Executive Assessment

The October 2025 baseline identified four structural findings: policy proliferation without implementation capacity; continental coordination functioning as guidance rather than governance; the resource gap as constitutive of the governance gap; and a bilateral coordination model outperforming multilateral frameworks in practice. All four findings remain valid. What has changed is the texture of evidence supporting them.

The G20 Johannesburg Summit (22–23 November 2025) delivered the “AI for Africa Initiative” as a confirmed legacy outcome — a voluntary multilateral platform for cooperation on AI infrastructure and capacity building. Nigeria finalised and revised its National AI Strategy, developed a multilingual foundational language model, and jumped 31 places in the Government AI Readiness Index. Kenya formally launched its National AI Strategy with a specific budget allocation. South Africa’s government declared data centres “critical as electricity, ports and transport networks” in its February 2026 budget statement. The Africa AI Declaration — endorsed by national ministers — announced an Africa AI Council under the Smart Africa Steering Committee and a $60 billion Africa AI Fund target.

Against these developments, the structural deficits largely hold. No functional continental AI coordination body has been established. The AU Commissioner publicly confirmed in January 2026 that 83{e31bf911d06dd91ac4b0846a01926c6e0cba1b3752e1873aecb4a21b5e07de05} of AI startup funding remains concentrated in four countries — the same figure as October. Kenya still has no AI-specific legislation in force. REC-level coordination (ECOWAS, EAC, SADC) remains fragmented. And the US boycott of the G20 Johannesburg Summit, combined with the incoming US presidency of the G20 for 2026, introduces a geopolitical variable that complicates the outlook for the AI for Africa Initiative’s continuity.

National Implementation: What Has Moved

Nigeria: From Strategy to Legislative Architecture

The October baseline assessed Nigeria as “governance-heavy” — substantial policy infrastructure without concrete funding or enforcement mechanisms. That characterisation requires partial revision.

Nigeria’s National AI Strategy was revised and published in September 2025, followed by the launch of the N-ATLAS multilingual large language model at the UN General Assembly in September 2025. N-ATLAS, developed by the National Centre for Artificial Intelligence and Robotics in collaboration with Awarri Technologies and built on Meta’s Llama-3 architecture, represents the first African-developed foundational AI model to reach public deployment at scale. It is not merely symbolic — it directly addresses the language access gap that the October baseline identified, and its development through NCAIR demonstrates that ISAR Global’s “governance-heavy” characterisation understated operational technical capacity.

More significantly, the legislative architecture is advancing. The National Digital Economy and E-Governance Bill, expected to pass by Q2 2026, would position NITDA as a “super-regulator” for the digital economy with authority to classify AI systems by risk, mandate algorithmic transparency, and accredit AI auditors. The September 2025 General Application and Implementation Directive from the Nigeria Data Protection Commission created an enforcement framework for the 2023 Data Protection Act with direct implications for AI systems. Nigeria jumped 31 places to 72nd in the 2025 Oxford Government AI Readiness Index.

The governance fragmentation problem identified in October persists, however. NITDA, the Central Bank of Nigeria, and the NDPC continue to operate overlapping mandates without a unified coordination mechanism. The January 2026 NITDA-KOICA Start-Up Digital Innovation Academy and the UK-Nigeria Enhanced Trade and Investment Partnership (UK investors contributing 65{e31bf911d06dd91ac4b0846a01926c6e0cba1b3752e1873aecb4a21b5e07de05} of new foreign capital) represent implementation activity — but activity concentrated in skills development and bilateral partnerships rather than the cross-agency coordination that would constitute genuinely functional governance.

Kenya: Formal Launch with Budget Commitment — Still No Legislation

Kenya formally launched its National AI Strategy 2025–2030 on 27 March 2025, with a specific budget allocation of KES 152 billion (approximately $1.14 billion) over five years. This is a material development: the October baseline noted “no concrete budgetary commitments identified.” A five-year budget envelope, however aspirational, represents a governance instrument absent in October.

A Draft AI Code of Practice is under public review. A Robotics and AI Bill is in development. The National AI Steering Committee, working under the Ministry of ICT and Digital Economy, has been constituted. Kenya’s Data Protection Commissioner imposed enforcement fines in late 2025, demonstrating regulatory willingness to act — though not yet on AI-specific grounds.

What remains unchanged is the absence of any AI-specific legislation in force. As of March 2026, White and Case’s regulatory tracker confirms: “there are currently no specific laws or regulations in Kenya that directly regulate AI.” The strategy commits to regulatory sandboxes and risk-based classification frameworks — but commits to developing them, not deploying them. The governance foundation is more visible than in October. It remains a foundation under construction.

Rwanda: Deepening Rather Than Plateauing

The October baseline assessed Rwanda as the continental exception — the only country with a fully operational governance structure. That assessment holds, with the added observation that Rwanda is now deepening rather than simply maintaining its position.

Rwanda launched a Digital Public Infrastructure strategy in March 2026, establishing the Centre of Digital Public Infrastructure (constituted in 2025) as the coordinating mechanism for foundational digital systems — digital identity, payment infrastructure, data exchange. The timeline includes digital identity system rollout by June 2026. The RISA Chief Executive Officer, speaking at Mobile World Congress Kigali in 2025, outlined Rwanda’s governance model as centred on four pillars — skills, infrastructure, governance, and collaboration — and positioned Rwanda explicitly as an export model for the region.

Rwanda has not yet adopted AI-specific binding legislation (the National AI Policy from April 2023 remains the primary instrument), but the DPI strategy represents institutional deepening that moves the implementation model beyond governance framework toward foundational infrastructure delivery.

South Africa: Infrastructure Momentum, Policy Still Developing

South Africa presents the most complex picture. The October baseline rated its G20 presidency as a “7/10 institutional effectiveness” exception to the general pattern. That rating now requires context.

The G20 Johannesburg Summit (November 2025) confirmed the “AI for Africa Initiative” as a key legacy outcome — a voluntary multilateral platform for cooperation on AI infrastructure, skills, and research. The Summit declaration addressed AI governance, calling for “safe, secure, and trustworthy AI development with human oversight and transparency.” India proposed hosting an AI Impact Summit in 2026. These are institutional outputs, not merely declarations.

On domestic policy, South Africa’s Draft National AI Policy is expected to be gazetted for public comment in March 2026 — the first time a specific timeline has attached to what was previously an indefinite process. The National Treasury’s February 2026 budget speech explicitly classified data infrastructure as “critical as electricity, ports and transport networks,” opening the door to accelerated depreciation, VAT relief on imported hardware, and faster grid connections for data centres. Hyperscaler commitments in South Africa now exceed $3 billion from Microsoft, AWS, and Google, with the data centre market projected to reach $5.28 billion by 2031.

The US boycott of the G20 Summit is the complicating variable. The AI for Africa Initiative is voluntary. Without US participation — and with the US holding the 2026 G20 presidency — the continuity and ambition of that initiative cannot be assumed. The Global Affairs Council notes that “there’s going to be likely some very visible pushback on some of the priorities that have been agreed upon under the South African presidency.”

Continental Layer: Structural Gaps Persist with New Institutional Signals

The October baseline’s central finding — that the AU Continental AI Strategy functions as guidance rather than operational governance — has not been fundamentally altered. No functional continental AI coordination body has been established. Phase 1 (2025–2026) implementation remains, by all available evidence, in the workshops-and-frameworks stage rather than the operational-institutions stage.

Two new developments require recording.

The Africa AI Declaration — endorsed by national ministers at a high-level dialogue — announced the establishment of the Africa AI Council under the Smart Africa Steering Committee “to ensure high-level engagement and strategic alignment with continental and global digital transformation efforts.” This is the first time a specific governing body with a defined mandate has been proposed at continental level. Whether it constitutes a functional coordination mechanism or an additional aspirational layer requires monitoring. ISAR Global’s assessment, pending further evidence, is cautious: Smart Africa has previously announced coordination mechanisms that have not achieved operational status.

The $60 billion Africa AI Fund announced in the Africa AI Declaration represents the most ambitious financing commitment yet made at continental level. The fund is described as “leveraging public, private, and philanthropic capital.” No operational mechanism — governance structure, disbursement criteria, or timeline — has been published. The comparison with the October baseline’s observation that “resource mobilisation remains aspirational without operational funding frameworks” applies with equal force to this announcement.

The AU Commissioner for Infrastructure and Energy publicly confirmed at a 2025 High-Level Policy Dialogue that 83{e31bf911d06dd91ac4b0846a01926c6e0cba1b3752e1873aecb4a21b5e07de05} of AI startup funding in Q1 2025 went to Kenya, Nigeria, South Africa, and Egypt — the same figure as the October baseline. That this figure is being cited by senior AU leadership as a problem to be addressed, rather than concealed as an embarrassment, represents a modest transparency gain. It does not represent a change in the underlying distribution pattern.

Infrastructure: Real Momentum, Concentrated Geography

The October baseline characterised Africa as holding 1.8{e31bf911d06dd91ac4b0846a01926c6e0cba1b3752e1873aecb4a21b5e07de05} of global large-scale data centres against 15{e31bf911d06dd91ac4b0846a01926c6e0cba1b3752e1873aecb4a21b5e07de05} of world population. This remains accurate in structural terms, though the trajectory is materially more active than October suggested.

The African hyperscale data centre market is forecast to surge from $6.7 billion in 2025 to over $28 billion by 2030. More than 220 data centres now operate across 38 African countries, with Morocco emerging as an unexpected hub — running over 23 facilities, supported by a 500MW renewable GPU strategy that leverages proximity to European markets. The IXAfrica campus in Nairobi entered a strategic partnership with Safaricom in 2025 to deliver AI-ready infrastructure. Cassava Technologies’ NVIDIA partnership is deploying 12,000 GPUs across its Africa AI Factory network.

The compute paradox identified in the baseline has become more visible, not less. A World Economic Forum analysis published in late 2025 notes that 7 million GPU hours of unmet demand for model training exist across Africa’s research and startup community over the next three years — while data centre operators simultaneously report under-utilised racks and slower-than-expected returns. The constraint has shifted from physical capacity to GPU access cost and uncertain workload demand. Global supply dynamics continue to favour hyperscalers over African buyers, leaving smaller actors facing steep prices and long delivery times.

The geographic concentration problem has a new dimension: the data centre boom is concentrated in North Africa (Morocco, Egypt) and South Africa. Sub-Saharan markets outside Kenya and Nigeria remain in early stages. The Africa Data Centres Association’s 2026 Economic Report confirms that outside South Africa — the continent’s most developed ecosystem — “occupancy growth remains gradual.”

The New Variable: US Disengagement and Geopolitical Uncertainty

The October 2025 baseline did not — could not — account for the degree of US disengagement from multilateral AI governance that has materialised since then. This requires explicit assessment.

President Trump did not attend the G20 Johannesburg Summit in November 2025. The Summit closed without a ceremonial handover to the incoming US presidency. The G20 troika mechanism — South Africa, the US (2026 host), and the UK (2027 host) — is under visible strain. Observers note that the US approach to the 2026 Miami Summit is expected to be “narrower and more transactional.”

For Africa’s AI governance agenda, the practical implication is that the AI for Africa Initiative, which was launched under South Africa’s G20 presidency and relies on voluntary multilateral cooperation, now faces an incoming host presidency that is unlikely to champion its continuity. The $60 billion Africa AI Fund target has no enforcement mechanism. The Africa AI Council has no operational structure. The infrastructure partnerships (Cassava-NVIDIA, Microsoft-G42-Kenya) that were framed as G20 outcomes are bilateral commercial arrangements that will proceed or not on commercial logic, largely independent of the G20 framework.

This is not a novel dynamic — the October baseline’s finding that bilateral arrangements consistently outperform multilateral coordination frameworks remains accurate. What has changed is the geopolitical framing. The multilateral scaffolding that was meant to coordinate and amplify bilateral activity is now more visibly fragile than it appeared in October.

ISAR Assessment: What Has Changed, What Holds, What Has Deteriorated

Nigeria’s N-ATLAS multilingual model represents the first African-developed foundational AI asset of genuine technical significance. The legislative architecture advancing through the National Assembly is more specific and binding than anything in the October baseline. Kenya’s budget commitment is a real governance instrument. Rwanda’s DPI strategy represents institutional deepening. South Africa’s budget reclassification of data centres creates a credible investment incentive framework. Data protection enforcement is measurably more active — 44 African countries now have data protection laws, with 38 functional DPAs. The Africa AI Declaration’s proposal of an Africa AI Council, whatever its current limitations, represents the first time a specific continental body has been named and mandated.

No functional continental AI coordination body has been established. The 83{e31bf911d06dd91ac4b0846a01926c6e0cba1b3752e1873aecb4a21b5e07de05} investment concentration in four countries is confirmed by AU leadership as a current, unresolved reality. Kenya has no AI-specific legislation in force. REC-level coordination (ECOWAS, EAC, SADC) remains systematically fragmented. The gap between continental frameworks and operational institutional capacity remains the defining characteristic of Africa’s AI governance landscape.

US disengagement from multilateral AI governance frameworks introduces a material risk to the continuity of G20-derived Africa AI commitments. The compute paradox — data centres being built while GPU access remains prohibitively expensive for most African AI practitioners — has sharpened rather than eased. Nigeria’s regulatory fragmentation, with NITDA, CBN, and NDPC operating parallel mandates, has not resolved; the incoming Digital Economy Bill may address this, or may create a fourth regulatory layer. The $60 billion Africa AI Fund and the Africa AI Council remain, as of this writing, announcements without operational architecture.

Morocco’s emergence as a significant data centre and AI infrastructure hub was not anticipated in the October baseline. Morocco now runs more data centres than South Africa by some counts, benefits from renewable energy overcapacity, and is positioned as a connectivity bridge between Africa and Europe. This is a material shift in the geography of African AI infrastructure investment — one that complicates the “four country concentration” narrative by introducing a fifth significant actor that was not in the baseline analysis.

Forward Assessment: March 2026

Three developments warrant monitoring over the coming quarter.

South Africa’s Draft National AI Policy is expected to be gazetted for public comment in March 2026. Whether the published draft represents a binding regulatory instrument or a further consultation document will determine whether South Africa advances from “policy development” to “regulatory implementation” in ISAR Global’s classification framework.

Nigeria’s National Digital Economy and E-Governance Bill is expected to pass by Q2 2026. If enacted as described, it would represent the first binding, risk-classified AI regulatory framework on the continent — a material shift from voluntary guidelines and sector-specific guidance. The fragmentation question — whether NITDA’s new authority coordinates with or duplicates CBN and NDPC mandates — will determine the legislation’s practical governance impact.

The Africa AI Council’s operational emergence, or absence, over the next six months will clarify whether the Africa AI Declaration represents a structural governance development or a high-profile diplomatic statement. The Smart Africa Steering Committee’s track record on operational delivery warrants scepticism. ISAR Global will monitor for evidence of secretariat establishment, membership formalisation, or working group activation.

Conclusion: The Gap Has Narrowed Modestly — The Structure Has Not Changed

The October 2025 baseline described a continent where governance frameworks significantly exceeded implementation capacity at every level: continental, regional, and national. Five months later, the frameworks have multiplied and several have acquired more concrete institutional expression. Nigeria and Kenya have moved perceptibly toward operational governance architecture. South Africa’s infrastructure investment environment has materially improved. Rwanda has deepened its institutional model.

The structure of the analysis, however, holds. Continental coordination remains aspirational. Investment concentration remains unchanged. The resource gap that makes governance infrastructure difficult to build has not been bridged by the announcements of funds that have not yet been capitalised. And the geopolitical context has become more, not less, uncertain.

The governance reality versus governance rhetoric gap in African AI has not closed. It has become more textured and more legible — which is itself a modest form of progress.