Why Africa’s SMEs need access to the infrastructure behind AI
- Moinak Maiti
While policymakers celebrate Africa’s booming AI adoption in mobile apps and fintech, small businesses tend to remain locked out of high-value innovation.
While policymakers celebrate Africa’s booming Artificial Intelligence (AI) adoption in mobile apps and fintech, small businesses tend to remain locked out of high-value innovation. This is not for a lack of entrepreneurial skill, but because the foundations are structurally hard to access for them.
Energy, semiconductors, and cloud infrastructure are part of the infrastructure that needs to become accessible for them to become untrapped from the bottom of the global digital economy, often at the low-value consumer tier. Policy foci have to, as argued in this piece, shift from a narrow framing of "digital skill" towards a layer-based, AI infrastructure-aware regulatory approach. A closer look at AI dissected into its five-layer infrastructure (Energy, Chips, Digital Infrastructure, Models, Applications) shows what must change to turn African SMEs from consumers into innovators.
From AI use to AI participation
The strength of Africa’s economy will depend on how effectively its SMEs use digital infrastructure. AI specifically is increasingly promoted by policy makers as a way to boost SME productivity, lower barriers to entry, spur innovation, and help small businesses compete in global markets. However, this straightforward development narrative is prone to conceal a deeper structural inequality underlying different layers of the AI systems, resulting in the limited participation of African SMEs in AI innovation.
Often, AI is portrayed as a standalone technology and a lack of its uptake as a general skills issue. However, to understand the barriers to SME’s utilising these technologies more frequently and effectively, it is worth looking at the five different layers AI comprises: Energy, Chips (processing and storage), Digital Infrastructures (hardware and cloud), Models (training and development), and Applications (user interfaces). These layers are interdependent, and unequal access at the foundational levels leads to possible constraints at the higher levels.
SMEs operating on the African continent are largely confined to the application layer of AI, while access to the underlying layers where much of the value and control are concentrated remains limited. This structural positioning reflects policy choices that prioritise simplistic notions of the digital divide, focused on connectivity and skills and usage, while neglecting the third-level divide: the ability to understand, access and participate across AI infrastructure layers. This results in the integration of the African SMEs into global AI ecosystems as consumers, rather than producers.
Making sense of AI inequality and policy failures across the AI layers
Looking at each AI layer and how they interconnect affords a more nuanced understanding of just how infrastructural inequality operates for African SMEs, and why conventional policy approaches have been inadequate to date.
Layer 1: Energy — Or the foundation of AI
Without electricity as a basic foundational requirement, the most advanced AI systems cannot operate. AI systems consume a vast amount of electricity, yet, the need for a reliable and uninterrupted supply of electricity, still a major challenge across many African countries, is rarely given sufficient attention in digital policy discussions.
Affordable and sustainable energy is particularly crucial for SMEs, who often lack the resources to invest in alternative energy solutions at scale. Failing to integrate energy and digital strategies, and instead focusing primarily on the technologies themselves, represents a critical oversight. AI and data centres can create enough predictable electricity demand to make new investment in generation, grids and storage financially viable, potentially expanding energy infrastructure that SMEs can also benefit from. It obscures the foundational inequalities that inform AI adoption and may amplify the risks associated with adopting AI technologies.
South Africa, for instance, has a relatively well developed electricity grid, but prolonged load shedding has demonstrated the consequences of an unreliable supply. North African countries like Egypt and Morocco are investing heavily in developing sustainable energy resources to power datacentres, while Kenya is, similarly, promoting renewable energy as part of its strategy for developing AI infrastructure. Without addressing energy as a binding constraint, the layers above it cannot function reliably.
Layer 2: Chips and storage
The global semiconductor supply chain has historically heavily relied on East Asia and North America. However, Africa is emerging as a region with significant potential and could play a vital role in diversifying geopolitically sensitive semiconductor supply chains. Semiconductors are fundamental to the chip layer underpinning AI. Advanced AI systems depend on semiconductors for processing power, data storage, computational intelligence and automation, and data security.
Currently, Africa produces a significant share of the critical minerals required for semiconductor manufacturing, including approximately 70% of the world’s cobalt from the Democratic Republic of the Congo and a substantial share of its platinum-group metals from South Africa. The continent also holds extensive reserves of graphite, manganese, and lithium.
Notwithstanding, African countries have minimal participation in semiconductor manufacturing and consequently limited bargaining power within global supply chains. Access to high-performance chips is therefore mediated through imports often at high cost from countries like Taiwan’s TSMC, US-based Nvidia and Qualcomm, and the Netherlands-based ASML. For African SMEs, this translates into indirect but significant constraints. Because chip supply is concentrated among a small number of global firms, African SMEs face higher costs, longer lead times, import dependence and limited bargaining power, making compute-intensive businesses more difficult to start and scale.
Policy discussions in Africa have largely overlooked this layer, treating access to hardware as an external constraint rather than a strategic issue. Without developing Africa’s own semiconductor strategy, comparable to the EU's European Chips Act and the US’s CHIPS and Science Act, regional cooperation, strategic procurement, or partnerships within African economies alone will not substantially reduce structural dependence on external actors.
Layer 3: Digital infrastructures
Africa faces a significant gap between the development of its digital infrastructure and its capacity to participate across the different layers of AI. The continent currently accounts for less than 1% of global data center capacity, despite being home to 18% of the world's population. Policy attention has concentrated particularly in internet connectivity, cloud computing, and data storage systems. While progress has been made in expanding connectivity, significant gaps remain in terms of quality, affordability, and distribution.
Current trends suggest that African countries are not necessarily adopting AI by upgrading old infrastructure, but by bypassing it, using AI to solve basic, immediate problems in agriculture, financial innovation, health, education, and local innovations. For instance, the "Silicon Savannah" ambition in Kenya has focused heavily on applications in agriculture, financial innovation, and health, while still heavily relying on foreign third-party cloud providers. This creates a paradox: African SMEs may have access to digital services without having access to the necessary infrastructure and resources needed to meaningfully innovate, and gain value from, AI ecosystems.
Layer 4: Models
Advanced AI models are increasingly important for forms of automation and decision-making that have the potential of serving African SMEs and publics. Developing, adapting, and deploying these models requires access to large and representative datasets, substantial computing capacity, and the infrastructure needed for storage, connectivity, and model development.
Some of the resources needed to make AI work are emerging through research institutions and laboratories in countries such as South Africa, startups providing APIs in Nigeria and Kenya, and government-backed AI labs in Rwanda and Egypt. Their convergence, however, remains largely concentrated in a small number of research institutions and laboratories. As a result, most SMEs operating in Africa rely on third-party pretrained models or APIs, limiting their ability to develop AI and ensure it serves people in local contexts.
Since global models are also often trained on data that inadequately reflects African contexts, this impacts how well they perform across languages and contexts. Existing policy frameworks have largely failed to address this AI layer, instead focusing on skills development without sufficiently tackling the structural barriers to model creation and adaptation. This is one key reason why the African SMEs remain primarily consumers of AI models, rather than contributors to their development and application.
Layer 5: Applications
The application layer is where most African SME’s AI-related activities are concentrated and where revenues are generated. This includes using AI tools for customer service, marketing, logistics, predictions, frauds, digital identities, diagnostics, and other purposes. Barriers to entry for SMEs are relatively low at this layer.
By operating primarily at the application layer, African SMEs capture only a small portion of the value generated by AI systems. This creates an illusion of inclusion in AI driven economies while masking the deeper inequalities that exist in the other layers of AI. In addition, telecommunications and internet costs remain high across the continent, affecting consumers more than producers in their adoption of AI services. From a policy perspective, this raises important questions about the goals of digital transformation. If African SMEs are confined to low-value segments of the AI economy, current approaches may reinforce rather than reduce structural inequalities.
Building an equitable AI ecosystem for African SMEs
The African Union Continental AI Strategy provides an important foundation for AI policy, governance, and capacity-building. However, with only around 22 of Africa’s 54 countries having formally published national AI strategies, there remain significant gaps to be bridged.
This is, not least of all, due to the strategy supporting AI applications and governance but not fully addressing inequalities across the AI stack. The layered approach outlined in this piece lends itself as a useful approach for African Union and national policymakers in order to strengthen energy, chips, digital infrastructure, models, and applications, enabling African SMEs to create value across the AI economy rather than remaining confined to its lowest-value segments.
Professor Moinak Maiti is the FirstRand Endowed Chair Professor in Financial Data Science, working at the intersection of academia, industry, and public policy to advance resilient and intelligent financial systems. His research spans Artificial Intelligence, Blockchain, Data Science, Digital Finance, and Financial Economics, with a focus on translating empirical insights into actionable policy. He participates in global policy platforms, including the G20 TechSprint and central bank climate dialogues. A Fellow of the Wits MIND Institute, he contributes to regulatory innovation and financial resilience. He also serves as journal editor and featured from time to time in expert interviews and international media.
This article was first published on Global Center on AI Governance.