The 2025 KPMG Africa CEO Outlook surveyed more than 130 chief executives across Southern, East and West Africa between August and September 2025, drawn from a global sample of 1,350 across 11 markets. Its headline finding is not that African firms are behind on AI. It is that they are ahead on commitment and behind on return, and the gap between those two facts is where most of the wasted money will go.
What African boards have already decided
Seventy-one per cent of African CEOs are investing in AI to improve efficiency, competitiveness and long-term resilience. Twenty-six per cent plan to allocate more than a fifth of their annual budget to AI, against a global average of 14 per cent. The allocation argument is over.
The regional variation is worth knowing, though the sub-samples are small. West African CEOs are the most likely to treat AI as a growth engine, at 65 per cent, against 40 per cent in East Africa and 38 per cent in Southern Africa. But in all three regions the question has moved from whether to invest to where the investment should go and in what order.
The gap that matters is size, not geography
A 2025 WTO and International Chamber of Commerce survey of 158 firms found that 41 per cent of small firms report using AI, against more than 60 per cent of large firms. By economy, 66 per cent of firms in high-income countries have adopted AI, against 27 per cent in low and lower-middle-income economies. The sample is small and the precision is limited, but the direction is consistent with everything else in the literature.
A mid-sized African firm is not primarily behind its African peers. It is behind large firms everywhere, because large firms absorb the fixed cost of integration across a bigger revenue base. That is a sequencing problem, not an ambition problem.
Global organisational AI adoption reached 88 per cent in 2025, up from 78 per cent in 2024 and 55 per cent in 2023, according to the Stanford HAI AI Index. But “everyone is using it” and “anyone is getting paid for it” are different claims.
Where the money returns, and where it stalls
McKinsey’s 2025 global survey found the same 88 per cent adoption rate. But only 39 per cent of organisations report any enterprise-level EBIT impact, and most of those attribute less than 5 per cent of EBIT to AI. Roughly 6 per cent of organisations qualify as high performers clearing 5 per cent. Adoption is near-universal; measured return is not.
The functions that pay first are the ones where volume is high, the output is already structured, and someone already owns the error rate: support triage and ticket routing, document and invoice processing, reconciliation, receivables ageing. Cash-flow forecasting and demand planning follow, but they require cleaner underlying data and a named owner of the forecast, which most mid-sized firms do not yet have.
McKinsey separately estimates that generative AI could unlock $61 billion to $103 billion in annual economic value across African markets, with more than 40 per cent of African institutions already experimenting or implementing significant solutions. The range is wide because the underlying infrastructure varies as much within Africa as between Africa and anywhere else.
What it actually costs a firm of 50 to 500 people
The published seat price for Microsoft 365 Copilot is USD 21 per user per month, discounted to USD 18 paid yearly through 30 September 2026. A 200-person firm licensing 40 people, a reasonable starting scope, pays roughly USD 8,600 to USD 10,000 a year. That is a rounding error against most mid-sized firms’ payroll.
Inference cost is collapsing. The cost of running a system at GPT-3.5 level fell 280-fold between November 2022 and October 2024, according to Stanford HAI’s 2025 AI Index. The licence and the compute are no longer the variables that decide anything.
The budget line that matters is the data work and the process ownership underneath, which is exactly what 96 per cent of African CEOs already identify as their constraint. The spending order follows: fix the data layer, then buy the seats.
For Mauritian readers, the Budget 2025-2026 provides start-ups and MSMEs with tax deductions on AI investments of up to Rs 150,000, roughly USD 3,300, enough for the licences and a meaningful pilot, though not for the integration work itself. The ICT sector contributed 5.6 per cent of GDP in 2024 and employed 34,500 people.
The constraints you cannot buy your way out of
Power. 72 per cent of firms in Sub-Saharan Africa experience electrical outages, according to World Bank Enterprise Surveys. Country readings range from 34 per cent in Tanzania to 92 per cent in South Africa, with survey years varying between 2020 and 2025. This is not background context. It is a design input that argues for asynchronous, batch-tolerant workloads over anything requiring always-on interactive availability.
Cloud geography. Every hyperscaler cloud region on the African continent sits in South Africa: Microsoft Azure’s sole African geography (South Africa North and South Africa West), AWS’s only African region (Cape Town, opened April 2020), and Google Cloud’s only African region (Johannesburg, opened January 2024). For a firm in Nairobi, Lagos or Port Louis, “the cloud” means a round trip to South Africa, with the latency, foreign-currency billing and data-residency implications that follow.
Skills. African CEOs describe a pricing and retention problem, not a pipeline problem. Only 15 per cent report generational gaps in critical future skills, against 30 per cent globally, and 23 per cent name competition with global technology firms and salary expectations as the main barrier to attracting AI talent. Stanford separately records South Africa among the three fastest-growing markets for AI engineering skills, alongside the UAE and Chile. The implication is to build internal capability rather than to recruit scarce specialists at global rates, which is consistent with the 67 per cent of African CEOs already redeploying staff into AI-enabled roles and the 88 per cent who expect headcount to increase.
A sequencing rule for the next four quarters
Thirty-two per cent of African CEOs name integrating AI into core operations as their top operational pressure. The difficulty is known. What follows from the evidence is an order of operations, not a list of initiatives.
First, instrument the data layer. If the firm cannot produce a clean customer, invoice or transaction record on demand, nothing built on top of it will return measurable value.
Second, pick one high-frequency process with a measurable error rate. Not the most ambitious use case, but the one where volume and structure make the return calculable.
Third, assign a named process owner, not a technology owner. The discipline of moving from a pilot to a measurable production deployment, which we examined in moving agentic AI from pilot to production, is an operational discipline before it is a technical one.
Fourth, licence the seats, connect the tools, and expand from the first process that works.
Key takeaways
- 71 per cent of African CEOs are investing in AI, and 26 per cent are allocating more than a fifth of their annual budget, nearly double the 14 per cent global average. The allocation argument is settled.
- Global AI adoption reached 88 per cent in 2025, but only 39 per cent of organisations report any enterprise EBIT impact and roughly 6 per cent clear 5 per cent. Adoption is near-universal; measured return is not.
- The gap that matters most is firm size, not geography: 41 per cent of small firms use AI against more than 60 per cent of large firms.
- Licensing costs roughly USD 8,600 to USD 10,000 a year for 40 seats. Inference cost fell 280-fold in two years. The expensive part is data and process work, not tools.
- 96 per cent of African CEOs cite data readiness as a challenge, the strongest signal of where the first investment should go.
- Every hyperscaler cloud region on the continent is in South Africa. 72 per cent of Sub-Saharan firms experience electrical outages. These constraints shape what can be deployed, not whether to deploy.
Frequently asked questions
How much are African firms spending on AI?
26 per cent of African CEOs plan to allocate more than 20 per cent of their annual budget to AI, nearly double the 14 per cent global average, according to the 2025 KPMG Africa CEO Outlook (130+ CEOs surveyed August to September 2025). The survey does not report a median spend figure. Published seat pricing for enterprise AI tools such as Microsoft 365 Copilot is USD 18 to 21 per user per month, making licensing a minor line item relative to the data and process work that determines whether the investment returns value.
Which business functions return value from AI first?
McKinsey’s 2025 survey found that only 39 per cent of organisations report enterprise-level EBIT impact from AI, despite 88 per cent adoption. The functions that pay first share three characteristics: high transaction volume, already-structured output, and an existing owner of the error rate. In practice this means support triage, document and invoice processing, reconciliation, and receivables ageing. Cash-flow forecasting and demand planning also show returns but require cleaner underlying data and a named process owner.
What should a 200-person firm expect to spend in year one?
Licensing 40 users on Microsoft 365 Copilot at current rates costs roughly USD 8,600 to USD 10,000 a year. Inference costs have fallen 280-fold in two years and are no longer a material variable. The larger cost is data preparation: structuring records, cleaning customer and transaction data, and assigning process ownership. 96 per cent of African CEOs cite data readiness as a challenge, which suggests that for most firms the data work will exceed the tooling cost in year one.
Is Africa behind on AI talent?
The picture is more nuanced than the usual narrative. Only 15 per cent of African CEOs report generational gaps in critical future skills, against 30 per cent globally, and 23 per cent name competition with global technology firms and salary expectations as the main barrier. Stanford HAI records South Africa among the three fastest-growing markets for AI engineering skills. The constraint is retention and pricing, not pipeline: 67 per cent of African CEOs are already redeploying staff into AI-enabled roles and 88 per cent expect headcount to increase.
What infrastructure constraints affect AI deployment in Africa?
Two constraints shape what can be deployed. First, power: 72 per cent of firms in Sub-Saharan Africa experience electrical outages, according to World Bank Enterprise Surveys (survey years 2020 to 2025 depending on country). This favours asynchronous, batch-tolerant AI workloads over always-on interactive services. Second, cloud geography: every hyperscaler region on the continent (AWS, Azure, Google Cloud) is located in South Africa, meaning firms elsewhere route through Johannesburg or Cape Town with corresponding latency and data-residency considerations.
Sources
- KPMG, 2025 KPMG Africa CEO Outlook, fielded 5 August to 10 September 2025. kpmg.com/za (press release) and kpmg.com/ng (insights)
- WTO and International Chamber of Commerce, WTO-ICC business survey on AI use in trade, 11 December 2025. wto.org
- Stanford Institute for Human-Centered Artificial Intelligence, 2026 AI Index Report. hai.stanford.edu
- Stanford Institute for Human-Centered Artificial Intelligence, 2025 AI Index Report. hai.stanford.edu
- McKinsey and Company, The state of AI in 2025: agents, innovation, and transformation, November 2025. mckinsey.com
- McKinsey and Company, Leading, not lagging: Africa’s gen AI opportunity, May 2025. mckinsey.com
- Microsoft, Microsoft 365 Copilot business pricing (South Africa storefront, USD excluding VAT), retrieved 28 July 2026. microsoft.com
- World Bank Enterprise Surveys, firms experiencing electrical outages (indicator IC.ELC.OUTG.ZS), dataset last updated 13 July 2026. data.worldbank.org
- Economic Development Board Mauritius, Budget 2025-2026: Artificial Intelligence, 9 June 2025. edbmauritius.org
