The Said & Done Ledger

1. Announced vs. deployed — the headline gap#

The single most important number in this Ledger: announced continental AI commitments exceed US$70bn. Verifiable deployed AI-specific capital, including infrastructure capex, is on the order of US$1–2bn.

Announced commitments#

Claims logged at events#

Events lane: about a week after a major continental AI event closes, each claim made on stage is logged here with its speaker, source and the date we check it. Status is as of the "logged" date; the check result is added in place, never overwritten.

22 Dec = 90 days after PAAIS closed, per Christabel Danso's 90-day test. Next event: Africa Tech Festival, Cape Town, 17–19 Nov 2026.

Ledger read: the deployed column is dominated by corporate infrastructure capex (Cassava, Google, Microsoft SA, the Nigerian data centres) and philanthropic programme money (Gates, IDRC, UNDP). The two largest headline numbers — the US$60bn fund and the US$10bn initiative — account for US$70bn of announcements and no verifiable deployment.

2. The startup count#

Baseline: 207 AI companies tracked across 17 countries.

Source for this section: TechCabal Insights, Africa's AI Builders, April 2026, analysing Dr Chinasa T. Okolo's Harvard Dataverse dataset.

Concentration. Nigeria 50 · South Africa 49 · Kenya 31 = 63% of all tracked. Ghana 13 · Tunisia 11 · Egypt 11. Regional: West 71 · Southern 53 · East 50 · North 26.

Maturity. Kenya is proportionally the most mature ecosystem — 42% at growth stage, vs 25% Nigeria, 24% South Africa.

Vertical movement. Software Development (48) functions partly as a catch-all. The real vertical story: Finance 22 · Agriculture 20 · Healthcare 20 · Education 14 — 37% of total. Education grew 2 → 14. Legal AI went 0 → 6, three of them in Nigeria, all founded 2024–25. Egypt grew 3 → 11 (+267%).

Next update: TechCabal's annual refresh, expected ~April 2027.

3. Compute — announced vs. online#

Primary source: IMF, Unlocking the Potential: AI in Sub-Saharan Africa, Departmental Paper, 13 Jul 2026, team led by Martin Schindler and Andrew Tiffin. Page numbers below are printed page numbers, verified page by page.**

The unit problem, stated first#

Africa's compute is quoted in three different units, and they tell three different stories. Quote the wrong one and the continent looks four to nine times better supplied than it is.

Source: IMF departmental paper, pp.15 and 18.

Capacity — the best-corroborated figure in the whole landscape, and the least quoted#

Not contested. Three independent sources converge:

The forward path — the share does not move#

Sources: IMF p.15; ADCA; Baker McKenzie.

Baker McKenzie's scale comparison: the United States alone holds close to 45% of global capacity, and the continent's entire projected 2030 capacity is "about the size of just one large American AI campus." The single site unveiled outside Abu Dhabi in May 2025 is 5 GW.

Demand side — the 5% figure and its provenance#

Affordability — the constraint the geography framing misses#

Off-continent processing#

Cassava GPU tracker — the continent's compute bellwether#

AI factories — switched on, and what "switched on" means#

AI-factory ladder#

Five rungs, each needing its own evidence: 1 announced · 2 sited and financed · 3 hardware on site · 4 commissioned · 5 serving customers (named customers count for more than an operator's word). A project sits at the highest rung it has evidence for; a missing rung below is named. One entry per announcement, not per country. Source: From Announced to Serving Customers, data as of 24 Oct 2026, which links each rung's evidence.

Headline-target sum: about 970MW with Nexus's April 500MW, or ~650MW with its October 180MW; about 820MW of it sits on rungs 1 and 2. The two facilities serving customers haven't published a megawatt figure.

4. National AI strategy status#

Published / adopted (16+): Mauritius (2018 — Africa's first) · Rwanda (2022/23) · Benin (2023) · Senegal (2023) · Côte d'Ivoire (SNIA 2030, 2024) · Ethiopia (2024) · Egypt (2025–30) · Kenya (2025–30) · Zambia (2024–26) · Namibia (2025) · Zimbabwe (2026–30) · Nigeria · Tunisia · Algeria · Ghana · Lesotho

Drafting / under development: DRC (Oct 2025) · Gabon · Congo-Brazzaville · Togo · Uganda · Tanzania (draft 2025) · Mozambique (Dec 2025) · Sierra Leone (2026) · Chad · Botswana

No confirmed standalone strategy: Malawi · Madagascar · Mali · Angola (folded into digital strategy) · Sudan · Libya

South Africa is the notable gap. The draft policy was withdrawn in late April 2026 after it was found to cite fictitious sources. The redrafted policy isn't expected before end-March 2027 (iAfrica, Oct 2026).

Continental: the AU Continental AI Strategy was endorsed July 2024 (Accra) — five focus areas, fifteen action points. A 2025 working paper found 83% of AI funding concentrated in four countries (Kenya, Nigeria, South Africa, Egypt) and governance structures "largely aspirational." It is single-authored and not peer-reviewed.

Ghana, dated targets: by December 2026, a National AI Delivery Framework with milestones to 2035 is finalised and AI capacity-building extends to 50 Digital Champions plus Chief Directors and Directors (MyJoyOnline, 7 Sep). On 1 Oct 2026 the minister announced a central technical review of all public-sector AI acquisitions: MoCDTI coordinates, NITA scopes each against interoperability, data protection, cybersecurity, sustainability and local skills transfer. Next check 22 Dec 2026.

First sub-regional convergence: on 6 July 2026, Benin, Burkina Faso, Côte d'Ivoire, Guinea, Mali and Senegal adopted shared AI governance guidelines led by Niyel (iAfrica).

5. The capital picture#

H1 2026: US$1.44bn across 146 deals vs 252 a year earlier (TechCabal Insights). One deal (Spiro, US$215m) pushed the half past prior pace. 63 M&A transactions vs 33. This is consolidation, not expansion.

Angels. ABAN: 5,000+ investors across 60+ networks in 37 countries; US$35m into 1,200+ startups over a decade (TechCabal). In 2025, responding groups deployed US$4.4m — with over 90% of individual angels writing cheques below US$25,000, up from 76% in 2024 (ABAN 2025 survey). More deals, smaller cheques.

The structural warning. Partech observed a shift toward later-stage deployment while early-stage stays constrained; Norrsken22's Lexi Novitske said secondaries "could be something that we have to look at" (Semafor). Launch Africa: "If nobody writes that cheque in 2026, there's no Series A class in 2029" (TechCabal).

The benchmark exit: InstaDeep → BioNTech, 2023. Upfront ~£362m cash and shares plus up to ~£200m milestones; Bloomberg valued it at as much as £562m (US$684m) (Bloomberg). Still the only nine-figure African AI exit.

6. The gender-capital gap#

Sources: Africa: The Big Deal, Jan 2026; TechCabal Insights; Disrupt Africa, Jul 2026.

What holds: representation is climbing (19.2% with a female co-founder) while the capital share falls (2.2% to female CEOs in 2025). The problem is depth of cheque, not presence of founders. Correctives: Janngo Capital (Fatoumata Bâ); Madica, which set out to build a portfolio with at least 50% gender diversity in leadership teams (Madica).

7. The Unresolved Register#

Both numbers stay. Each row carries the reason they diverge.

Unverifiable — flagged, not published as fact#

• Cassava Cape Town "no capacity left to sell" — one outlet's account (Cape Town ETC) of a CEO interview on CNN's Connecting Africa; the CNN original not found; no customer named. Carried as a company claim.
• Africa AI Fund (US$60bn) — no legal vehicle, capital call or first close found. Smart Africa's US$18.7bn (Accra, Sep 2026) is unbroken-down; see §7
• HAVAIC US$50m 2026 fund close — single low-quality aggregator source
• "Africa AI Summit 2026" (Cape Town, 3–5 Nov) — the organiser listed Huang, Nadella, Pichai, Altman as invited. The dates have passed, and public reporting reviewed as of 13 Nov 2026 doesn't describe the event taking place
• ~98% of African data processed off-continent — widely repeated; public reporting reviewed as of 13 Nov 2026 traces it to no study
• DRC US$8.7bn digital infrastructure — flagged by the reporting source itself
• Sunbird "FarmPredict" reaching 2m farmers — uncorroborated listicle claim
• Cameroon, Gabon, Congo-Brazzaville, Chad, CAR, Equatorial Guinea, Burundi, South Sudan, Somalia, Eritrea, Guinea-Bissau, Sierra Leone, Liberia, Gambia, Lesotho, Eswatini, Comoros, Djibouti — no verifiable AI-specific investor, fund or accelerator with deployed capital found. Programme participation ≠ investment.

8. The counterweight — what the optimism has to answer#

The IMF's estimate (Unlocking the Potential: AI in Sub-Saharan Africa, 21 Jul 2026):

Source: IMF; summary via IMF news.

Lead author Martin Schindler called the current-conditions figure "little more than a rounding error." Binding constraints: electricity (nearly half the region lacks reliable access), connectivity (38% internet usage in 2024), skills (fewer than a quarter of higher-ed students in STEM; ~9% tertiary enrolment). Sub-Saharan Africa ranks lowest on the IMF's AI Preparedness Index.

The dependency critique. Rest of World (May 2026) documented how far African open-source AI initiatives depend on Meta grants and Google Cloud. On the Anthropic–Rwanda MoU specifically, Ayantola Alayande (Global Center on AI Governance) called it "about adoption, really, rather than building local capacity" and "a nice way to put market capture", warning of vendor lock-in (Devex, Mar 2026).

The deployment critique. Stephen Deng (DFS Lab), responding to the IMF paper: calling for a revolution in long-tail AI use while naming venture capital as a core funder, without addressing ability and willingness to pay, "is simply promoting the more efficient distribution of fragility" (Deng on X; TechCabal Insights).

9. Origins — the history behind the numbers#

Everything above is 2025–26 data. None of it started then. Full account: The Foundations of African AI.

Source: The Foundations of African AI, which links each event.

Why it's in the Ledger: none of the capital, compute or startup activity tracked above is a green-field event. It's the newest layer on three decades of mostly self-funded infrastructure that predates the 2025 pledge cycle by up to thirty-one years (SAT-3/WASC, 2001, vs. the Kigali declaration, 2025).