The global AI ecosystem today is largely built on centralised, cloud-based infrastructure. Yet that infrastructure is often not adapted to the needs and constraints of low- and middle-income countries (LMICs), where connectivity, power reliability, and costs limit access.
Edge AI offers a promising alternative. By running models directly on or near the device where data is generated and processing data locally rather than in distant data centres, it can reduce latency, lower bandwidth costs, improve privacy and reliability, and help close emerging compute gaps as AI adoption grows.
The compute gap in LMICs
Today, most LMICs depend on costly international cloud services hosted in the US, China, and Europe, where most large-scale data centres and compute clusters are concentrated. Africa, for example, accounts for less than 1% of global data centre capacity despite being home to 18% of the world’s population. Meanwhile, US-based technology firms alone are expected to invest over $300 billion on AI infrastructure in 2025 – more than the combined GDP of many LMICs. This imbalance risks leaving them dependent on foreign providers, with limited ability to shape or benefit from the emerging AI economy.
Across LMICs, startups, research labs, and small enterprises face a fragmented and often fragile compute landscape, shaped by overlapping technical, financial, and policy barriers. Cloud tokens and grants offer important and much needed support to local innovators, but sustained AI use remains expensive.
Variable USD costs – from cloud API fees to data transfer and storage – are compounded with volatile exchange rates, high bandwidth costs, and limited credit access. Local hosting options are often scarce, concentrated in major cities with limited redundancy, cooling capacity, and power reliability.
This fragmented landscape stems from a convergence of structural barriers including:
- Unreliable power: frequent blackouts drive up costs and undermine continuous operations, forcing reliance on expensive diesel generators.
- High costs: cloud usage, hardware, and bandwidth remain expensive, often worsened by currency volatility and tariffs.
- Skills shortages: the lack of trained professionals in model optimisation, orchestration, or secure data management means that even available infrastructure is often underused or misconfigured.
- Connectivity barriers: low bandwidth and high latency constrain real-time AI applications and slow model updates.
- Policy uncertainty: inconsistent rules around data protection and cross-border transfers create uncertainty for both users and investors.
These barriers become even sharper as demand grows for compute-intensive models like generative AI and reasoning systems. More than half of the startups applying to the recent GSMA’s Innovation Fund for Impactful AI proposed using GenAI, with many referencing large language models. While this shows local appetite for cutting edge innovations, scaling such solutions will be difficult without fit-for-purpose compute.
The opportunity for edge AI
Edge AI spans a wide spectrum of devices, from high-end smartphones to microcontrollers and embedded sensors. Together, these form an increasingly capable and increasingly affordable compute layer that is already widely distributed across LMICs.
Unlike cloud-based AI, which depends on stable power, high bandwidth and costly data transfers, edge AI brings computation directly to where data is generated, enabling faster, cheaper and more resilient services. At the most constrained end lies TinyML, ultra-efficient models that run on milliwatts of power and require no continuous connectivity. TinyML is already being applied in agriculture, healthcare, and biodiversity monitoring from Brazil and Peru to India, Kenya and Malaysia, demonstrating its potential to deliver value in resource-constrained settings.
At the higher end, smartphones are emerging as the primary edge device in LMICs, serving as the backbone of digital ecosystems. By 2030, according to the GSMA data, an estimated 81% of mobile connections in Africa will be via smartphones, up from 51% in 2023. Smartphone penetration already exceeds 80% in Asia-Pacific and Latin America. This growth creates a significant opportunity to embed locally relevant AI capabilities directly into devices. Thanks to increasingly powerful chipsets, more efficient models, and accessible open-source tools, smartphones can now perform tasks such as speech recognition, image classification, and lightweight language-model inference, all without relying on cloud infrastructure.
Together, these trends create a major opportunity for LMICs to benefit from AI in ways that do not depend on high-capacity cloud infrastructure. Edge deployments can lower bandwidth and cloud costs, offer greater resilience in low-connectivity settings, strengthen privacy and data sovereignty, and enable more locally relevant applications. For many innovators, this is becoming a practical pathway to overcome compute constraints, support cost-sensitive business models, and reach underserved communities.
Why now, and what next?
Despite growing interest, there is still limited evidence on the feasibility and economics of edge AI in resource-constrained settings. Existing studies including recent GSMA research, have mainly examined model training, compute pathways, and large-scale cloud infrastructure. The next step is to look more closely at real-world readiness and the business case for edge deployments in LMICs.
Our upcoming mixed-methods research will bring fresh evidence in this area. The research will combine quantitative modelling of market readiness and unit economics with qualitative case studies documenting edge deployments across Africa, South Asia, and Latin America. We will explore:
- The cost and performance trade-offs between edge and cloud inference;
- The potential of TinyML, reasoning models, and distributed learning to support low-resource innovation;
- Economic productivity effects in key sectors such as agriculture, health, and education;
- Policy and regulatory enablers for sustainable and inclusive edge AI ecosystems.
By grounding technical foresight in economic and social realities, this work aims to help industry, investors, and policymakers chart a path toward scalable and equitable edge AI in the Global South.
For continued insights and future developments on this topic, subscribe to our newsletter.
This project is funded by the UK Foreign, Commonwealth & Development Office and supported by the GSMA and its members.
