Who Captures Value and Who Does Not

Not all African ventures benefit equally from AI

MIT Sloan's December 2025 analysis is direct: while the world's superpowers treat AI as a matter of national security and industrial policy, many African actors are still treating it as a side project. Optimism that masks this structural reality is dangerous.

The most consequential framing number comes from Partech's January 2026 Africa Tech Venture Capital Report: AI captured 50 percent of global VC dollars in 2025, with 60 percent of that capital deployed in $100M-plus rounds; the equivalent megadeal share in Africa's equity market is 15 percent. OpenAI and Anthropic's combined raises - $40bn and $16bn respectively - alone captured 14 percent of total global VC. Partech tracks African tech funding at $4.1 billion in 2025 (including debt) against Disrupt Africa's $1.64 billion (equity-only) and Africa: The Big Deal's $3.2 billion-plus (disclosed deals above $100k) - the trackers diverge by methodology rather than underlying reality.

Africa is not missing the AI wave through lack of interest. The continent is structurally misaligned with the specifics of AI investing, which concentrates in frontier-model capex, data centre infrastructure, and the US and Chinese talent clusters that produce foundation models - rather than the applied-AI, product-and-distribution ventures that characterise African scaling. Of the cumulative $1.25 billion raised by African AI startups from 2019 through Q1 2025, 87 percent concentrated in the Big Four - Nigeria, Kenya, South Africa, and Egypt.

The most consequential decisions about Africa's AI future are being made by hyperscalers, sovereign wealth funds, and development finance institutions allocating compute infrastructure, data centre capacity and model training resources. Cassava Technologies' partnership with NVIDIA is a multi-year commitment of up to $720 million to deploy GPU infrastructure across five African markets. IFC's $100 million debt facility to Raxio Group in April 2025 - its largest-ever investment in African digital infrastructure - supports expansion across seven countries. Microsoft's ZAR 5.4 billion commitment expands cloud and AI infrastructure in South Africa. The African data centre market has approximately 360 MW of active capacity against a 1.2 GW pipeline that the Africa Data Centres Association projects will leave Africa's share of global capacity broadly unchanged as hyperscale expansion accelerates elsewhere.

Infrastructure decisions made in the next 24 to 36 months will determine whether African ventures participate in the AI economy as builders of AI-enabled products designed for African markets or as consumers of tools built for other markets. The builder position generates intellectual property, technical capability, and economic value that compounds within the African ecosystem. The consumer position generates dependency that compounds outside it. The Africa Declaration on Artificial Intelligence, adopted in Kigali on 4 April 2025 and endorsed by all 55 AU member states, proposes a $60 billion Africa AI Fund. No capital has been committed, no fund manager announced, no anchor LPs publicly named. The declaration is a statement of political intent.

Current conditions are unfavourable. Africa holds approximately 0.6 percent of global data centre capacity against 18 percent of the world's population, per the Africa Data Centres Association. African innovators face 7 million GPU hours of unmet compute demand over the next three years, per WEF December 2025 analysis. The binding constraint is power: grid instability, high energy costs, and the absence of reliable transmission infrastructure undermine the value of physical compute investment.

Four fault lines determine which African ventures sit on the builder side of the divide. Each is individually documented. Their interaction determines the venture's position.

Fault line 1: Data

The most durable AI-enabled competitive advantages come from proprietary data that reflects the specific characteristics of the market being served. Data architecture is a founding-stage decision, not a growth-stage one. The collapse of 54gene - Nigeria's flagship genomics venture, which raised $45 million at a peak $170 million valuation and amassed 100,000-plus Nigerian genomes across 300 ethnic groups before shutting down in September 2023 - illustrates the inverse. In July 2025 founder Abasi Ene-Obong filed a Federal High Court petition alleging that lead investors Cathay AfricInvest Innovation Fund and Adjuvant Capital had forced the company into bankruptcy after rejecting a $110 million rescue package, and were preparing to sell the biobank asset for $3 million. A Lagos Federal High Court injunction in August 2025 blocked the sale that would have placed 100,000 Nigerians' genetic data "in the hands of the highest bidder." Proprietary data produces a moat only when the venture captures the long-run economic value of it. The extractive pressure on African data assets originates from investor boards as readily as from foreign acquirers.

Fault line 2: Talent

Africa accounts for three percent of the global AI talent pool. Only 31 percent of African universities offer dedicated AI programmes, and AI represents 1.5 percent of specialised digital enrolments. Approximately 70,000 skilled Africans emigrate annually, representing a $2 billion loss (AUDA-NEPAD). Large technology multinationals compete for the talent that remains at compensation levels African scaling ventures cannot match.

Fault line 3: Compute

The divide is tiered and structural. The first tier - ventures with access to international compute through cloud partnerships or direct infrastructure relationships - can build and train models at the frontier. The second tier - ventures working with inference-only access to pre-trained models through standard API connections - can deploy AI applications but cannot build the proprietary AI assets that generate the most durable competitive advantages. The divide does not map onto venture quality or founder capability. It maps onto structural access to compute. The Cassava-NVIDIA deployment, Microsoft's South African investment, and IFC's Raxio facility begin to address the capacity gap. The power constraint remains.

Fault line 4: Language and cultural context

Global AI models perform materially worse on African languages, in African cultural contexts, and in informal market conditions. Masakhane's January 2026 Request for Proposals, supported by Google.org, FCDO, IDRC and the Gates Foundation, funds work on 50 African languages - building on a 2025 call that received 93 applications from 22 countries. The goal is empowering one billion Africans with locally relevant AI tools by 2029. Ventures that build AI systems genuinely calibrated to African linguistic and cultural contexts hold competitive advantages international providers cannot easily replicate.

The four fault lines interact multiplicatively. A venture with a strong data asset but no compute access ends up structurally consumer regardless of how much proprietary data it has collected - which is the position most African scaling ventures currently occupy. Structural weakness on any one of the four constrains the value the other three can produce.

The value-capture question operates with particular intensity in AI. The most consequential African AI ventures are headquartered offshore: Cassava Technologies in the UK, InstaDeep in London at the time of its $682 million BioNTech acquisition, Raxio in the Netherlands, Kera Health and Intron Health in Delaware. The exit value, ongoing IP, and tax base of these ventures accrue overwhelmingly outside African jurisdictions. The structural tension with the Kigali Declaration's call for data sovereignty is clear: the prevailing venture capital stack requires offshore incorporation. Lelapa AI, still incorporated in Johannesburg, is a rare exception.