Deploying edge AI as sovereignty infrastructure

When the Hunga Tonga-Hunga Ha’apai eruption severed Tonga’s submarine cable in 2022, the country was cut off from reliable internet access for weeks.

The event was a warning about the way modern digital systems are built.

Most artificial intelligence (AI) deployed across our region routes its computation through data centres thousands of kilometres away. That connection holds until it does not.

Cable cuts, cyclones, earthquakes, power failures, and bandwidth constraints do not arrive on a convenient schedule. They often coincide with the moments when public decisions matter most.

If an AI system depends entirely on a distant cloud, it becomes weakest when governments need it most. This is not a temporary coverage gap. It is architectural.

In a recent IEEE paper, I argued that digital sovereignty cannot be achieved only through policies, data protection laws, or hosting contracts.

Those things matter, but they do not change where computation happens. If the model runs somewhere else, control is partly somewhere else too.

Edge AI offers a different design choice. It means deploying AI directly on local hardware, such as smartphones, sensors, cameras, small servers, or microcontroller-class devices, rather than depending entirely on centralised cloud data centres.

The question we are not asking

Digital sovereignty is usually treated as a policy goal. Governments write strategies, pass laws, negotiate hosting terms, and ask whether data should be stored locally.

But the harder question is not only where the data sits. It is where the intelligence runs.

If every useful AI function depends on a remote data centre, then governments remain dependent on the availability, pricing, rules, and resilience of infrastructure outside their direct control.

Cloud services can be valuable, but cloud dependence should not be mistaken for sovereignty.

This reaches well beyond small island states, although islands feel it first.

Archipelagic nations, remote provinces, border districts, rural health systems, and disaster-prone regions across Asia all share the same exposure: connectivity that is assumed but not guaranteed.

What the evidence shows
 

The evidence for edge AI should be stated carefully. It does not show that every public AI service can run offline, or that the cloud should be abandoned.

It shows something narrower and more important: useful AI functions can be designed to keep working locally when connectivity is weak or unavailable.

In autonomous web-based GIS research, Mahdi Nazari Ashani and colleagues tested a browser-based small language model (SLM) for geospatial function-calling. Instead of sending every spatial query to a cloud model, the model could run on the client side in a web browser.

In the reported evaluation, the client side SLM achieved strong accuracy and removed the need for server-based inference. That is not a disaster deployment story; it is a feasibility signal.

Other technical work points in the same direction. Deeploy, for example, demonstrates energy-efficient deployment of SLMs on microcontroller-class hardware.

This is not the same as proving a full public sector service in the field, but it matters because it shows that language model inference is moving closer to devices with severe memory and energy constraints.

That makes local AI more plausible for low-resource and unreliable-connectivity environments.

There is also evidence that local knowledge can improve decision tools when it is treated seriously.

Research on hybrid precipitation forecasting has shown that combining machine learning with local or indigenous forecasting knowledge can outperform scientific or local forecasts alone in agricultural settings.

For governments, the lesson is not to extract community knowledge into machines, but to design systems where communities retain consent, interpretation, and control.

Together, these examples point to a practical design principle: AI systems for critical public functions should not be built as if the network will always be available.

Cloud systems carry broad capability. Edge systems carry narrower capability, but they can keep running when the link drops.

For disaster response, rural health, local agriculture, environmental monitoring, and frontline public services, a narrower system that keeps working may be more valuable than a broader system that disappears during a crisis.

The economics point the same way

A government spending around US$500,000 a year on cloud subscriptions may be renting capability without building enough of its own.

The same money could deploy 5,000 edge devices and train 50 local AI engineers.

Cloud spending often creates recurring dependency, while edge investment can front-load more of the cost and keep some capability closer to home.

Digital literacy, which is about teaching people to use platforms built elsewhere, isn’t the same as digital capability to build your own. Only the second produces sovereignty.

One caution: Bringing indigenous knowledge into these systems only works when communities are partners, not data sources, with real consent and control over interpretation.

The risk of extraction never goes away, and the governance to manage it is a technical requirement, not an afterthought.

The choice is not between cloud and edge. It is between remaining a downstream consumer of systems designed elsewhere and building the capacity to shape systems that reflect local reality.

The technical barriers are surmountable.

Whether governments choose to build that capability is the question that now matters.

If you're interested to read more of Mohamed's columns, you can click here.

---------------------------------------

Mohamed Shareef is a former Minister of State for Environment, Climate Change and Technology in the Maldives (2021-2023). He previously served as Permanent Secretary of Science and Technology Ministry (2019-2021) and the Chief Information Officer at the National Centre for Information Technology (2009-2014) and led the development of the country’s national digital public infrastructure. He also served in the academia including as a researcher at the United Nations University. He currently serves as Senior Advisor for Digital Transformation at Nexia Maldives.