The wrong debate is whether schools should choose between modern AI and the internet. They can use both, but for different jobs.
A school can keep Starlink for administration, updates, approved research and communication while running a selected AI model on a computer or server inside the school. Learners and teachers then reach that local assistant over the school Wi-Fi or LAN without every routine question becoming an unrestricted web session.
Why this matters now
UNESCO and UNICEF both describe a real dual-use problem in digital education: connected technology expands learning opportunities while also increasing privacy, safety, governance and child-protection risks. AI adds another layer because learners can receive incorrect or inappropriate outputs if systems are adopted without supervision and policy.
Local AI does not magically make a child safe. But it gives the school another control point. The institution can choose the model, the connected knowledge, the users, the network path and the conditions under which the open internet becomes necessary.
A practical school architecture
Imagine three layers. The first is a local school AI service for everyday explanations, practice questions, teacher preparation and approved school resources. The second is restricted internet access for approved educational sites and supervised research. The third is wider internet access for staff or separately governed use.
This is very different from saying, “Here is Starlink; every learner now has the whole internet.” Connectivity remains valuable, but the school decides where it belongs.
Parents have the same opportunity at home
Some families can run an approved local model on a capable computer or home server. A child can use that assistant for homework and explanations while social media, general browsing and other internet access remain subject to separate household rules.
The value is not only privacy. It is also resilience, predictable access and the ability to make the learning tool itself distinct from the wider internet.
Organisations should think the same way
A lodge, NGO, farm, workshop or business can keep Starlink and still run private AI locally. Internal procedures, training material, manuals and selected documents can remain on the organisation's own equipment. Routine work can continue during connectivity interruptions, while difficult approved tasks can still use frontier cloud models.
The future is not cloud versus local. The useful architecture is local where control and resilience matter, cloud where frontier capability justifies it, and clear rules between the two.
Talunza's role
Talunza is not asking schools or organisations to buy GPUs for prestige. We first ask what the AI must do, how many users need it, what information should stay local and how the network should be governed. Then we size the hardware, software and support around the real workload.
Zimbabwe has a chance to adopt AI differently: not by copying every uncontrolled internet habit of richer markets, but by combining connectivity with local intelligence, offline capability and practical governance from the beginning.