Artificial Intelligence (AI) is recognised as a general-purpose technology capable of compressing developmental trajectories, transcending historical constraints, and catalysing broad-based economic progress in the Global South. In Africa particularly, an increasingly prominent Promethean discourse imagines AI as possessing a uniquely elastic and scalar capacity for cross-sectoral integration, capable of generating developmental gains across healthcare, education, agriculture, finance, and public administration; the dream, in its fullest iteration, being of a continent that inherits abundance without enduring the long, grinding ordeal of industrialisation that was, in any case, never fully permitted to run its course. In their more modest expression, these visions are not unsubstantiated – cases of AI-assisted healthcare delivery in Kenya and AI-enabled education services in Rwanda, inter alia, lend credence to AI’s elasticity. Yet, given the scale of capital investment and depths of institutional capacity required for their realisation, the conditions underpinning these claims nevertheless warrant closer examination.
The proposition that technological diffusion can ‘leapfrog’ developmental stages or circumvent incumbent technologies is not new. In African developmental discourse, the era of tractors and mechanisation gave way to broadcasting and television, and the latter to mobile phones, the canonical leapfrogging success story. Yet despite the material improvements mobile connectivity has facilitated across the continent, with mobile contributing $230 billion (~8% of GDP) to Africa in 2025, its proliferation has proven largely incapable of overcoming the structural developmental constraints confronting Africans. While technology can help reduce poverty, poverty constrains technological adoption: even mobile phones, the poster child of leapfrogging, remain out of reach for much of the continent, with more than half of Africa’s population only anticipated to become mobile subscribers by 2030.
Rather than interrogating the faith in AI’s transformative potential as such, this analysis examines the material and institutional conditions under which AI leapfrogging in Africa might realise the developmental gains promised by its enthusiasts. To this end, the intellectual lineage of the leapfrogging thesis is reconstructed and its core premises applied to mobile connectivity in Africa. The extent to which conditions historically thought necessary for leapfrogging were present in the mobile case is then examined, before mobile connectivity and AI are compared as developmental technologies to assess whether the developmental realities of the former’s leapfrog meaningfully illustrate the latter. On this basis, the analysis concludes by outlining the conditions required for AI leapfrogging to approximate its developmental promise on the continent.
Credits: United Nations
Gerschenkron and the Conditions of Late Development
There is broad consensus within the literature regarding what constitutes technological leapfrogging. Steinmueller, for example, defines the process as the ‘bypassing [of] stages in capacity-building or investment though which countries were previously required to pass during economic development’. From this perspective, leapfrogging does not imply transcending development as such, but circumventing technological pathways historically associated with it. Elaborating the properties a prospective leapfrogging technology must possess, Mutiso identifies three qualifying attributes: first, the technology must already have been proven in more mature markets; second, it must offer functionality superior or comparable to incumbent technologies; and third, it must permit workarounds to obsolete or absent infrastructure. Embedded within this understanding is a distinct developmental logic: countries least committed to legacy technologies possess the greatest capacity to benefit from technological transition. It is this proposition that finds its classical formulation in Gerschenkron’s theory of economic backwardness.
Gerschenkron starts from a simple premise: the greater a country’s economic backwardness, the greater the potential speed and intensity of industrialisation made possible through adopting advanced technologies. Backwardness, in other words, does not merely constitute a condition of deprivation, but a latent developmental advantage enabling late industrialisers to bypass stages previously traversed by the industrialised. Diverging from the Promethean discourse surrounding AI, however, Gerschenkron never regarded this relationship as deterministic. The developmental gains associated with leapfrogging remained contingent upon two moderating conditions.
The first condition concerns access to external technological resources – such as foreign machinery, capital, technical expertise, and industrial know-how. Leapfrogging, after all, presupposes that late industrialisers can acquire technologies already matured elsewhere. The second concerns the domestic capacity to mobilise and effectively utilise these resources once acquired. Here, Gerschenkron placed particular emphasis on what he termed ‘institutional substitutes’ – the political, financial, and organisational arrangements capable of compensating for the institutional deficiencies characteristic of economically backward states. These substitutes, together with broader industrial potentialities (such as labour supply and skill, resource bases, and structural readiness), determine whether, and to what extent, imported technologies can be absorbed, diffused, and operationalised within the domestic economy.
Gerschenkron thus conceived leapfrogging not as deterministic, but as characterised by a persistent tension between developmental potentialities and actualities. The mere existence of an advanced technology capable of bypassing legacy infrastructure does not guarantee developmental outcomes. Indeed, even once Gerschenkron’s conditions are met, because leapfrogging technologies are often capital-intensive and labour-saving, the developmental trajectories they generate are frequently structurally uneven: capable of producing significant gains in accumulation and productivity without corresponding expansions in employment, wages, or broad-based economic agency.
Mobile Connectivity and the Lightness of Diffusion
Mobile connectivity appeared, at least initially, to approximate the Gerschenkronian promise. Due to limited fixed-line infrastructure, African economies entered the telecommunications era without facing the sunk costs, institutional inertia, and technological lock-in associated with legacy systems. In this respect, economic ‘backwardness’ assumed its developmental character: the relative absence of older infrastructures created the possibility of bypassing them altogether. While Africa had only three fixed-line subscribers per 100 inhabitants in 2003, Europe had forty-two. Yet by that year, mobile subscriptions in both continents had already surpassed fixed-line subscriptions, standing at 6.1 per 100 inhabitants in Africa and 56 in Europe. The difference was that Europe’s transition to mobile connectivity followed decades of investment in fixed-line systems, whereas mobile diffusion in Africa occurred without the prior construction of extensive legacy telecommunications infrastructure.
The absence of legacy infrastructure, moreover, had no significant stymying effect on Africa’s uptake of mobile phones. By 2003, Africa had 51.8 million mobile subscribers, reflecting a 1000 percent increase in five years. Between 2002 and 2007, mobile subscriptions increased by 49 percent annually in Africa, more than double Europe’s 17 percent annual increase. As African telecommunications markets matured, mobile phones evolved from simple communications tools into service-delivery platforms with mHealth, mEducation, and mAgriculture innovations shifting the developmental paradigm surrounding mobile phones from one that reduced communication and coordination costs to one that could transform lives.
The first of Gerschenkron’s moderating conditions – access to external technology and industrial know-how – was largely satisfied. The expansion of mobile connectivity across the continent depended upon imported handsets, foreign telecommunications capital, and substantial external infrastructural investment, particularly from multinational telecommunications firms. Simultaneously, institutional substitutes, Gerschenkron’s second moderating condition, partially compensated for the infrastructural and financial deficiencies characteristic of economically peripheral states. Telecommunications liberalisation, private-sector participation, and mobile-specific regulatory reforms enabled rapid network expansion where state-led infrastructural development remained comparatively weak. Mobile systems themselves subsequently operated as substitutes for absent formal infrastructures, particularly in finance and communications. Kenya’s M-Pesa, for example, transformed the mobile phone into an alternative financial architecture through which remittances, payments, and credit could circulate despite limited banking penetration.
Credits: Valentin Cimino – @ciminix
It is here that the conditional nature of leapfrogging becomes visible. Compared with AI, mobile connectivity represents a comparatively ‘light’ technology, capable of diffusing amid infrastructural scarcity, intermittent state capacity, limited industrial development, and a modest resource base. Yet even this relative ‘lightness’ did not overcome affordability constraints and low digital literacy, which continue to exclude large segments of the population from meaningful digital participation. Thus, despite mobile connectivity’s developmental potential in Africa, ordinary developmental constraints such as poverty and limited education continue to inhibit uptake. AI, by contrast, depends upon dense energy, data, and computational infrastructures. In Gerschenkronian terms, the resource base required for AI diffusion is substantially greater than that required for mobile telephony, raising the threshold conditions under which leapfrogging can occur.
Artificial Intelligence and the Weight of Infrastructure
To understand whether AI can function as a leapfrogging technology capable of delivering the developmental gains its proponents promise, it is necessary to consider the structures through which access to AI may be secured across Africa. AI diffusion is unlikely to occur through a politically neutral process of technology transfer. Implicit in Gerschenkron’s first moderating condition – access to external technology – is the recognition that AI’s growing political salience has caused geopolitical and strategic considerations to dominate a calculus structured around control over compute infrastructure, standards-setting, and dependence upon integrated technology stacks.
Given that access to frontier AI is subject to political processes and the strategic interests of exporting powers, three simplified pathways for Africa’s attainment of frontier AI capabilities merit consideration. In the first, a single American or Chinese frontier AI firm (e.g., OpenAI, Anthropic, or DeepSeek) exports an end-to-end AI stack to an African state or grouping of states, including models, cloud infrastructure, standards, and technical dependencies. In the second, rather than importing frontier AI itself, African states import foreign know-how and relevant technologies to build indigenous frontier models through domestic investment in compute, talent, and infrastructure. In the third, African states organise through institutions such as the African Union (AU), pooling purchasing power to negotiate coordinated access to frontier capabilities while establishing open-source and open-weight standards intended to reduce technological lock-in and enable local adaptation.
Credits: Smart Africa
Each scenario hinges on a distinct institutional substitute: foreign frontier firms in the first, the developmental state in the second, and coordinating institutions such as the AU in the third. Yet they do not produce the same developmental outcomes. Under the first scenario, outcomes are shallowest, as foreign corporations compensate for domestic deficits in capital, compute, infrastructure, and technical labour, externalising AI’s productive foundations and rendering the technology consumed rather than socially embedded within domestic production structures. Under the second, while sovereignty over frontier capabilities promises the greatest developmental gains, indigenous frontier models remain distant, with Oxford Insight’s Global AI Index placing African countries among ‘waking up’ and ‘nascent’ nations in AI investment and innovation. This leaves the third scenario which, through institutional mediation, could secure frontier capabilities without the risks of vertical and horizontal lock-in. Yet despite its strategic advantages, even this scenario is likely to confront constraints imposed by electricity, data, and connectivity.
Ironically, many leapfrogging narratives surrounding AI in Africa overlook the fact that the continent has yet to realise the full developmental potential of an earlier general-purpose technology: electricity. Around 43 percent of Africa’s population – roughly 600 million people – still lack access to power. The contradiction was stark in 2023, when South Africa’s then-Deputy Minister of Higher Education, Science and Technology, Buti Manemala, visited Lengua, one of Africa’s fastest supercomputers. The tour of Cape Town’s Centre for High Performance Computing should have showcased the infrastructure underpinning AI and machine learning. Instead, Manamela was there to assess how loadshedding – South Africa’s rolling blackouts – had affected the facility and what mitigating measures might be adopted.
Such candour from policymakers is rare. Across the growing number of AI and digital strategies emerging on the continent, particularly in sub-Saharan Africa where energy deficits are most acute, the power problem is scarcely mentioned. Smart Africa’s Artificial Intelligence for Africa Blueprint, for example, contains no mention of electricity, while the African Union’s Continental Artificial Intelligence Strategy deals with the issue in just three sentences. Electricity is only one of the obstacles that Africa’s AI ambitions confront. Africa is endemically data poor, particularly in sectors such as energy on which developmental gains would be contingent. Where data is less scarce, it is siloed and inadequately structured for AI applications. Moreover, despite some progress in connectivity infrastructure, Africa still accounts for less than 1 percent of global data centre capacity.
Paying to Play
Gerschenkron’s framework provides a powerful heuristic through which mobile connectivity and AI can be analysed as prospective leapfrogging technologies. It is through this framework that the contingent nature of leapfrogging becomes visible. Neither mobile connectivity nor AI possess an inherent developmental quality. Rather, their developmental potential remains conditional upon access to external technologies and, more importantly, upon the domestic capacity to absorb and operationalise them within existing economic structures.
This contingency is most pronounced in Gershenkron’s second condition, particularly as it relates to the industrial potentialities of the prospective leapfrogging state(s). The comparison between mobile connectivity and AI is instructive here. Mobile connectivity was a comparatively low-threshold technology, capable of diffusing amid infrastructural scarcity, weak state capacity, and limited industrial development. This relative ‘lightness’ contributed significantly to its rapid proliferation across the continent. AI, by contrast, depends upon dense energy, data, and computational infrastructures. Its developmental promise therefore rests upon a substantially larger industrial resource base than that required for mobile telephony. Yet there are few indications that electricity generation, data infrastructure, or compute capacity across the continent are expanding at the scale necessary to sustain widespread AI diffusion.
Even mobile connectivity, despite its modest infrastructural demands, remained constrained by ordinary developmental problems. Poverty, low literacy levels, uneven connectivity infrastructure, and limited purchasing power all inhibited the extent to which mobile technologies could realise their developmental potential. Multinational telecommunications firms may have functioned as institutional substitutes for weak state-led infrastructural development, but the gains associated with mobile diffusion nevertheless remained conditioned by the underlying industrial potentialities of the societies into which the technology diffused. The implications for AI are therefore significant. If a comparatively low-threshold technology such as mobile connectivity remained constrained by these structural conditions, then the threshold conditions for AI leapfrogging are considerably higher still.
The core tension of AI leapfrogging in Africa is that meaningful participation in the AI economy presupposes the very forms of industrial development leapfrogging narratives imply can be circumvented. Insofar as AI remains dependent upon electricity generation, data infrastructures, compute capacity, technical labour, and institutional coordination, Africa cannot bypass the industrial foundations on which AI systems rest.
Recommended Readings
Mutiso, Rose. ‘Five Rules for Technology Leapfrogging in Africa.’ Science. 31 July, 2025.
