTN\u2019s Bayobab, Orange, Vodafone, Telecom Egypt, stc\u2019s center3 and WIOCC. Researchers writing in The Information Society note that each of these two systems dwarfs the combined capacity of Africa\u2019s other cables by roughly an order of magnitude. The Atlantic Council\u2019s Digital Forensic Research Lab has found that Google, Meta, Microsoft and Amazon between them own or lease around 70 percent of global undersea bandwidth.<\/p>\n\n\n\n
The result is genuinely double-edged, and pretending otherwise would be dishonest. West Africa went dark in 2024 partly because it was leaning on a handful of ageing systems; the new cables are precisely the redundancy the region lacked, and African operators hold real equity in one of them. But the redundancy arrived courtesy of the two companies whose own services generate much of the traffic. Africa got a better road, built by the trucking company.<\/p>\n\n\n\n
Compute is failing the test faster than anything else<\/strong><\/h2>\n\n\n\nWhich brings the argument to the layer where the answer is worst and the clock runs quickest.<\/p>\n\n\n\n
Oxford researchers mapping the world\u2019s publicly available AI compute found it in just 32 countries, with the United States and China operating the overwhelming majority of specialized AI data centres between them. Most of the world sits in what the authors call compute deserts. Africa\u2019s presence on that map is in the single digits.<\/p>\n\n\n\n
The scarcity is felt at the level of an individual researcher. An analysis of Zindi\u2019s network of 11,000 African data scientists found that only about five percent had access to the computing power their work required, and that the first one percent were simply those with GPUs physically in their offices. The GSMA notes that a single GPU costs the equivalent of 75 percent of GDP per capita in Kenya and 69 percent in Senegal. The absolute price is much the same everywhere. That is exactly the problem.<\/p>\n\n\n\n
The most serious African answer is Cassava Technologies, which announced in March 2025 that it would build the continent\u2019s first AI factory in South Africa with Nvidia, targeting deployment by that June. By November 2025 it had acquired 12,000 Nvidia GPUs and launched GPU-as-a-service; in March 2026 it announced deployment in South Africa with Nigeria, Kenya, Egypt and Morocco to follow; by May 2026 the Cape Town facility was preparing to go live and a second was planned for Johannesburg. Its stated ambition is 12,000 to 13,000 GPUs across the entire continent.<\/p>\n\n\n\n
For scale: a single site in Memphis, Tennessee was reported in January 2026 at around 555,000 GPUs. Published figures for that cluster vary by source; the order of magnitude does not. Cassava\u2019s continental target is roughly two percent of one American building.<\/p>\n\n\n\n
That comparison is not a verdict on Cassava, which is doing something difficult and necessary that nobody else on the continent is attempting at scale. It isolates the variable that actually matters: on current trajectories Africa will not out-build the frontier, and any sovereignty strategy premised on doing so is a strategy for permanent disappointment.<\/p>\n\n\n\n
\u00a0Fifty-five sovereignties, one uneconomic market<\/strong><\/h2>\n\n\n\nIf every country insists that its citizens\u2019 data live within its own borders, Africa gets fifty-odd subscale data centres, each too small to be efficient, each sitting on a grid that cannot reliably power it, each raising the cost of serving its own market. Sovereignty pursued at the national grain produces less resilience, not more \u2014 because the smallest facilities are the ones most likely to fail the cut test in the least dramatic way: by being switched off during load-shedding.<\/p>\n\n\n\n
The African Union has, to its credit, seen this coming. Its Data Policy Framework, the Continental AI Strategy adopted in July 2024 and the AfCFTA Digital Trade Protocol ratified in February 2024 all lean the same way \u2014 harmonised rules, conditional rather than absolute localisation, and free movement of data within the continent. The strategy\u2019s own ambition is for AI to add $1.5 trillion to African economies by 2030. That is a projection, and a stretching one, but the direction it sets is unmistakable.<\/p>\n\n\n\n
Implementation is the problem. The Malabo Convention, the AU\u2019s binding instrument on cybersecurity and personal data, was adopted in 2014 and took until June 2023 to enter into force. On the AU\u2019s own July 2024 tally, 21 member states had signed and 16 had ratified, out of 55. South Africa is not among them; nor, on that list, is Egypt. By a separate count, 22 of 54 countries had a national AI strategy by 2026, up from 16 in 2024.<\/p>\n\n\n\n
The Malawi finding points at the way out. A country serving 3 percent of its content locally and 62 percent regionally has not failed; it has pooled. Regional hosting is what national sovereignty looks like once it is made affordable. The unit that can actually pass the cut test in Africa is not the country. It is the region.<\/p>\n\n\n\n
The constraint has moved to the power station<\/strong><\/h2>\n\n\n\nOn 24 July 2026, the City of Cape Town approved two hyperscale data centres that will together draw around 174 megawatts. That single approval is close to half of the continent\u2019s entire installed data centre capacity today. McKinsey expects Africa to move from roughly 0.4 GW now to between 1.5 and 2.2 GW by 2030, requiring $10\u201320 billion of investment.<\/p>\n\n\n\n
The evidence keeps pointing the same way. Microsoft\u2019s $1 billion Kenyan project with G42 is built around geothermal power, with 100 MW planned. Nigeria\u2019s Kasi Cloud commissioned a carrier-neutral campus in Lekki in June 2026, designed to scale to about 100 MW. Raxio reported that its contracted power grew sixfold in the first half of 2026 against the same period a year earlier. In each case the pitch to capital is not bandwidth. It is firm power.<\/p>\n\n\n\n
And the constraint is severe. Around 600 million Africans still live without electricity \u2014 a figure that shifts with methodology, from about 565 million in the Tracking SDG7 baseline to roughly 590 million in the IEA\u2019s most recent investment report, but never in a direction that changes the argument. Per capita electricity use in sub-Saharan Africa has been broadly flat for three decades. Meanwhile a large data centre can consume up to 19 million litres of water a day for cooling, on figures put to South African councillors \u2014 which is why, when eThekwini approved an AI data centre agreement in Durban in March 2026, a senior opposition councillor withheld support pending transparency on its energy and water demand.<\/p>\n\n\n\n
That councillor was doing something more sophisticated than obstruction. He was asking the sovereignty question in its correct, current form: not whose flag is on the building, but who bears the cost of keeping it running, and whose supply gets shorter when it does.<\/p>\n\n\n\n
The unglamorous agenda<\/strong><\/h2>\n\n\n\nFor founders. <\/strong>The sovereign layer is the one Africa has always won, rails, rules and the last mile. Payment infrastructure, identity, clearing, logistics, health records. Compute is a rented input, and the firms that survive the next decade will be the ones that treat it as such while owning the relationship with the customer and the licence that governs it.<\/p>\n\n\n\nFor investors. <\/strong>The scarce asset in African digital infrastructure is no longer fibre; it is firm, dispatchable, contractable power, and the transmission and storage to move it. A hyperscale tenant is the best offtaker a private power project has ever been offered on this continent: predictable, round-the-clock, multi-decade demand. Cape Town\u2019s 174 megawatts is the demand signal. The generation to meet it is the trade.<\/p>\n\n\n\nFor ministers. <\/strong>Localisation mandates without regional harmonisation are a tax on your own economy. The higher-return moves are cheaper and far less announceable: fund the internet exchange, subsidise local caching and CDN presence, ratify the continental instruments so data can move within Africa as freely as it already moves to Frankfurt, and buy substitutability. Open standards and exit-ready contracts are how you stop a foreign compliance department from becoming your policy.<\/p>\n\n\n\nThe rockslide off Abidjan did the continent an unintended favour. It ran, for free, the experiment no government would ever have commissioned: it switched off the outside world for a few weeks and let everybody see which parts of their national life were actually theirs. Payments held. Exchanges held. Locally hosted content held. Everything routed through somewhere else went dark.<\/p>\n\n\n\n
The next cut will not be geological, and it will not announce itself with an outage graph. It will arrive as a sanctions list, a licence revocation, an export control on a chip, a clause invoked by a compliance officer eight thousand kilometres away. Africa cannot stop that switch from existing. Nobody can.<\/p>\n\n\n\n
What it can still decide, layer by layer and within this decade, is how much of ordinary life is wired through it.<\/p>\n\n\n\n
A note on sources and method<\/strong>
Outage figures are NetBlocks network measurements as reported by the BBC and Reuters on 14 March 2024, with restoration dates from cable operators and the Internet Society\u2019s April 2024 outage report. Mobile money data is the GSMA\u2019s State of the Industry Report on Mobile Money 2026 (covering 2025). Content-locality measurements are preliminary results from the Internet Society Pulse research fellowship, published June 2026. Compute geography draws on Lehdonvirta, W\u00fa and Hawkins, \u2018Compute North vs. Compute South\u2019 (AIES 2024) and subsequent Oxford Internet Institute mapping reported in June 2025; African compute-access figures are from Alliance4ai and Zindi via UNDP, and the GSMA.
Where credible sources disagree, on Liberia\u2019s outage depth, on Africa\u2019s installed data centre capacity, on the number of people without electricity, and on the Microsoft\u2013ICC episode, the disagreement is stated in the text and the figure used is identified. Forward-looking figures from McKinsey and the African Union are projections and are marked as such. No figure in this feature has been estimated or interpolated by Villpress.<\/strong><\/pre>","protected":false},"excerpt":{"rendered":"On 14 March 2024, a stretch of seabed off C\u00f4te d\u2019Ivoire gave way. Four submarine cables,\u00a0WACS, SAT-3, ACE and MainOne,\u00a0failed within hours of one another. There was no attacker, no ransom note, nobody to negotiate with. A rock had moved. By that afternoon C\u00f4te d\u2019Ivoire\u2019s internet was running at roughly four percent of its ordinary […]<\/p>\n","protected":false},"author":1,"featured_media":12287,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"_acf_changed":false,"_mi_skip_tracking":false,"footnotes":""},"categories":[2109],"tags":[2247],"ppma_author":[331],"class_list":["post-12279","post","type-post","status-publish","format-standard","has-post-thumbnail","category-villpress-intelligence","tag-infastructure"],"acf":[],"authors":[{"term_id":331,"user_id":1,"is_guest":0,"slug":"pastakutmanwen","display_name":"Staff Writer","avatar_url":{"url":"https:\/\/villpress.com\/wp-content\/uploads\/2025\/05\/Logo.png","url2x":"https:\/\/villpress.com\/wp-content\/uploads\/2025\/05\/Logo.png"},"author_category":"1","first_name":"Staff","last_name":"Writer","user_url":"http:\/\/villpress.com","job_title":"Staffs At Villpress","description":"The Villpress Staff Writers are an in-house team of experienced editors and industry experts dedicated to producing clear, insightful content. As part of Villpress, they cover the latest trends and innovations across business, technology, artificial intelligence, advertising, and more, delivering stories that inform, engage, and add real value to readers."}],"_links":{"self":[{"href":"https:\/\/villpress.com\/de\/wp-json\/wp\/v2\/posts\/12279","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/villpress.com\/de\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/villpress.com\/de\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/villpress.com\/de\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/villpress.com\/de\/wp-json\/wp\/v2\/comments?post=12279"}],"version-history":[{"count":2,"href":"https:\/\/villpress.com\/de\/wp-json\/wp\/v2\/posts\/12279\/revisions"}],"predecessor-version":[{"id":12288,"href":"https:\/\/villpress.com\/de\/wp-json\/wp\/v2\/posts\/12279\/revisions\/12288"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/villpress.com\/de\/wp-json\/wp\/v2\/media\/12287"}],"wp:attachment":[{"href":"https:\/\/villpress.com\/de\/wp-json\/wp\/v2\/media?parent=12279"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/villpress.com\/de\/wp-json\/wp\/v2\/categories?post=12279"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/villpress.com\/de\/wp-json\/wp\/v2\/tags?post=12279"},{"taxonomy":"author","embeddable":true,"href":"https:\/\/villpress.com\/de\/wp-json\/wp\/v2\/ppma_author?post=12279"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}
<\/figure>\n\n\n\nCut the cable to Europe and a Kenyan sending money to Kisumu is inconvenienced rather than stranded: the transaction clears on domestic rails, under a domestic licence, at a price set by a domestic regulator, settled in domestic currency. Nigerian bank apps and USSD services did stumble in March 202, but that was a connectivity failure upstream of the payment system, not a failure of who controls it.<\/p>\n\n\n\n
Here is the part worth sitting with. None of that sovereignty came from owning the technology. The handsets are foreign. The towers are foreign. The silicon is foreign. Safaricom did not build a semiconductor industry before launching M-Pesa. What Kenya\u2019s regulators did instead was let a telco hold customer float, and then insist the resulting rails stay licensed, switched and settled at home. Sovereignty was won at the layer where the rules are written and the customer is touched, not at the layer where the equipment is manufactured.<\/p>\n\n\n\n
The pattern is now being extended deliberately. The Pan-African Payment and Settlement System, launched by Afreximbank and the African Union in January 2022, exists so that a Nigerian importer can pay a Kenyan supplier in naira and shillings without the payment detouring through a correspondent bank in London or New York. The Bank of Central African States joined in July 2026. Progress is real but slower than the announcements imply: by early 2025, fifteen central banks had signed up while live commercial banks existed in only twelve countries. A continental switch is easy to launch and hard to fill.<\/p>\n\n\n\n
Even so, the direction is unambiguous. Payments remain the only layer on the continent where the honest answer to the cut test is: mostly fine.<\/p>\n\n\n\n
\u00a0Why the data went abroad, and why the law could not bring it back<\/strong><\/h2>\n\n\n\nMore than forty African countries have passed data protection or localization frameworks over the past decade. Institutions including the IFC and the GSMA nonetheless estimate that upwards of 80 percent of African data is hosted outside Africa. The gap between the statute and the server is the most important number in this story, and it has almost nothing to do with political will.<\/p>\n\n\n\n<\/figure>\n\n\n\nMalawi\u2019s number, though, carries a twist. A further 62 percent of its content is served regionally, overwhelmingly out of South Africa. Tanzania shows the same shape: 10 percent local, 44 percent regional. The countries with almost no domestic hosting are not simply dependent on Frankfurt and Virginia. They are increasingly dependent on Johannesburg. What looks like failure at the national scale is, at the continental scale, the early outline of a solution.<\/p>\n\n\n\n
The economics explain the rest. Africa had roughly 223 data centres across 38 countries by mid-2025, with South Africa, Kenya and Nigeria accounting for about 41 percent of them. Capacity here is expensive \u2014 on industry estimates around $10 million per megawatt, and closer to $15 million in Nigeria, where operators report grid supply of about four hours a day and diesel makes up the difference. Nigeria\u2019s seventeen facilities together draw roughly 137 megawatts. A law can require that data stay in the country. It cannot conjure the megawatt the data would sit on.<\/p>\n\n\n\n
Estimates of what the continent actually has diverge, and it is worth saying which to trust. The Africa Data Centres Association put operational capacity at 307 MW in 2025; McKinsey works from about 0.4 GW; ITWeb, citing industry figures, puts Africa at 0.6 percent of global capacity. One commercial research house reports a 2025 figure roughly three times higher, apparently on a broader definition of IT load. The lower cluster is consistent across independent sources, and it is the one used here.<\/p>\n\n\n\n
The kill switch is a contract<\/strong><\/h2>\n\n\n\nIn March 2024 there was an actual physical interruption, but the more relevant threat is contractual control. After a U.S. executive order in May 2025, the International Criminal Court\u2019s chief prosecutor lost access to a Microsoft-hosted email account and moved to a Swiss provider. Microsoft disputes that it suspended service to the court and in February 2026 asked a UK parliamentary committee to correct its earlier testimony; the facts remain contested. The practical consequence was already clear: by October 2025 the ICC began migrating to OpenDesk, an open-source platform supported by Germany\u2019s state centre for digital sovereignty. Gartner expects European sovereign-cloud spending to more than triple between 2025 and 2027.No cable was cut in The Hague. An account was.<\/p>\n\n\n\n
A data centre you own, running software you licence, under a legal system you do not vote in, is a warehouse,\u00a0not a sovereign.<\/em><\/pre>\n\n\n\nStating where a diagnostic fails is what makes it usable, so state it plainly: the cut test misses every dependency that is legal or commercial rather than physical. Under section 2713 of the US CLOUD Act, American providers can be compelled to produce data in their possession, custody or control regardless of which country it is stored in. Localisation answers the question \u201cwhere does the disk sit?\u201d It does not answer \u201cwho can be ordered to unlock it?\u201d<\/p>\n\n\n\n
The same logic runs through the cables. Google\u2019s Equiano, at 144 Tbps, is wholly owned by Google. Meta\u2019s 2Africa, at 180 Tbps and completed in November 2025, encircles the continent across 45,000 kilometres and 33 countries \u2014 built by a consortium in which Meta sits alongside MTN\u2019s Bayobab, Orange, Vodafone, Telecom Egypt, stc\u2019s center3 and WIOCC. Researchers writing in The Information Society note that each of these two systems dwarfs the combined capacity of Africa\u2019s other cables by roughly an order of magnitude. The Atlantic Council\u2019s Digital Forensic Research Lab has found that Google, Meta, Microsoft and Amazon between them own or lease around 70 percent of global undersea bandwidth.<\/p>\n\n\n\n
The result is genuinely double-edged, and pretending otherwise would be dishonest. West Africa went dark in 2024 partly because it was leaning on a handful of ageing systems; the new cables are precisely the redundancy the region lacked, and African operators hold real equity in one of them. But the redundancy arrived courtesy of the two companies whose own services generate much of the traffic. Africa got a better road, built by the trucking company.<\/p>\n\n\n\n
Compute is failing the test faster than anything else<\/strong><\/h2>\n\n\n\nWhich brings the argument to the layer where the answer is worst and the clock runs quickest.<\/p>\n\n\n\n
Oxford researchers mapping the world\u2019s publicly available AI compute found it in just 32 countries, with the United States and China operating the overwhelming majority of specialized AI data centres between them. Most of the world sits in what the authors call compute deserts. Africa\u2019s presence on that map is in the single digits.<\/p>\n\n\n\n
The scarcity is felt at the level of an individual researcher. An analysis of Zindi\u2019s network of 11,000 African data scientists found that only about five percent had access to the computing power their work required, and that the first one percent were simply those with GPUs physically in their offices. The GSMA notes that a single GPU costs the equivalent of 75 percent of GDP per capita in Kenya and 69 percent in Senegal. The absolute price is much the same everywhere. That is exactly the problem.<\/p>\n\n\n\n
The most serious African answer is Cassava Technologies, which announced in March 2025 that it would build the continent\u2019s first AI factory in South Africa with Nvidia, targeting deployment by that June. By November 2025 it had acquired 12,000 Nvidia GPUs and launched GPU-as-a-service; in March 2026 it announced deployment in South Africa with Nigeria, Kenya, Egypt and Morocco to follow; by May 2026 the Cape Town facility was preparing to go live and a second was planned for Johannesburg. Its stated ambition is 12,000 to 13,000 GPUs across the entire continent.<\/p>\n\n\n\n
For scale: a single site in Memphis, Tennessee was reported in January 2026 at around 555,000 GPUs. Published figures for that cluster vary by source; the order of magnitude does not. Cassava\u2019s continental target is roughly two percent of one American building.<\/p>\n\n\n\n
That comparison is not a verdict on Cassava, which is doing something difficult and necessary that nobody else on the continent is attempting at scale. It isolates the variable that actually matters: on current trajectories Africa will not out-build the frontier, and any sovereignty strategy premised on doing so is a strategy for permanent disappointment.<\/p>\n\n\n\n
\u00a0Fifty-five sovereignties, one uneconomic market<\/strong><\/h2>\n\n\n\nIf every country insists that its citizens\u2019 data live within its own borders, Africa gets fifty-odd subscale data centres, each too small to be efficient, each sitting on a grid that cannot reliably power it, each raising the cost of serving its own market. Sovereignty pursued at the national grain produces less resilience, not more \u2014 because the smallest facilities are the ones most likely to fail the cut test in the least dramatic way: by being switched off during load-shedding.<\/p>\n\n\n\n
The African Union has, to its credit, seen this coming. Its Data Policy Framework, the Continental AI Strategy adopted in July 2024 and the AfCFTA Digital Trade Protocol ratified in February 2024 all lean the same way \u2014 harmonised rules, conditional rather than absolute localisation, and free movement of data within the continent. The strategy\u2019s own ambition is for AI to add $1.5 trillion to African economies by 2030. That is a projection, and a stretching one, but the direction it sets is unmistakable.<\/p>\n\n\n\n
Implementation is the problem. The Malabo Convention, the AU\u2019s binding instrument on cybersecurity and personal data, was adopted in 2014 and took until June 2023 to enter into force. On the AU\u2019s own July 2024 tally, 21 member states had signed and 16 had ratified, out of 55. South Africa is not among them; nor, on that list, is Egypt. By a separate count, 22 of 54 countries had a national AI strategy by 2026, up from 16 in 2024.<\/p>\n\n\n\n
The Malawi finding points at the way out. A country serving 3 percent of its content locally and 62 percent regionally has not failed; it has pooled. Regional hosting is what national sovereignty looks like once it is made affordable. The unit that can actually pass the cut test in Africa is not the country. It is the region.<\/p>\n\n\n\n
The constraint has moved to the power station<\/strong><\/h2>\n\n\n\nOn 24 July 2026, the City of Cape Town approved two hyperscale data centres that will together draw around 174 megawatts. That single approval is close to half of the continent\u2019s entire installed data centre capacity today. McKinsey expects Africa to move from roughly 0.4 GW now to between 1.5 and 2.2 GW by 2030, requiring $10\u201320 billion of investment.<\/p>\n\n\n\n
The evidence keeps pointing the same way. Microsoft\u2019s $1 billion Kenyan project with G42 is built around geothermal power, with 100 MW planned. Nigeria\u2019s Kasi Cloud commissioned a carrier-neutral campus in Lekki in June 2026, designed to scale to about 100 MW. Raxio reported that its contracted power grew sixfold in the first half of 2026 against the same period a year earlier. In each case the pitch to capital is not bandwidth. It is firm power.<\/p>\n\n\n\n
And the constraint is severe. Around 600 million Africans still live without electricity \u2014 a figure that shifts with methodology, from about 565 million in the Tracking SDG7 baseline to roughly 590 million in the IEA\u2019s most recent investment report, but never in a direction that changes the argument. Per capita electricity use in sub-Saharan Africa has been broadly flat for three decades. Meanwhile a large data centre can consume up to 19 million litres of water a day for cooling, on figures put to South African councillors \u2014 which is why, when eThekwini approved an AI data centre agreement in Durban in March 2026, a senior opposition councillor withheld support pending transparency on its energy and water demand.<\/p>\n\n\n\n
That councillor was doing something more sophisticated than obstruction. He was asking the sovereignty question in its correct, current form: not whose flag is on the building, but who bears the cost of keeping it running, and whose supply gets shorter when it does.<\/p>\n\n\n\n
The unglamorous agenda<\/strong><\/h2>\n\n\n\nFor founders. <\/strong>The sovereign layer is the one Africa has always won, rails, rules and the last mile. Payment infrastructure, identity, clearing, logistics, health records. Compute is a rented input, and the firms that survive the next decade will be the ones that treat it as such while owning the relationship with the customer and the licence that governs it.<\/p>\n\n\n\nFor investors. <\/strong>The scarce asset in African digital infrastructure is no longer fibre; it is firm, dispatchable, contractable power, and the transmission and storage to move it. A hyperscale tenant is the best offtaker a private power project has ever been offered on this continent: predictable, round-the-clock, multi-decade demand. Cape Town\u2019s 174 megawatts is the demand signal. The generation to meet it is the trade.<\/p>\n\n\n\nFor ministers. <\/strong>Localisation mandates without regional harmonisation are a tax on your own economy. The higher-return moves are cheaper and far less announceable: fund the internet exchange, subsidise local caching and CDN presence, ratify the continental instruments so data can move within Africa as freely as it already moves to Frankfurt, and buy substitutability. Open standards and exit-ready contracts are how you stop a foreign compliance department from becoming your policy.<\/p>\n\n\n\nThe rockslide off Abidjan did the continent an unintended favour. It ran, for free, the experiment no government would ever have commissioned: it switched off the outside world for a few weeks and let everybody see which parts of their national life were actually theirs. Payments held. Exchanges held. Locally hosted content held. Everything routed through somewhere else went dark.<\/p>\n\n\n\n
The next cut will not be geological, and it will not announce itself with an outage graph. It will arrive as a sanctions list, a licence revocation, an export control on a chip, a clause invoked by a compliance officer eight thousand kilometres away. Africa cannot stop that switch from existing. Nobody can.<\/p>\n\n\n\n
What it can still decide, layer by layer and within this decade, is how much of ordinary life is wired through it.<\/p>\n\n\n\n
A note on sources and method<\/strong>
Outage figures are NetBlocks network measurements as reported by the BBC and Reuters on 14 March 2024, with restoration dates from cable operators and the Internet Society\u2019s April 2024 outage report. Mobile money data is the GSMA\u2019s State of the Industry Report on Mobile Money 2026 (covering 2025). Content-locality measurements are preliminary results from the Internet Society Pulse research fellowship, published June 2026. Compute geography draws on Lehdonvirta, W\u00fa and Hawkins, \u2018Compute North vs. Compute South\u2019 (AIES 2024) and subsequent Oxford Internet Institute mapping reported in June 2025; African compute-access figures are from Alliance4ai and Zindi via UNDP, and the GSMA.
Where credible sources disagree, on Liberia\u2019s outage depth, on Africa\u2019s installed data centre capacity, on the number of people without electricity, and on the Microsoft\u2013ICC episode, the disagreement is stated in the text and the figure used is identified. Forward-looking figures from McKinsey and the African Union are projections and are marked as such. No figure in this feature has been estimated or interpolated by Villpress.<\/strong><\/pre>","protected":false},"excerpt":{"rendered":"On 14 March 2024, a stretch of seabed off C\u00f4te d\u2019Ivoire gave way. Four submarine cables,\u00a0WACS, SAT-3, ACE and MainOne,\u00a0failed within hours of one another. There was no attacker, no ransom note, nobody to negotiate with. A rock had moved. By that afternoon C\u00f4te d\u2019Ivoire\u2019s internet was running at roughly four percent of its ordinary […]<\/p>\n","protected":false},"author":1,"featured_media":12287,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"_acf_changed":false,"_mi_skip_tracking":false,"footnotes":""},"categories":[2109],"tags":[2247],"ppma_author":[331],"class_list":["post-12279","post","type-post","status-publish","format-standard","has-post-thumbnail","category-villpress-intelligence","tag-infastructure"],"acf":[],"authors":[{"term_id":331,"user_id":1,"is_guest":0,"slug":"pastakutmanwen","display_name":"Staff Writer","avatar_url":{"url":"https:\/\/villpress.com\/wp-content\/uploads\/2025\/05\/Logo.png","url2x":"https:\/\/villpress.com\/wp-content\/uploads\/2025\/05\/Logo.png"},"author_category":"1","first_name":"Staff","last_name":"Writer","user_url":"http:\/\/villpress.com","job_title":"Staffs At Villpress","description":"The Villpress Staff Writers are an in-house team of experienced editors and industry experts dedicated to producing clear, insightful content. As part of Villpress, they cover the latest trends and innovations across business, technology, artificial intelligence, advertising, and more, delivering stories that inform, engage, and add real value to readers."}],"_links":{"self":[{"href":"https:\/\/villpress.com\/de\/wp-json\/wp\/v2\/posts\/12279","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/villpress.com\/de\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/villpress.com\/de\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/villpress.com\/de\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/villpress.com\/de\/wp-json\/wp\/v2\/comments?post=12279"}],"version-history":[{"count":2,"href":"https:\/\/villpress.com\/de\/wp-json\/wp\/v2\/posts\/12279\/revisions"}],"predecessor-version":[{"id":12288,"href":"https:\/\/villpress.com\/de\/wp-json\/wp\/v2\/posts\/12279\/revisions\/12288"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/villpress.com\/de\/wp-json\/wp\/v2\/media\/12287"}],"wp:attachment":[{"href":"https:\/\/villpress.com\/de\/wp-json\/wp\/v2\/media?parent=12279"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/villpress.com\/de\/wp-json\/wp\/v2\/categories?post=12279"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/villpress.com\/de\/wp-json\/wp\/v2\/tags?post=12279"},{"taxonomy":"author","embeddable":true,"href":"https:\/\/villpress.com\/de\/wp-json\/wp\/v2\/ppma_author?post=12279"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}
<\/figure>\n\n\n\nMalawi\u2019s number, though, carries a twist. A further 62 percent of its content is served regionally, overwhelmingly out of South Africa. Tanzania shows the same shape: 10 percent local, 44 percent regional. The countries with almost no domestic hosting are not simply dependent on Frankfurt and Virginia. They are increasingly dependent on Johannesburg. What looks like failure at the national scale is, at the continental scale, the early outline of a solution.<\/p>\n\n\n\n
The economics explain the rest. Africa had roughly 223 data centres across 38 countries by mid-2025, with South Africa, Kenya and Nigeria accounting for about 41 percent of them. Capacity here is expensive \u2014 on industry estimates around $10 million per megawatt, and closer to $15 million in Nigeria, where operators report grid supply of about four hours a day and diesel makes up the difference. Nigeria\u2019s seventeen facilities together draw roughly 137 megawatts. A law can require that data stay in the country. It cannot conjure the megawatt the data would sit on.<\/p>\n\n\n\n
Estimates of what the continent actually has diverge, and it is worth saying which to trust. The Africa Data Centres Association put operational capacity at 307 MW in 2025; McKinsey works from about 0.4 GW; ITWeb, citing industry figures, puts Africa at 0.6 percent of global capacity. One commercial research house reports a 2025 figure roughly three times higher, apparently on a broader definition of IT load. The lower cluster is consistent across independent sources, and it is the one used here.<\/p>\n\n\n\n
The kill switch is a contract<\/strong><\/h2>\n\n\n\nIn March 2024 there was an actual physical interruption, but the more relevant threat is contractual control. After a U.S. executive order in May 2025, the International Criminal Court\u2019s chief prosecutor lost access to a Microsoft-hosted email account and moved to a Swiss provider. Microsoft disputes that it suspended service to the court and in February 2026 asked a UK parliamentary committee to correct its earlier testimony; the facts remain contested. The practical consequence was already clear: by October 2025 the ICC began migrating to OpenDesk, an open-source platform supported by Germany\u2019s state centre for digital sovereignty. Gartner expects European sovereign-cloud spending to more than triple between 2025 and 2027.No cable was cut in The Hague. An account was.<\/p>\n\n\n\n
A data centre you own, running software you licence, under a legal system you do not vote in, is a warehouse,\u00a0not a sovereign.<\/em><\/pre>\n\n\n\nStating where a diagnostic fails is what makes it usable, so state it plainly: the cut test misses every dependency that is legal or commercial rather than physical. Under section 2713 of the US CLOUD Act, American providers can be compelled to produce data in their possession, custody or control regardless of which country it is stored in. Localisation answers the question \u201cwhere does the disk sit?\u201d It does not answer \u201cwho can be ordered to unlock it?\u201d<\/p>\n\n\n\n
The same logic runs through the cables. Google\u2019s Equiano, at 144 Tbps, is wholly owned by Google. Meta\u2019s 2Africa, at 180 Tbps and completed in November 2025, encircles the continent across 45,000 kilometres and 33 countries \u2014 built by a consortium in which Meta sits alongside MTN\u2019s Bayobab, Orange, Vodafone, Telecom Egypt, stc\u2019s center3 and WIOCC. Researchers writing in The Information Society note that each of these two systems dwarfs the combined capacity of Africa\u2019s other cables by roughly an order of magnitude. The Atlantic Council\u2019s Digital Forensic Research Lab has found that Google, Meta, Microsoft and Amazon between them own or lease around 70 percent of global undersea bandwidth.<\/p>\n\n\n\n
The result is genuinely double-edged, and pretending otherwise would be dishonest. West Africa went dark in 2024 partly because it was leaning on a handful of ageing systems; the new cables are precisely the redundancy the region lacked, and African operators hold real equity in one of them. But the redundancy arrived courtesy of the two companies whose own services generate much of the traffic. Africa got a better road, built by the trucking company.<\/p>\n\n\n\n
Compute is failing the test faster than anything else<\/strong><\/h2>\n\n\n\nWhich brings the argument to the layer where the answer is worst and the clock runs quickest.<\/p>\n\n\n\n
Oxford researchers mapping the world\u2019s publicly available AI compute found it in just 32 countries, with the United States and China operating the overwhelming majority of specialized AI data centres between them. Most of the world sits in what the authors call compute deserts. Africa\u2019s presence on that map is in the single digits.<\/p>\n\n\n\n
The scarcity is felt at the level of an individual researcher. An analysis of Zindi\u2019s network of 11,000 African data scientists found that only about five percent had access to the computing power their work required, and that the first one percent were simply those with GPUs physically in their offices. The GSMA notes that a single GPU costs the equivalent of 75 percent of GDP per capita in Kenya and 69 percent in Senegal. The absolute price is much the same everywhere. That is exactly the problem.<\/p>\n\n\n\n
The most serious African answer is Cassava Technologies, which announced in March 2025 that it would build the continent\u2019s first AI factory in South Africa with Nvidia, targeting deployment by that June. By November 2025 it had acquired 12,000 Nvidia GPUs and launched GPU-as-a-service; in March 2026 it announced deployment in South Africa with Nigeria, Kenya, Egypt and Morocco to follow; by May 2026 the Cape Town facility was preparing to go live and a second was planned for Johannesburg. Its stated ambition is 12,000 to 13,000 GPUs across the entire continent.<\/p>\n\n\n\n
For scale: a single site in Memphis, Tennessee was reported in January 2026 at around 555,000 GPUs. Published figures for that cluster vary by source; the order of magnitude does not. Cassava\u2019s continental target is roughly two percent of one American building.<\/p>\n\n\n\n
That comparison is not a verdict on Cassava, which is doing something difficult and necessary that nobody else on the continent is attempting at scale. It isolates the variable that actually matters: on current trajectories Africa will not out-build the frontier, and any sovereignty strategy premised on doing so is a strategy for permanent disappointment.<\/p>\n\n\n\n
\u00a0Fifty-five sovereignties, one uneconomic market<\/strong><\/h2>\n\n\n\nIf every country insists that its citizens\u2019 data live within its own borders, Africa gets fifty-odd subscale data centres, each too small to be efficient, each sitting on a grid that cannot reliably power it, each raising the cost of serving its own market. Sovereignty pursued at the national grain produces less resilience, not more \u2014 because the smallest facilities are the ones most likely to fail the cut test in the least dramatic way: by being switched off during load-shedding.<\/p>\n\n\n\n
The African Union has, to its credit, seen this coming. Its Data Policy Framework, the Continental AI Strategy adopted in July 2024 and the AfCFTA Digital Trade Protocol ratified in February 2024 all lean the same way \u2014 harmonised rules, conditional rather than absolute localisation, and free movement of data within the continent. The strategy\u2019s own ambition is for AI to add $1.5 trillion to African economies by 2030. That is a projection, and a stretching one, but the direction it sets is unmistakable.<\/p>\n\n\n\n
Implementation is the problem. The Malabo Convention, the AU\u2019s binding instrument on cybersecurity and personal data, was adopted in 2014 and took until June 2023 to enter into force. On the AU\u2019s own July 2024 tally, 21 member states had signed and 16 had ratified, out of 55. South Africa is not among them; nor, on that list, is Egypt. By a separate count, 22 of 54 countries had a national AI strategy by 2026, up from 16 in 2024.<\/p>\n\n\n\n
The Malawi finding points at the way out. A country serving 3 percent of its content locally and 62 percent regionally has not failed; it has pooled. Regional hosting is what national sovereignty looks like once it is made affordable. The unit that can actually pass the cut test in Africa is not the country. It is the region.<\/p>\n\n\n\n
The constraint has moved to the power station<\/strong><\/h2>\n\n\n\nOn 24 July 2026, the City of Cape Town approved two hyperscale data centres that will together draw around 174 megawatts. That single approval is close to half of the continent\u2019s entire installed data centre capacity today. McKinsey expects Africa to move from roughly 0.4 GW now to between 1.5 and 2.2 GW by 2030, requiring $10\u201320 billion of investment.<\/p>\n\n\n\n
The evidence keeps pointing the same way. Microsoft\u2019s $1 billion Kenyan project with G42 is built around geothermal power, with 100 MW planned. Nigeria\u2019s Kasi Cloud commissioned a carrier-neutral campus in Lekki in June 2026, designed to scale to about 100 MW. Raxio reported that its contracted power grew sixfold in the first half of 2026 against the same period a year earlier. In each case the pitch to capital is not bandwidth. It is firm power.<\/p>\n\n\n\n
And the constraint is severe. Around 600 million Africans still live without electricity \u2014 a figure that shifts with methodology, from about 565 million in the Tracking SDG7 baseline to roughly 590 million in the IEA\u2019s most recent investment report, but never in a direction that changes the argument. Per capita electricity use in sub-Saharan Africa has been broadly flat for three decades. Meanwhile a large data centre can consume up to 19 million litres of water a day for cooling, on figures put to South African councillors \u2014 which is why, when eThekwini approved an AI data centre agreement in Durban in March 2026, a senior opposition councillor withheld support pending transparency on its energy and water demand.<\/p>\n\n\n\n
That councillor was doing something more sophisticated than obstruction. He was asking the sovereignty question in its correct, current form: not whose flag is on the building, but who bears the cost of keeping it running, and whose supply gets shorter when it does.<\/p>\n\n\n\n
The unglamorous agenda<\/strong><\/h2>\n\n\n\nFor founders. <\/strong>The sovereign layer is the one Africa has always won, rails, rules and the last mile. Payment infrastructure, identity, clearing, logistics, health records. Compute is a rented input, and the firms that survive the next decade will be the ones that treat it as such while owning the relationship with the customer and the licence that governs it.<\/p>\n\n\n\nFor investors. <\/strong>The scarce asset in African digital infrastructure is no longer fibre; it is firm, dispatchable, contractable power, and the transmission and storage to move it. A hyperscale tenant is the best offtaker a private power project has ever been offered on this continent: predictable, round-the-clock, multi-decade demand. Cape Town\u2019s 174 megawatts is the demand signal. The generation to meet it is the trade.<\/p>\n\n\n\nFor ministers. <\/strong>Localisation mandates without regional harmonisation are a tax on your own economy. The higher-return moves are cheaper and far less announceable: fund the internet exchange, subsidise local caching and CDN presence, ratify the continental instruments so data can move within Africa as freely as it already moves to Frankfurt, and buy substitutability. Open standards and exit-ready contracts are how you stop a foreign compliance department from becoming your policy.<\/p>\n\n\n\nThe rockslide off Abidjan did the continent an unintended favour. It ran, for free, the experiment no government would ever have commissioned: it switched off the outside world for a few weeks and let everybody see which parts of their national life were actually theirs. Payments held. Exchanges held. Locally hosted content held. Everything routed through somewhere else went dark.<\/p>\n\n\n\n
The next cut will not be geological, and it will not announce itself with an outage graph. It will arrive as a sanctions list, a licence revocation, an export control on a chip, a clause invoked by a compliance officer eight thousand kilometres away. Africa cannot stop that switch from existing. Nobody can.<\/p>\n\n\n\n
What it can still decide, layer by layer and within this decade, is how much of ordinary life is wired through it.<\/p>\n\n\n\n
A note on sources and method<\/strong>
Outage figures are NetBlocks network measurements as reported by the BBC and Reuters on 14 March 2024, with restoration dates from cable operators and the Internet Society\u2019s April 2024 outage report. Mobile money data is the GSMA\u2019s State of the Industry Report on Mobile Money 2026 (covering 2025). Content-locality measurements are preliminary results from the Internet Society Pulse research fellowship, published June 2026. Compute geography draws on Lehdonvirta, W\u00fa and Hawkins, \u2018Compute North vs. Compute South\u2019 (AIES 2024) and subsequent Oxford Internet Institute mapping reported in June 2025; African compute-access figures are from Alliance4ai and Zindi via UNDP, and the GSMA.
Where credible sources disagree, on Liberia\u2019s outage depth, on Africa\u2019s installed data centre capacity, on the number of people without electricity, and on the Microsoft\u2013ICC episode, the disagreement is stated in the text and the figure used is identified. Forward-looking figures from McKinsey and the African Union are projections and are marked as such. No figure in this feature has been estimated or interpolated by Villpress.<\/strong><\/pre>","protected":false},"excerpt":{"rendered":"On 14 March 2024, a stretch of seabed off C\u00f4te d\u2019Ivoire gave way. Four submarine cables,\u00a0WACS, SAT-3, ACE and MainOne,\u00a0failed within hours of one another. There was no attacker, no ransom note, nobody to negotiate with. A rock had moved. By that afternoon C\u00f4te d\u2019Ivoire\u2019s internet was running at roughly four percent of its ordinary […]<\/p>\n","protected":false},"author":1,"featured_media":12287,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"_acf_changed":false,"_mi_skip_tracking":false,"footnotes":""},"categories":[2109],"tags":[2247],"ppma_author":[331],"class_list":["post-12279","post","type-post","status-publish","format-standard","has-post-thumbnail","category-villpress-intelligence","tag-infastructure"],"acf":[],"authors":[{"term_id":331,"user_id":1,"is_guest":0,"slug":"pastakutmanwen","display_name":"Staff Writer","avatar_url":{"url":"https:\/\/villpress.com\/wp-content\/uploads\/2025\/05\/Logo.png","url2x":"https:\/\/villpress.com\/wp-content\/uploads\/2025\/05\/Logo.png"},"author_category":"1","first_name":"Staff","last_name":"Writer","user_url":"http:\/\/villpress.com","job_title":"Staffs At Villpress","description":"The Villpress Staff Writers are an in-house team of experienced editors and industry experts dedicated to producing clear, insightful content. As part of Villpress, they cover the latest trends and innovations across business, technology, artificial intelligence, advertising, and more, delivering stories that inform, engage, and add real value to readers."}],"_links":{"self":[{"href":"https:\/\/villpress.com\/de\/wp-json\/wp\/v2\/posts\/12279","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/villpress.com\/de\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/villpress.com\/de\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/villpress.com\/de\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/villpress.com\/de\/wp-json\/wp\/v2\/comments?post=12279"}],"version-history":[{"count":2,"href":"https:\/\/villpress.com\/de\/wp-json\/wp\/v2\/posts\/12279\/revisions"}],"predecessor-version":[{"id":12288,"href":"https:\/\/villpress.com\/de\/wp-json\/wp\/v2\/posts\/12279\/revisions\/12288"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/villpress.com\/de\/wp-json\/wp\/v2\/media\/12287"}],"wp:attachment":[{"href":"https:\/\/villpress.com\/de\/wp-json\/wp\/v2\/media?parent=12279"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/villpress.com\/de\/wp-json\/wp\/v2\/categories?post=12279"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/villpress.com\/de\/wp-json\/wp\/v2\/tags?post=12279"},{"taxonomy":"author","embeddable":true,"href":"https:\/\/villpress.com\/de\/wp-json\/wp\/v2\/ppma_author?post=12279"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}
Stating where a diagnostic fails is what makes it usable, so state it plainly: the cut test misses every dependency that is legal or commercial rather than physical. Under section 2713 of the US CLOUD Act, American providers can be compelled to produce data in their possession, custody or control regardless of which country it is stored in. Localisation answers the question \u201cwhere does the disk sit?\u201d It does not answer \u201cwho can be ordered to unlock it?\u201d<\/p>\n\n\n\n
The same logic runs through the cables. Google\u2019s Equiano, at 144 Tbps, is wholly owned by Google. Meta\u2019s 2Africa, at 180 Tbps and completed in November 2025, encircles the continent across 45,000 kilometres and 33 countries \u2014 built by a consortium in which Meta sits alongside MTN\u2019s Bayobab, Orange, Vodafone, Telecom Egypt, stc\u2019s center3 and WIOCC. Researchers writing in The Information Society note that each of these two systems dwarfs the combined capacity of Africa\u2019s other cables by roughly an order of magnitude. The Atlantic Council\u2019s Digital Forensic Research Lab has found that Google, Meta, Microsoft and Amazon between them own or lease around 70 percent of global undersea bandwidth.<\/p>\n\n\n\n
The result is genuinely double-edged, and pretending otherwise would be dishonest. West Africa went dark in 2024 partly because it was leaning on a handful of ageing systems; the new cables are precisely the redundancy the region lacked, and African operators hold real equity in one of them. But the redundancy arrived courtesy of the two companies whose own services generate much of the traffic. Africa got a better road, built by the trucking company.<\/p>\n\n\n\n
Compute is failing the test faster than anything else<\/strong><\/h2>\n\n\n\nWhich brings the argument to the layer where the answer is worst and the clock runs quickest.<\/p>\n\n\n\n
Oxford researchers mapping the world\u2019s publicly available AI compute found it in just 32 countries, with the United States and China operating the overwhelming majority of specialized AI data centres between them. Most of the world sits in what the authors call compute deserts. Africa\u2019s presence on that map is in the single digits.<\/p>\n\n\n\n
The scarcity is felt at the level of an individual researcher. An analysis of Zindi\u2019s network of 11,000 African data scientists found that only about five percent had access to the computing power their work required, and that the first one percent were simply those with GPUs physically in their offices. The GSMA notes that a single GPU costs the equivalent of 75 percent of GDP per capita in Kenya and 69 percent in Senegal. The absolute price is much the same everywhere. That is exactly the problem.<\/p>\n\n\n\n
The most serious African answer is Cassava Technologies, which announced in March 2025 that it would build the continent\u2019s first AI factory in South Africa with Nvidia, targeting deployment by that June. By November 2025 it had acquired 12,000 Nvidia GPUs and launched GPU-as-a-service; in March 2026 it announced deployment in South Africa with Nigeria, Kenya, Egypt and Morocco to follow; by May 2026 the Cape Town facility was preparing to go live and a second was planned for Johannesburg. Its stated ambition is 12,000 to 13,000 GPUs across the entire continent.<\/p>\n\n\n\n
For scale: a single site in Memphis, Tennessee was reported in January 2026 at around 555,000 GPUs. Published figures for that cluster vary by source; the order of magnitude does not. Cassava\u2019s continental target is roughly two percent of one American building.<\/p>\n\n\n\n
That comparison is not a verdict on Cassava, which is doing something difficult and necessary that nobody else on the continent is attempting at scale. It isolates the variable that actually matters: on current trajectories Africa will not out-build the frontier, and any sovereignty strategy premised on doing so is a strategy for permanent disappointment.<\/p>\n\n\n\n
\u00a0Fifty-five sovereignties, one uneconomic market<\/strong><\/h2>\n\n\n\nIf every country insists that its citizens\u2019 data live within its own borders, Africa gets fifty-odd subscale data centres, each too small to be efficient, each sitting on a grid that cannot reliably power it, each raising the cost of serving its own market. Sovereignty pursued at the national grain produces less resilience, not more \u2014 because the smallest facilities are the ones most likely to fail the cut test in the least dramatic way: by being switched off during load-shedding.<\/p>\n\n\n\n
The African Union has, to its credit, seen this coming. Its Data Policy Framework, the Continental AI Strategy adopted in July 2024 and the AfCFTA Digital Trade Protocol ratified in February 2024 all lean the same way \u2014 harmonised rules, conditional rather than absolute localisation, and free movement of data within the continent. The strategy\u2019s own ambition is for AI to add $1.5 trillion to African economies by 2030. That is a projection, and a stretching one, but the direction it sets is unmistakable.<\/p>\n\n\n\n
Implementation is the problem. The Malabo Convention, the AU\u2019s binding instrument on cybersecurity and personal data, was adopted in 2014 and took until June 2023 to enter into force. On the AU\u2019s own July 2024 tally, 21 member states had signed and 16 had ratified, out of 55. South Africa is not among them; nor, on that list, is Egypt. By a separate count, 22 of 54 countries had a national AI strategy by 2026, up from 16 in 2024.<\/p>\n\n\n\n
The Malawi finding points at the way out. A country serving 3 percent of its content locally and 62 percent regionally has not failed; it has pooled. Regional hosting is what national sovereignty looks like once it is made affordable. The unit that can actually pass the cut test in Africa is not the country. It is the region.<\/p>\n\n\n\n
The constraint has moved to the power station<\/strong><\/h2>\n\n\n\nOn 24 July 2026, the City of Cape Town approved two hyperscale data centres that will together draw around 174 megawatts. That single approval is close to half of the continent\u2019s entire installed data centre capacity today. McKinsey expects Africa to move from roughly 0.4 GW now to between 1.5 and 2.2 GW by 2030, requiring $10\u201320 billion of investment.<\/p>\n\n\n\n
The evidence keeps pointing the same way. Microsoft\u2019s $1 billion Kenyan project with G42 is built around geothermal power, with 100 MW planned. Nigeria\u2019s Kasi Cloud commissioned a carrier-neutral campus in Lekki in June 2026, designed to scale to about 100 MW. Raxio reported that its contracted power grew sixfold in the first half of 2026 against the same period a year earlier. In each case the pitch to capital is not bandwidth. It is firm power.<\/p>\n\n\n\n
And the constraint is severe. Around 600 million Africans still live without electricity \u2014 a figure that shifts with methodology, from about 565 million in the Tracking SDG7 baseline to roughly 590 million in the IEA\u2019s most recent investment report, but never in a direction that changes the argument. Per capita electricity use in sub-Saharan Africa has been broadly flat for three decades. Meanwhile a large data centre can consume up to 19 million litres of water a day for cooling, on figures put to South African councillors \u2014 which is why, when eThekwini approved an AI data centre agreement in Durban in March 2026, a senior opposition councillor withheld support pending transparency on its energy and water demand.<\/p>\n\n\n\n
That councillor was doing something more sophisticated than obstruction. He was asking the sovereignty question in its correct, current form: not whose flag is on the building, but who bears the cost of keeping it running, and whose supply gets shorter when it does.<\/p>\n\n\n\n
The unglamorous agenda<\/strong><\/h2>\n\n\n\nFor founders. <\/strong>The sovereign layer is the one Africa has always won, rails, rules and the last mile. Payment infrastructure, identity, clearing, logistics, health records. Compute is a rented input, and the firms that survive the next decade will be the ones that treat it as such while owning the relationship with the customer and the licence that governs it.<\/p>\n\n\n\nFor investors. <\/strong>The scarce asset in African digital infrastructure is no longer fibre; it is firm, dispatchable, contractable power, and the transmission and storage to move it. A hyperscale tenant is the best offtaker a private power project has ever been offered on this continent: predictable, round-the-clock, multi-decade demand. Cape Town\u2019s 174 megawatts is the demand signal. The generation to meet it is the trade.<\/p>\n\n\n\nFor ministers. <\/strong>Localisation mandates without regional harmonisation are a tax on your own economy. The higher-return moves are cheaper and far less announceable: fund the internet exchange, subsidise local caching and CDN presence, ratify the continental instruments so data can move within Africa as freely as it already moves to Frankfurt, and buy substitutability. Open standards and exit-ready contracts are how you stop a foreign compliance department from becoming your policy.<\/p>\n\n\n\nThe rockslide off Abidjan did the continent an unintended favour. It ran, for free, the experiment no government would ever have commissioned: it switched off the outside world for a few weeks and let everybody see which parts of their national life were actually theirs. Payments held. Exchanges held. Locally hosted content held. Everything routed through somewhere else went dark.<\/p>\n\n\n\n
The next cut will not be geological, and it will not announce itself with an outage graph. It will arrive as a sanctions list, a licence revocation, an export control on a chip, a clause invoked by a compliance officer eight thousand kilometres away. Africa cannot stop that switch from existing. Nobody can.<\/p>\n\n\n\n
What it can still decide, layer by layer and within this decade, is how much of ordinary life is wired through it.<\/p>\n\n\n\n
A note on sources and method<\/strong>
Outage figures are NetBlocks network measurements as reported by the BBC and Reuters on 14 March 2024, with restoration dates from cable operators and the Internet Society\u2019s April 2024 outage report. Mobile money data is the GSMA\u2019s State of the Industry Report on Mobile Money 2026 (covering 2025). Content-locality measurements are preliminary results from the Internet Society Pulse research fellowship, published June 2026. Compute geography draws on Lehdonvirta, W\u00fa and Hawkins, \u2018Compute North vs. Compute South\u2019 (AIES 2024) and subsequent Oxford Internet Institute mapping reported in June 2025; African compute-access figures are from Alliance4ai and Zindi via UNDP, and the GSMA.
Where credible sources disagree, on Liberia\u2019s outage depth, on Africa\u2019s installed data centre capacity, on the number of people without electricity, and on the Microsoft\u2013ICC episode, the disagreement is stated in the text and the figure used is identified. Forward-looking figures from McKinsey and the African Union are projections and are marked as such. No figure in this feature has been estimated or interpolated by Villpress.<\/strong><\/pre>","protected":false},"excerpt":{"rendered":"On 14 March 2024, a stretch of seabed off C\u00f4te d\u2019Ivoire gave way. Four submarine cables,\u00a0WACS, SAT-3, ACE and MainOne,\u00a0failed within hours of one another. There was no attacker, no ransom note, nobody to negotiate with. A rock had moved. By that afternoon C\u00f4te d\u2019Ivoire\u2019s internet was running at roughly four percent of its ordinary […]<\/p>\n","protected":false},"author":1,"featured_media":12287,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"_acf_changed":false,"_mi_skip_tracking":false,"footnotes":""},"categories":[2109],"tags":[2247],"ppma_author":[331],"class_list":["post-12279","post","type-post","status-publish","format-standard","has-post-thumbnail","category-villpress-intelligence","tag-infastructure"],"acf":[],"authors":[{"term_id":331,"user_id":1,"is_guest":0,"slug":"pastakutmanwen","display_name":"Staff Writer","avatar_url":{"url":"https:\/\/villpress.com\/wp-content\/uploads\/2025\/05\/Logo.png","url2x":"https:\/\/villpress.com\/wp-content\/uploads\/2025\/05\/Logo.png"},"author_category":"1","first_name":"Staff","last_name":"Writer","user_url":"http:\/\/villpress.com","job_title":"Staffs At Villpress","description":"The Villpress Staff Writers are an in-house team of experienced editors and industry experts dedicated to producing clear, insightful content. As part of Villpress, they cover the latest trends and innovations across business, technology, artificial intelligence, advertising, and more, delivering stories that inform, engage, and add real value to readers."}],"_links":{"self":[{"href":"https:\/\/villpress.com\/de\/wp-json\/wp\/v2\/posts\/12279","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/villpress.com\/de\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/villpress.com\/de\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/villpress.com\/de\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/villpress.com\/de\/wp-json\/wp\/v2\/comments?post=12279"}],"version-history":[{"count":2,"href":"https:\/\/villpress.com\/de\/wp-json\/wp\/v2\/posts\/12279\/revisions"}],"predecessor-version":[{"id":12288,"href":"https:\/\/villpress.com\/de\/wp-json\/wp\/v2\/posts\/12279\/revisions\/12288"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/villpress.com\/de\/wp-json\/wp\/v2\/media\/12287"}],"wp:attachment":[{"href":"https:\/\/villpress.com\/de\/wp-json\/wp\/v2\/media?parent=12279"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/villpress.com\/de\/wp-json\/wp\/v2\/categories?post=12279"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/villpress.com\/de\/wp-json\/wp\/v2\/tags?post=12279"},{"taxonomy":"author","embeddable":true,"href":"https:\/\/villpress.com\/de\/wp-json\/wp\/v2\/ppma_author?post=12279"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}
If every country insists that its citizens\u2019 data live within its own borders, Africa gets fifty-odd subscale data centres, each too small to be efficient, each sitting on a grid that cannot reliably power it, each raising the cost of serving its own market. Sovereignty pursued at the national grain produces less resilience, not more \u2014 because the smallest facilities are the ones most likely to fail the cut test in the least dramatic way: by being switched off during load-shedding.<\/p>\n\n\n\n
The African Union has, to its credit, seen this coming. Its Data Policy Framework, the Continental AI Strategy adopted in July 2024 and the AfCFTA Digital Trade Protocol ratified in February 2024 all lean the same way \u2014 harmonised rules, conditional rather than absolute localisation, and free movement of data within the continent. The strategy\u2019s own ambition is for AI to add $1.5 trillion to African economies by 2030. That is a projection, and a stretching one, but the direction it sets is unmistakable.<\/p>\n\n\n\n
Implementation is the problem. The Malabo Convention, the AU\u2019s binding instrument on cybersecurity and personal data, was adopted in 2014 and took until June 2023 to enter into force. On the AU\u2019s own July 2024 tally, 21 member states had signed and 16 had ratified, out of 55. South Africa is not among them; nor, on that list, is Egypt. By a separate count, 22 of 54 countries had a national AI strategy by 2026, up from 16 in 2024.<\/p>\n\n\n\n
The Malawi finding points at the way out. A country serving 3 percent of its content locally and 62 percent regionally has not failed; it has pooled. Regional hosting is what national sovereignty looks like once it is made affordable. The unit that can actually pass the cut test in Africa is not the country. It is the region.<\/p>\n\n\n\n
The constraint has moved to the power station<\/strong><\/h2>\n\n\n\nOn 24 July 2026, the City of Cape Town approved two hyperscale data centres that will together draw around 174 megawatts. That single approval is close to half of the continent\u2019s entire installed data centre capacity today. McKinsey expects Africa to move from roughly 0.4 GW now to between 1.5 and 2.2 GW by 2030, requiring $10\u201320 billion of investment.<\/p>\n\n\n\n
The evidence keeps pointing the same way. Microsoft\u2019s $1 billion Kenyan project with G42 is built around geothermal power, with 100 MW planned. Nigeria\u2019s Kasi Cloud commissioned a carrier-neutral campus in Lekki in June 2026, designed to scale to about 100 MW. Raxio reported that its contracted power grew sixfold in the first half of 2026 against the same period a year earlier. In each case the pitch to capital is not bandwidth. It is firm power.<\/p>\n\n\n\n
And the constraint is severe. Around 600 million Africans still live without electricity \u2014 a figure that shifts with methodology, from about 565 million in the Tracking SDG7 baseline to roughly 590 million in the IEA\u2019s most recent investment report, but never in a direction that changes the argument. Per capita electricity use in sub-Saharan Africa has been broadly flat for three decades. Meanwhile a large data centre can consume up to 19 million litres of water a day for cooling, on figures put to South African councillors \u2014 which is why, when eThekwini approved an AI data centre agreement in Durban in March 2026, a senior opposition councillor withheld support pending transparency on its energy and water demand.<\/p>\n\n\n\n
That councillor was doing something more sophisticated than obstruction. He was asking the sovereignty question in its correct, current form: not whose flag is on the building, but who bears the cost of keeping it running, and whose supply gets shorter when it does.<\/p>\n\n\n\n
The unglamorous agenda<\/strong><\/h2>\n\n\n\nFor founders. <\/strong>The sovereign layer is the one Africa has always won, rails, rules and the last mile. Payment infrastructure, identity, clearing, logistics, health records. Compute is a rented input, and the firms that survive the next decade will be the ones that treat it as such while owning the relationship with the customer and the licence that governs it.<\/p>\n\n\n\nFor investors. <\/strong>The scarce asset in African digital infrastructure is no longer fibre; it is firm, dispatchable, contractable power, and the transmission and storage to move it. A hyperscale tenant is the best offtaker a private power project has ever been offered on this continent: predictable, round-the-clock, multi-decade demand. Cape Town\u2019s 174 megawatts is the demand signal. The generation to meet it is the trade.<\/p>\n\n\n\nFor ministers. <\/strong>Localisation mandates without regional harmonisation are a tax on your own economy. The higher-return moves are cheaper and far less announceable: fund the internet exchange, subsidise local caching and CDN presence, ratify the continental instruments so data can move within Africa as freely as it already moves to Frankfurt, and buy substitutability. Open standards and exit-ready contracts are how you stop a foreign compliance department from becoming your policy.<\/p>\n\n\n\nThe rockslide off Abidjan did the continent an unintended favour. It ran, for free, the experiment no government would ever have commissioned: it switched off the outside world for a few weeks and let everybody see which parts of their national life were actually theirs. Payments held. Exchanges held. Locally hosted content held. Everything routed through somewhere else went dark.<\/p>\n\n\n\n
The next cut will not be geological, and it will not announce itself with an outage graph. It will arrive as a sanctions list, a licence revocation, an export control on a chip, a clause invoked by a compliance officer eight thousand kilometres away. Africa cannot stop that switch from existing. Nobody can.<\/p>\n\n\n\n
What it can still decide, layer by layer and within this decade, is how much of ordinary life is wired through it.<\/p>\n\n\n\n
A note on sources and method<\/strong>
Outage figures are NetBlocks network measurements as reported by the BBC and Reuters on 14 March 2024, with restoration dates from cable operators and the Internet Society\u2019s April 2024 outage report. Mobile money data is the GSMA\u2019s State of the Industry Report on Mobile Money 2026 (covering 2025). Content-locality measurements are preliminary results from the Internet Society Pulse research fellowship, published June 2026. Compute geography draws on Lehdonvirta, W\u00fa and Hawkins, \u2018Compute North vs. Compute South\u2019 (AIES 2024) and subsequent Oxford Internet Institute mapping reported in June 2025; African compute-access figures are from Alliance4ai and Zindi via UNDP, and the GSMA.
Where credible sources disagree, on Liberia\u2019s outage depth, on Africa\u2019s installed data centre capacity, on the number of people without electricity, and on the Microsoft\u2013ICC episode, the disagreement is stated in the text and the figure used is identified. Forward-looking figures from McKinsey and the African Union are projections and are marked as such. No figure in this feature has been estimated or interpolated by Villpress.<\/strong><\/pre>","protected":false},"excerpt":{"rendered":"On 14 March 2024, a stretch of seabed off C\u00f4te d\u2019Ivoire gave way. Four submarine cables,\u00a0WACS, SAT-3, ACE and MainOne,\u00a0failed within hours of one another. There was no attacker, no ransom note, nobody to negotiate with. A rock had moved. By that afternoon C\u00f4te d\u2019Ivoire\u2019s internet was running at roughly four percent of its ordinary […]<\/p>\n","protected":false},"author":1,"featured_media":12287,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"_acf_changed":false,"_mi_skip_tracking":false,"footnotes":""},"categories":[2109],"tags":[2247],"ppma_author":[331],"class_list":["post-12279","post","type-post","status-publish","format-standard","has-post-thumbnail","category-villpress-intelligence","tag-infastructure"],"acf":[],"authors":[{"term_id":331,"user_id":1,"is_guest":0,"slug":"pastakutmanwen","display_name":"Staff Writer","avatar_url":{"url":"https:\/\/villpress.com\/wp-content\/uploads\/2025\/05\/Logo.png","url2x":"https:\/\/villpress.com\/wp-content\/uploads\/2025\/05\/Logo.png"},"author_category":"1","first_name":"Staff","last_name":"Writer","user_url":"http:\/\/villpress.com","job_title":"Staffs At Villpress","description":"The Villpress Staff Writers are an in-house team of experienced editors and industry experts dedicated to producing clear, insightful content. As part of Villpress, they cover the latest trends and innovations across business, technology, artificial intelligence, advertising, and more, delivering stories that inform, engage, and add real value to readers."}],"_links":{"self":[{"href":"https:\/\/villpress.com\/de\/wp-json\/wp\/v2\/posts\/12279","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/villpress.com\/de\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/villpress.com\/de\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/villpress.com\/de\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/villpress.com\/de\/wp-json\/wp\/v2\/comments?post=12279"}],"version-history":[{"count":2,"href":"https:\/\/villpress.com\/de\/wp-json\/wp\/v2\/posts\/12279\/revisions"}],"predecessor-version":[{"id":12288,"href":"https:\/\/villpress.com\/de\/wp-json\/wp\/v2\/posts\/12279\/revisions\/12288"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/villpress.com\/de\/wp-json\/wp\/v2\/media\/12287"}],"wp:attachment":[{"href":"https:\/\/villpress.com\/de\/wp-json\/wp\/v2\/media?parent=12279"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/villpress.com\/de\/wp-json\/wp\/v2\/categories?post=12279"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/villpress.com\/de\/wp-json\/wp\/v2\/tags?post=12279"},{"taxonomy":"author","embeddable":true,"href":"https:\/\/villpress.com\/de\/wp-json\/wp\/v2\/ppma_author?post=12279"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}
For founders. <\/strong>The sovereign layer is the one Africa has always won, rails, rules and the last mile. Payment infrastructure, identity, clearing, logistics, health records. Compute is a rented input, and the firms that survive the next decade will be the ones that treat it as such while owning the relationship with the customer and the licence that governs it.<\/p>\n\n\n\n For investors. <\/strong>The scarce asset in African digital infrastructure is no longer fibre; it is firm, dispatchable, contractable power, and the transmission and storage to move it. A hyperscale tenant is the best offtaker a private power project has ever been offered on this continent: predictable, round-the-clock, multi-decade demand. Cape Town\u2019s 174 megawatts is the demand signal. The generation to meet it is the trade.<\/p>\n\n\n\n For ministers. <\/strong>Localisation mandates without regional harmonisation are a tax on your own economy. The higher-return moves are cheaper and far less announceable: fund the internet exchange, subsidise local caching and CDN presence, ratify the continental instruments so data can move within Africa as freely as it already moves to Frankfurt, and buy substitutability. Open standards and exit-ready contracts are how you stop a foreign compliance department from becoming your policy.<\/p>\n\n\n\n The rockslide off Abidjan did the continent an unintended favour. It ran, for free, the experiment no government would ever have commissioned: it switched off the outside world for a few weeks and let everybody see which parts of their national life were actually theirs. Payments held. Exchanges held. Locally hosted content held. Everything routed through somewhere else went dark.<\/p>\n\n\n\n The next cut will not be geological, and it will not announce itself with an outage graph. It will arrive as a sanctions list, a licence revocation, an export control on a chip, a clause invoked by a compliance officer eight thousand kilometres away. Africa cannot stop that switch from existing. Nobody can.<\/p>\n\n\n\n What it can still decide, layer by layer and within this decade, is how much of ordinary life is wired through it.<\/p>\n\n\n\n On 14 March 2024, a stretch of seabed off C\u00f4te d\u2019Ivoire gave way. Four submarine cables,\u00a0WACS, SAT-3, ACE and MainOne,\u00a0failed within hours of one another. There was no attacker, no ransom note, nobody to negotiate with. A rock had moved. By that afternoon C\u00f4te d\u2019Ivoire\u2019s internet was running at roughly four percent of its ordinary […]<\/p>\n","protected":false},"author":1,"featured_media":12287,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"_acf_changed":false,"_mi_skip_tracking":false,"footnotes":""},"categories":[2109],"tags":[2247],"ppma_author":[331],"class_list":["post-12279","post","type-post","status-publish","format-standard","has-post-thumbnail","category-villpress-intelligence","tag-infastructure"],"acf":[],"authors":[{"term_id":331,"user_id":1,"is_guest":0,"slug":"pastakutmanwen","display_name":"Staff Writer","avatar_url":{"url":"https:\/\/villpress.com\/wp-content\/uploads\/2025\/05\/Logo.png","url2x":"https:\/\/villpress.com\/wp-content\/uploads\/2025\/05\/Logo.png"},"author_category":"1","first_name":"Staff","last_name":"Writer","user_url":"http:\/\/villpress.com","job_title":"Staffs At Villpress","description":"The Villpress Staff Writers are an in-house team of experienced editors and industry experts dedicated to producing clear, insightful content. As part of Villpress, they cover the latest trends and innovations across business, technology, artificial intelligence, advertising, and more, delivering stories that inform, engage, and add real value to readers."}],"_links":{"self":[{"href":"https:\/\/villpress.com\/de\/wp-json\/wp\/v2\/posts\/12279","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/villpress.com\/de\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/villpress.com\/de\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/villpress.com\/de\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/villpress.com\/de\/wp-json\/wp\/v2\/comments?post=12279"}],"version-history":[{"count":2,"href":"https:\/\/villpress.com\/de\/wp-json\/wp\/v2\/posts\/12279\/revisions"}],"predecessor-version":[{"id":12288,"href":"https:\/\/villpress.com\/de\/wp-json\/wp\/v2\/posts\/12279\/revisions\/12288"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/villpress.com\/de\/wp-json\/wp\/v2\/media\/12287"}],"wp:attachment":[{"href":"https:\/\/villpress.com\/de\/wp-json\/wp\/v2\/media?parent=12279"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/villpress.com\/de\/wp-json\/wp\/v2\/categories?post=12279"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/villpress.com\/de\/wp-json\/wp\/v2\/tags?post=12279"},{"taxonomy":"author","embeddable":true,"href":"https:\/\/villpress.com\/de\/wp-json\/wp\/v2\/ppma_author?post=12279"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}A note on sources and method<\/strong>
Outage figures are NetBlocks network measurements as reported by the BBC and Reuters on 14 March 2024, with restoration dates from cable operators and the Internet Society\u2019s April 2024 outage report. Mobile money data is the GSMA\u2019s State of the Industry Report on Mobile Money 2026 (covering 2025). Content-locality measurements are preliminary results from the Internet Society Pulse research fellowship, published June 2026. Compute geography draws on Lehdonvirta, W\u00fa and Hawkins, \u2018Compute North vs. Compute South\u2019 (AIES 2024) and subsequent Oxford Internet Institute mapping reported in June 2025; African compute-access figures are from Alliance4ai and Zindi via UNDP, and the GSMA.
Where credible sources disagree, on Liberia\u2019s outage depth, on Africa\u2019s installed data centre capacity, on the number of people without electricity, and on the Microsoft\u2013ICC episode, the disagreement is stated in the text and the figure used is identified. Forward-looking figures from McKinsey and the African Union are projections and are marked as such. No figure in this feature has been estimated or interpolated by Villpress.<\/strong><\/pre>","protected":false},"excerpt":{"rendered":"