By: Sir Roger Jantio, Senior Managing Director & CEO of Sterling Merchant Finance Ltd
Africa’s optimal AI path lies in application development and strategic use of global computing resources rather than replicating expensive data center buildouts. India’s emergence as an AI superpower through its massive user base and application ecosystem, rather than infrastructure ownership, validates this leapfrogging approach. The infrastructure debate should be considered settled.
The question now shifts from whether to build infrastructure to how to build applications. Small Language Models (SLMs) represent Africa’s most practical entry point: sector-focused AI systems that leverage the technical shift away from massive models while addressing uniquely African challenges in agriculture, healthcare, finance, and education. Beyond these domains, SLMs also create space for Africa to digitize and safeguard cultural heritage, preserve indigenous knowledge, and support sustainable approaches to nature and environmental stewardship — areas that are equally central to the continent’s transformation.
This article provides a concrete framework for developing African SLMs across three dimensions: technical requirements that minimize infrastructure costs, data strategies that balance sovereignty with utility, and business models that enable development without government infrastructure spending.
Technical Requirements: What SLMs Actually Need
The shift from large language models to small, specialized ones fundamentally changes infrastructure economics. Understanding what SLMs actually require—from computing power to deployment infrastructure—reveals that Africa can build sophisticated AI applications without nationally-owned data centers.
The Computing Reality Behind Small Language Models
A 9-billion-parameter model from Nvidia recently outperformed models 40 times larger on specific tasks. This represents an order of magnitude difference in required computing resources, transforming what’s possible for resource-constrained markets.
Consider a practical example: training a specialized agricultural advisory model for African crop patterns requires computing resources available through cloud services for thousands of dollars, not the millions needed for comprehensive general-purpose models. Fine-tuning an existing model to understand Swahili farming terminology or West African climate patterns requires even less, often achievable on university computing clusters or modest cloud allocations.
Leveraging Existing Infrastructure Without Sovereignty Requirements
African SLM development does not require nationally-owned data centers. Instead, it can operate effectively through already available computing layers. Global cloud providers like AWS, Google Cloud, and Azure compete aggressively for African markets, offering educational credits and startup programs. These platforms handle the heavy computational work of initial training. Regional resources—university computing clusters, private sector data centers, or commercial cloud edge nodes—can then adapt these models to local contexts. Finally, modern smartphones and basic servers run the trained models for actual use, making sophisticated AI accessible without massive infrastructure investment.
Transfer Learning as the Strategic Advantage
Transfer learning makes this approach particularly powerful. Rather than building models from scratch, African developers can start with models trained on global agricultural data and adapt them with African-specific information: local crop varieties, regional climate patterns, indigenous farming practices, and traditional knowledge systems. This leverages billions in investment dollars in foundational model development while focusing African resources on the contextual adaptation that creates unique value.
This technical reality contradicts conventional wisdom about infrastructure requirements. M-Pesa transformed African finance without requiring Visa’s global infrastructure—it adapted existing mobile networks to solve African payment challenges. SLM development follows the same principle: adapt existing AI infrastructure to address African needs rather than replicating global infrastructure buildouts.
Data Sovereignty: Balancing Control and Utility
Data represents both Africa’s greatest AI opportunity and its most contentious challenge. The continent generates vast information from mobile payment transactions, agricultural sensors, healthcare systems, and social platforms. Yet legitimate concerns about data extraction often slow progress, creating a perceived trade-off between sovereignty and utility.
Reframing Sovereignty: Control Over Use, Not Location
True data sovereignty isn’t about where servers are physically located. It’s about who controls how data gets used, who benefits from insights generated, and what protections exist against exploitation by foreign companies or unfair data extraction practices.
The question isn’t whether to use data for AI development—it’s how to ensure Africans benefit. An agricultural SLM trained partly on global crop data but delivering valuable recommendations to African farmers represents successful value capture, even if some processing happens offshore. What matters is the governance framework, not the physical location.
Technical Approaches Enabling Sovereignty Without Infrastructure
Several technical approaches enable data sovereignty without requiring sovereign infrastructure. Federated learning allows models to train across distributed datasets without centralizing sensitive information. A healthcare SLM could learn from patient data across hospitals in Kenya, Nigeria, and Ghana without that data ever leaving local systems.
Differential privacy techniques, increasingly standard in AI development, protect individual privacy while enabling model training. African institutions can contribute data to training processes while preventing the extraction of individual records.
Data cooperatives and trusts provide governance structures that enable contributors to maintain control. African farmers contributing agricultural data through cooperatives can mandate that resulting SLMs remain affordable and accessible to contributing communities. Ghana’s Farmers’ Information Network, for example, could contribute planting and harvest data to train agricultural SLMs while ensuring member farmers receive free or subsidized access to the resulting tools.
Language and Cultural Representation Without Massive Investment
African languages and cultural contexts present both challenges and opportunities. African languages represent less than 0.1% of internet content, creating fundamental bias in AI systems. Yet SLM development offers a path to correction without massive infrastructure investment. Rather than attempting comprehensive language models for all African languages simultaneously, focused approaches deliver immediate value.
Quality matters more than volume for SLMs. A few thousand high-quality examples in specific domains—agricultural extension conversations in Yoruba, medical consultations in Amharic, financial advisory interactions in Zulu—can produce effective specialized models. University linguistics departments, cultural organizations, and sector specialists can contribute domain expertise that produces superior models compared to generic approaches.
Business Models That Work Without Government Infrastructure Spending
Sustainable SLM development requires funding mechanisms independent of government infrastructure investment. Multiple proven models demonstrate viability, drawing lessons from Africa’s existing technology success stories. This section examines how private sector models, development finance mechanisms, research institution approaches, and strategic government enablement can fund SLM development without direct infrastructure spending.
Private Sector-Led Development Models
African fintech companies already show how to develop and monetize sophisticated digital services without infrastructure ownership. Flutterwave processes billions in transactions without owning payment networks. Paystack serves hundreds of thousands of businesses without building banking infrastructure. These companies leverage existing systems while creating distinctive value through applications. SLM development follows the same pattern.
Consider agricultural advisory platforms using SLMs to deliver personalized farming guidance. Such platforms generate subscription revenue from farmers and development organizations while scaling across markets. A service that helps smallholder farmers optimize planting schedules, identify crop diseases through phone photos, and access market pricing information could charge modest fees—perhaps $2-5 monthly—that farmers recoup through improved yields. Development organizations and agribusiness companies might sponsor access for specific farmer cooperatives, creating blended revenue streams.
Mobile operators seeking revenue beyond basic connectivity offer another proven channel. Safaricom’s success with M-Pesa demonstrates how telecommunications companies can monetize value-added services. Conversational AI assistants in local languages, voice-activated agricultural information services, and automated customer support create new revenue streams while providing social value.
Development Finance and Impact Investment Mechanisms
Development finance and impact investment provide crucial early-stage support. Healthcare diagnostic tools might receive development grants while generating subscription revenue from clinics and hospitals. Blended finance structures—combining philanthropic capital with commercial investment—enable development of SLMs with social impact alongside commercial viability.
Challenge funds focused on demonstrated capability rather than promised infrastructure, redirect development resources effectively. Rather than funding data centers, institutions like the African Development Bank could sponsor competitions for best-performing agricultural SLMs or healthcare diagnostic models.
University and Research Institution Approaches
African universities developing SLMs as open-source projects can attract research funding while building institutional capacity. The University of Dar es Salaam, for example, could develop a Swahili educational SLM as an open-source project, attracting funding from education-focused foundations while training the next generation of AI researchers.
Regional collaboration amplifies limited resources. A consortium of agricultural universities across East Africa could pool resources to develop crop advisory SLMs, sharing development costs while adapting models for different national contexts.
Intellectual Property and Benefit-Sharing Frameworks
Intellectual property frameworks shape whether private investment can coexist with broad access. Tiered licensing models allow SLMs developed with public funding or community data to provide free or low-cost access for African users while permitting commercial licensing for external markets.
Community benefit agreements ensure value flows to data contributors. Developers using farmer cooperative data might commit to free access for contributing members, local hiring for support roles, and reinvestment of a portion of revenues into agricultural extension services.
Government Enablement Without Infrastructure Spending
Government roles in enabling SLM development don’t require infrastructure spending but do demand strategic action. Regulatory clarity on data use, AI deployment, and liability reduces investment risk and enables private sector activity. Making government-collected data available for SLM development provides valuable training data without cost. Investing in AI literacy and technical skills creates the talent pool for SLM development. Most importantly, the government’s willingness to procure AI services rather than requiring infrastructure ownership creates immediate markets for private developers.
Rwanda’s approach to drone delivery services offers an instructive parallel. Rather than demanding that delivery companies build local manufacturing facilities, the government focused on regulatory frameworks that enabled rapid deployment of services. Zipline began delivering medical supplies within months, creating immediate health benefits without waiting for infrastructure buildouts. SLM development can follow similar logic.
Implementation Pathways
African institutions ready to pursue SLM development should consider phased approaches that deliver value quickly while building toward larger ambitions.
- First six months: identify high-priority sectors where SLMs offer clear value, map existing data sources, engage potential partners, and define success metrics. This requires modest investment—primarily staff time and strategic planning.
- Next twelve months: pilot SLM projects in priority sectors using cloud infrastructure, establish data governance frameworks, test business models, and build technical capacity. This phase involves real expenditure but focuses on learning and validation rather than large-scale deployment or premature scaling.
- Eighteen to thirty-six months: expand successful approaches across markets and sectors, refine business models, establish regional collaboration mechanisms, and position African SLMs for global markets. By this point, successful models should be generating revenue and attracting commercial investment.
From Strategy to Application
The infrastructure debate established that Africa should leverage global computing investments rather than replicate expensive data centers. SLM development provides the concrete pathway from strategy to implementation.
Technical requirements are modest—accessible through existing cloud services and regional computing resources. Data sovereignty concerns can be addressed through governance frameworks rather than infrastructure ownership. Business models combining private investment, development finance, and research collaboration enable sustainable development without government infrastructure spending.
The opportunity is immediate. As global AI shifts toward smaller, specialized models, African developers can build applications addressing local challenges while contributing to global AI diversity. Success requires focusing resources on application development, strategic partnerships, and enabling policies rather than infrastructure replication.
Africa’s demographic advantages, unique challenges, and growing technical capacity position the continent to follow India’s trajectory—becoming an AI leader through user base and application innovation rather than infrastructure ownership. The question isn’t whether Africa can succeed in AI, but whether leaders will choose the efficient path of application development over the costly trap of infrastructure replication.
The tools are available. The business models work. The question is whether Africa’s leaders will seize the moment.
Sir Roger Jantio is a seasoned finance expert serving as Senior Managing Director & CEO of Sterling Merchant Finance Ltd in Washington, DC, Chairman and Managing Partner of Condona Capital LLC, President of Sterling Africa Growth Fund, and President of African Overseas Private Investment Corporation. With extensive experience in public-private partnership (PPP) structuring and project finance across emerging markets, Roger has led over 100 global projects in sectors like banking, insurance, telecom, airports, ports, power, energy, and infrastructure. His expertise spans limited recourse project financing, financial modeling, risk analysis, and leading multidisciplinary teams in bid strategy, capital structuring, and due diligence. Specializing in infrastructure, renewable energy, oil & gas, mining, and more, Roger is a trusted leader in driving innovative financial solutions for complex projects.
