The World Bank’s ‘Learning Poverty’ crisis presents a staggering challenge: the majority of ten-year-olds in low- and middle-income countries cannot read a simple text.[1] This is not simply an access problem. Even children who are in school are not receiving teaching adapted to their needs. For too long, development agencies have wrestled with a seemingly impossible trade-off: the choice between achieving scale or delivering personalisation.
The Current Landscape
Organisations like Ubongo have shown what distribution excellence looks like, using mass media to deliver ‘edutainment’ to millions of children across Africa.[2] Similarly, platforms like Kolibri provide offline access to curated educational content, bridging the digital divide in disconnected environments.[3] These models are essential, yet they deliver a one-to-many, one-size-fits-all, curriculum.
The future of education, we are told, is Generative AI: an adaptive, one-to-one tutor for every child. Wheeler Institute research on responsible Generative AI deployment notes that the technology “has the potential to democratise knowledge,” yet warns of risks like over-reliance.[6] This tension is sharpened by a practical question: how can we deploy cloud-based, data-intensive large language models in regions with intermittent, low-bandwidth, or non-existent connectivity?
Breaking the Binary: The Business Opportunity
There is a prevailing assumption that advanced, personalised AI cannot function where infrastructure is weak. Call it the ‘Connectivity Trap.’ It is a view that no longer holds, and it represents precisely the kind of market inefficiency that business can address. A third way is now emerging. Rather than relying on distant cloud servers, decentralised ‘Edge AI’ brings the processing power, a compressed AI brain, directly onto the device itself.
Technical Spotlight: Defining Key Terms
- Edge AI (Offline Intelligence) describes AI systems that process data directly on the user’s hardware, such as a tablet or phone, rather than sending it to a central server over the internet. This means the software functions fully without any connection at all.
- SLMs: ‘Small Language Models’ (SLMs)are compact versions of AI models, engineered to fit within the limited storage and memory of a standard, low-cost tablet. They sacrifice scale, not capability, retaining the specific knowledge needed for tutoring while running on a fraction of the computing power.
- On-Device Inference refers to the moment the AI actually thinks. The tablet’s processor runs the model locally, generating answers and adjusting the curriculum in real time. No data leaves the device, which means faster responses and built-in privacy protection for students.
The case for such an approach is grounded in the scale of the challenge. Professor Elias Papaioannou’s Wheeler Institute research on education across Africa demonstrates that despite an unprecedented expansion of school access, “nearly one third [of African children] still do not finish primary school and more than half do not finish secondary school.”[4] It points to the market frictions shaping the success of school expansion, and asks when increases in educational expenditure are most likely to pay off. In the context of AI-powered education, this insight carries a direct implication: simply building more schools, or more cloud infrastructure, will not work without understanding the local conditions that shape outcomes.
A past Wheeler Institute panel on AI for Africa reinforces this point. The participants argued, “African governments and businesses should prioritise adaptation and application rather than attempts to compete with the US and China in building foundational models.”[5] Training frontier AI systems has become prohibitively expensive, but crucially, “the cost of using those models is falling quickly,”[5] creating a genuine opening for education innovators.
Applied to EdTech, this means a powerful, adaptive AI tutor can run locally on a low-cost tablet, independent of a constant internet connection, creating both social impact and business value. Crucially, because Edge AI is calibrated and deployed locally, it directly addresses one of the central risks of mainstream AI: that models trained on under representative global data can fail the very communities they are meant to serve.
Evidence-Based Impact
Edge AI offers a powerful way to scale personalisation. Its offline-first architecture reduces implementation costs without sacrificing effectiveness. Yet responsible deployment matters. Wheeler-funded research by Tong Wang, Kamalini Ramdas and Monika Heller examines how AI communication styles affect understanding and decision-making in education contexts, and warns that “over-reliance on AI-generated information and the spread of misinformation are concerns.”[6] The researchers note that poorly calibrated tools risk exacerbating “existing disparities, particularly in education and healthcare, where accurate, personalised information is crucial for effective outcomes.”[6] By grounding AI in local contexts rather than relying on distant, generic cloud models, Edge AI offers a more responsible path to deployment.
Continued, Wheeler Institute funded research by Tong Wang, Kamalini Ramdas and Monika Heller on equitable access to online learning further underscores the opportunity: “Generative AI could substantially empower personalised learning, by tailoring education to meet learners’ different needs or preferences.”[7] Yet the same research warns that “promising technologies like blockchain and AI can exacerbate disparities if not made accessible to all.”[7] Applied to the current landscape the implication for practitioners is clear: locally deployed, device based AI is one of the most direct means of ensuring these technologies reach those who need them most.
A New Investment Thesis
What emerges is more than a technical solution; it is a new, investable thesis. The next frontier for tech-focused investment is not in building more centralised, infrastructure-heavy platforms. It lies in funding a new generation of connectivity independent, distributed AI-powered ventures.
For investors and business leaders, this approach offers compelling advantages. First, eliminating the dependency on constant connectivity fundamentally reshapes the unit economics of educational technology in emerging markets.[5] Second, solutions become viable in previously underserved regions, dramatically expanding the addressable market. Third, the model demonstrates that educational equity and profitable enterprise need not be mutually exclusive. Edge AI in education represents one of the clearest examples of how to circumvent these infrastructure gaps rather than waiting for them to close.
The opportunity is clear. When business treats educational inequity not as a charitable cause but as a market inefficiency, it unlocks solutions that deliver both social impact and financial return. AI does not have to remain in the cloud, accessible only to the connected. It can be placed directly into the hands of the students the current system fails to reach.
[1] World Bank, The State of Global Learning Poverty: 2022 Update. https://www.worldbank.org/en/topic/education/publication/state-of-global-learning-poverty
[2] Ubongo, Impact. https://www.ubongo.org/impacts/
[3] Learning Equality, Kolibri. https://learningequality.org/kolibri/
[4] Papaioannou, E., Michalopoulos, S. and Figueiredo Walter, T., “Education for all, occupational choice and business formation in Africa,” Wheeler Institute Research Portal. https://wheelerinstituteresearch.org/project/education-for-all-occupational-choice-and-business-formation-in-africa/
[5] Wheeler Institute for Business and Development, “AI for Africa: Harnessing AI for jobs, growth and investment,” WheelerBlog, 17 December 2025.
[6] Wang, T., Ramdas, K. and Heller, M., “Applying Generative AI with Responsibility,” Wheeler Institute Research Portal. https://wheelerinstituteresearch.org/project/applying-generative-ai-with-responsibility/
[7] Wang, T., Ramdas, K. and Heller, M., “Increasing equitable access to online learning platforms,” Wheeler Institute Research Portal. https://wheelerinstituteresearch.org/project/increasing-equitable-access-to-online-learning-platforms/
About the writer
Nasreen Begum is an MBA 2027 candidate at London Business School. Prior to joining the school, she worked at Engine AI as a Product Manager, where she directed the development of autonomous data agents for the financial sector. Nasreen is focused on advancing equitable education in developing economies. She has a particular interest in leveraging technology and AI to overcome infrastructure barriers and improve access to learning in underserved regions
Student voice
The Wheeler Institute for Business and Development is seeking to understand, illuminate and offer solutions to the challenges faced by the developing world, with an aim to identify the role of business in addressing these challenges and a focus on the implications and actions for those in developing countries. In support of our students, we approach this blog section as a reflective platform and a space where individuals can generate debate as long-term agents of positive change. This article is solely authored by a student and reflects their individual research, opinion and point of view and is not based on research led or supported by the Wheeler Institute.